Putting Skeptics in Their Place

This book is about the nature of skeptical arguments and their role in
philosophical inquiry. John Greco defends three theses: that a number
of historically prominent skeptical arguments make no obvious mistake
and therefore cannot be easily dismissed; that the analysis of skeptical
arguments is philosophically useful and important and should therefore
have a central place in the methodology of philosophy; and that taking
skeptical arguments seriously requires us to adopt an externalist, reliabilist
epistemology.
Greco argues that the importance of skeptical arguments is methodological. Specifically, skeptical arguments act as heuristic devices for
highlighting plausible but mistaken assumptions about the nature of
knowledge, thereby requiring us to replace these assumptions with
something better. Consequently, the analysis of skeptical arguments drives
positive epistemology. It is further argued that taking skeptical arguments
seriously requires us to adopt a version of 'Virtue epistemology," or a
theory of knowledge that makes intellectual virtue central in the analysis
of knowledge. This methodology has consequences for moral and
religious epistemology; in particular, a theory of moral perception is
defended.
This book will be of interest to professionals and graduate students in
epistemology and moral philosophy.

John Greco is Associate Professor of Philosophy at Fordham University.

CAMBRIDGE STUDIES IN PHILOSOPHY

General editor ERNEST SOSA (Brown University)
Advisory editors:
J O N A T H A N D A N C Y (University of Reading)

JOHN H A L D A N E (University of St. Andrews)

G I L B E R T HARMAN (Princeton University)
F R A N K J A C K S O N (Australian National University)
WILLIAM G. LYCAN (University of North Carolina at Chapel Hill)
S Y D N E Y S H O E M A K E R (Cornell University)

J U D I T H j . THOMSON (Massachusetts Institute of Technology)

R E C E N T TITLES:
LYNNE R U D D E R BAKER

Explaining Attitudes

ROBERT A. WILSON Cartesian Psychology and Physical Minds

BARRY MAUND Colours
MICHAEL DEVITT

MICHAEL ZIMMERMAN

Coming to Our Senses

The Concept of Moral Obligation

MICHAEL S T O C K E R with ELIZABETH HEGEMAN

SYDNEY SHOEMAKER
NORTON NELKIN

Valuing Emotions

The First-Person Perspective and Other Essays

Consciousness and the Origins of Thought

M A R K LANCE and JOHN O'LEARY H A W T H O R N E

The Grammar of Meaning

D. M . A R M S T R O N G A World of States of Affairs

PIERRE JACOB What Minds Can Do
ANDRE GALLOIS The World Without the Mind Within
FRED FELDMAN Utilitarianism, Hedonism, and Desert
LAURENCE B O N J O U R In Defense of Pure Reason
DAVID LEWIS Papers in Philosophical Logic
WAYNE DAVIS

DAVID C O C K B U R N

DAVID LEWIS

Implicature

Other Times

Papers on Metaphysics and Epistemology

RAYMOND MARTIN

ANNETTE BARNES

MICHAEL B R A T M A N
AMIE T H O M A S S O N

Self-Concern

Seeing through Self-Deception

Faces of Intention

Fiction and Metaphysics

DAVID LEWIS Papers on Ethics and Social Philosophy

FRED DRETSKE Perception, Knowledge, and Belief
LYNNE R U D D E R BAKER

Persons and Bodies

Putting Skeptics in
Their Place
THE NATURE OF SKEPTICAL
ARGUMENTS AND THEIR ROLE IN
PHILOSOPHICAL INQUIRY

John Greco
Fordham University

CAMBRIDGE
UNIVERSITY PRESS

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521772631
John Greco 2000
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2000
This digitally printed version 2007
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data
Greco, John
Putting skeptics in their place : the nature of skeptical arguments and their role in
philosophical inquiry / John Greco
p. cm. - (Cambridge studies in philosophy)
Includes bibliographical references and index
ISBN 0-521-77263-X (hardcover)
1. Skepticism.

I. Title.

II. Series.

B837 .G74 2000

149'.73-dc21
99-042103
ISBN 978-0-521-77263-1 hardback
ISBN 978-0-521-04553-7 paperback

To Lizabeth and Sofia

Contents

Preface
1

page xiii

The Nature of Skeptical Arguments and Their Role in

Philosophical Inquiry
1. The Three Theses
2. Strategy and Outline of the Book
3. Three Criteria for an Adequate Theory of Knowledge

1
2
6
15

Skepticism about the World: Part One Reconstructions

1. An Argument from Hume
2. An Argument from Descartes
3. Need the Skeptic Challenge All of Our Knowledge at Once?
4. Two More Ways Not to Understand Descartes' Argument
5. Discriminating Evidence
6. Supporting Evidence

25
25
34
39
51
54
58

Skepticism about the World: Part Two Dismissive

Responses
1. Charges of Self-Refutation
2. Pragmatic and Rhetorical Responses
3. More Dismissive Responses
4. Transcendental Arguments

61
61
64
69
71

Skepticism about the World: Part Three Dualism, Realism,

and Representationalism
/. What Is Necessary for Skepticism about the World
1. The Modern Ontology and Skepticism
2. Realism and Skepticism
3. Representationalism and Skepticism

77
83
83
88
90

II.
1.
2.
3.
4.

What Is Sufficient for Skepticism about the World

The Big Mistake
The Modern Ontology Again
Representationalism Again
Conclusions

94
95
102
103
106

T h e Argument from an Infinite Regress of Reasons

1. The Regress Argument and Strong Particularism
2. Foundationalism
3. Contextualism
4. Coherentism
5. The Objection to Traditional Coherence Theories
6. The Objection to Nontraditional Coherence Theories
7. The Role of Sensory Appearances
8. Conclusions

108
108
118
121
126
128
130
134
135

Hume's Skepticism about Unobserved Matters of Fact

I. Hume's Arguments and Dismissive Responses
1. The Arguments in Sections IV and VII of the Enquiry
2. Objection (a): Hume Relies on an Inadequate Theory
of Ideas
3. Objection (b): Hume Requires Absolute Certainty
for Knowledge
II. The Standard Objection against Hume
1. The Objection
2. Deductive, Inductive, and Nonsupportive Inferences
3. An Objection Considered (against Our Interpretation
of Hume)
4. The Lesson of Hume's Skeptical Argument

137
138
138

Agent Reliabilism
/. Simple Reliabilism
1. Simple Reliabilism: The Big Idea
2. Why Skeptical Arguments Go Wrong
3. From Processes to Virtues
//. Agent Reliabilism and the Question of Subjective Justification
1. Knowing That One Knows
2. Understanding That One Knows
3. Sosa's Perspectivism
4. From Perspectives to Dispositions
5. The Place of Epistemic Norms
6. The Place of Epistemic Responsibility
7. Conclusions

164
165
165
167
174
180
181
184
187
190
192
200
202

141
145
146
147
149
155
159

Agent Reliabilism and the Relevant Sense of

' 'Relevant Possibility''
1. Two Tasks for Working out This Approach
2. The Relevant Sense of "Relevant Possibility"
3. A Theory of Virtues and a Virtue Theory of Knowledge
4. Agent Reliabilism and Relevant Possibilities
Moral and Religious Epistemology
/. Religious Epistemology
1. Plantinga's Rejection of the Evidentialist Objection to
Religious Belief
2. Alston's Defense of Religious Perception
IT. Moral Epistemology
1. Three Kinds of Moral Epistemology
2. A Theory of Empirical Perception
3. A Theory of Moral Perception
4. Some Traditional Objections to Moral Perception
///. General Conclusions

Bibliography
Index

204
205
207
211
217
220
221
222
226
231
232
235
241
244
248
255
261

XI

Preface

This book is largely the result of snide remarks from my colleagues at

Fordham University. I arrived there very interested in skeptical problems, and people like Vincent Colapietro and Merold Westphal would
wonder, out loud, why. One time, after I had presented a paper to the
department on Humean skepticism, Chris Gowans bluntly asked me
why anyone should find this sort of thing interesting. On another occasion, I was asked why I had my heart set on raising the dead Hume
had already been refuted, so why bother doing it again? Remarks like
this were only half serious, but they were frequent enough to set me
thinking about why I was so interested in skeptical arguments. It certainly was not because I thought that skepticism might be true. In fact,
in this sense I was far less skeptical than most of my colleagues. Rather,
I came to realize that my interest in skepticism was methodological.
Along with a great many other epistemologists, I was interested in
skeptical arguments because I thought that they could teach substantive
lessons about the nature of knowledge and evidence. It was part of my
methodology, in fact, to assume that skepticism is false, and that skeptical
arguments must go wrong somewhere. The trick was to say where, and
to learn the philosophical lesson contained therein.
A second thing that set me thinking was an observation about the
profession: it seemed to me that nearly all philosophers have a story
about why skepticism is wrong. Even people who are not interested in
epistemology usually have a neat refutation of skepticism that they take
for granted and think closes the case. None of these stories seemed even
remotely plausible to me, but many nevertheless seemed to have the
status of philosophical lore, with different lore existing in different circles. Finally, I was impressed that so many philosophers found the recent

xni

reliabilist trend in analytic philosophy to be entirely counterintuitive,

almost a nonstarter. I became convinced that all of these things were
related. It was partly because people thought that they had an easy
answer to skepticism that they dismissed it as a non-problem. And it was
partly because they thought skepticism to be a non-problem that they
could not see the motivations for reliabilism. Finally, only a methodology that took skeptical arguments seriously could drive home either of
these points. You need to take the arguments seriously to see that they
cannot be easily refuted. And it is only after seeing this that you can
appreciate the necessity of adopting some form of reliabilism in the
theory of knowledge.
I take it that the views about skepticism just described have wide
currency in philosophy. As I have already said, nearly everyone in
philosophy has a story about skepticism that is incompatible with the
one I am telling in this book. Accordingly, I take myself to be writing
for a broad philosophical audience, including philosophers from a wide
range of traditions, and including philosophers who are not particularly
interested in epistemology. Finally, I hope also to have said some things
that will be of interest to epistemologists, in particular. Specifically, the
current trend toward reliabilist theories is not ubiquitous in analytic
epistemology. On the contrary, a variety of objections have been directed against externalism in general and reliabilism in particular. By
defending a version of reliabilism that is also a version of virtue epistemology, I hope to show how a reliabilist epistemology can be developed
to meet the most powerful of those objections. For example, a virtue
account allows us to see how knowledge is subjectively appropriate as
well as objectively reliable. At the same time, developing the position as
a version of virtue epistemology places it in a historical context going
back at least to Aristotle, which helps to further address the counterintuitiveness of the position.

Acknowledgments
I am indebted to many people for my thinking in this book. The most
influential of these were my teachers in epistemology at Brown University: Roderick Chisholm, Ernest Sosa, and James Van Cleve. My continuing debt to Sosa will be obvious to anyone who is familiar with the
territory. Indeed, the extent of this debt is sometimes embarrassing, but
I decided early on that it was better to be right than original, and so that

xiv

is what I have tried to do. In graduate school I considered a number of

people to be my teachers in books and essays. These were William
Alston, Robert Audi, Laurence Bonjour, Alvin Goldman, Alvin Plantinga, John Pollock, and Barry Stroud.
I am indebted to many more people for their comments on the
manuscript or on relevant material. Keith Lehrer, Frederick Schmitt, and
Michael Williams read all of the manuscript and provided me with many
valuable comments, as did an anonymous referee. I thank all four for
helping me to think through the relevant issues and to improve the
manuscript on the basis of their careful criticisms. Robert Audi, Matthias
Steup, Ernest Sosa, James Van Cleve, Margaret Walker, and Linda Zagzebski all read parts of the manuscript and also provided me with countless valuable suggestions and criticisms. I am sure that I have not adequately addressed all of the concerns that they have raised; I can only
promise that I will try to think more about these issues in days to come.
So many other people have also read or heard relevant material and
have provided me with all sorts of comments, ranging from useful advice
to insightful criticisms. These include William Alston, Jose Bermudez,
Judith Bradford, Lizabeth Byrne, Norris Clarke S.J., Stewart Cohen,
Vincent Colapietro, Terence Cuneo, Brian Davies, Wayne Davis, Michael DePaul, Keith DeRose, Chris Gowans, Stephen Grimm, Christopher Hookway, Terence Horgan, John Jenkins, Jude Jones, Jaegwon
Kim, Hilary Kornblith, Joseph Koterski S. J., Matthew Kuenning, James
Marsh, David Martens, Richard McCombs, Robert Meyers, Dennis
Monokroussos, Ralf Muller, Hilde Nelson, Carmen Pace, George Pappas, Rob Philp, Alvin Plantinga, Vincent Potter S. J., Curt Purcell, Alan
Rhoda, Bob Roth S. J., Michael Sharkey, John Stolt, Dennis Sweet,
Daryl Tress, John Van Buren, Merold Westphal, Michael Williams,
Nicole Williams, Nicholas Wolterstorff, John Zeis, and Michael Zimmerman. I am sure there are more who should have been mentioned.
In what follows I draw from a number of my previously published
essays. Chapter 4 expands on material from "Modern Ontology and the
Problems of Epistemology," American Philosophical Quarterly 32 (1995),
while Chapter 6 expands on material from "The Force of Hume's
Skepticism about Unobserved Matters of Fact," Journal of Philosophical
Research 23 (1998). Chapter 8 draws quite heavily from "Virtue Epistemology and the Relevant Sense of 'Relevant Possibility/ " Southern
Journal of Philosophy 32 (1994). Finally, Chapter 9 expands on material
from "Perception as Interpretation," in Michael Baur, ed., Texts and

xv

Their Interpretation, Proceedings of the American Catholic Philosophical

Association (New York: American Catholic Philosophical Association,
1999). I would like to thank the editors of these publications for their
permission to use the relevant material.

xvi

1
The Nature of Skeptical Arguments
and Their Role in Philosophical
Inquiry

This book has three major theses: (1) that a number of historically
prominent skeptical arguments make no obvious mistake and therefore
cannot be easily dismissed; (2) that the analysis of skeptical arguments is
philosophically useful and important and should therefore have a central
place in the methodology of philosophy, particularly in the methodology
of epistemology; and (3) that taking skeptical arguments seriously requires us to adopt an externalist, reliabilist epistemology. More specifically, it motivates a position that I call "agent reliabilism," which is an
externalist version of virtue epistemology.
If these theses are correct, then many philosophers have misunderstood the nature of skeptical arguments and their role in philosophical
inquiry. For example, many philosophers think that skepticism poses no
philosophically interesting problem. According to this view, skeptical
arguments rest on some obvious mistake, such as a quest for absolute
certainty or a demand for immutable foundations, and can therefore
easily be dismissed. Others think that skepticism rests on a substantive
philosophical mistake, but that skeptical arguments teach no epistemological lessons. For example, many philosophers think that skepticism is
rooted in a bad ontology. On this view skeptical arguments assume an
ontological dualism between knowing mind and material object of
knowledge and can therefore be rejected by rejecting the offending
dualism. Others have thought that skepticism is rooted in representationalism, and still others that it is rooted in realism. Finally, some philosophers have appreciated that skepticism is indeed an epistemological problem, but have tried to solve it by remaining within a traditional,
internalist epistemology. Against all of these positions, I argue that the
recent externalist revolution in epistemology is necessary for a quite

traditional reason: to adequately address a range of well-known skeptical

arguments.1
1. THE THREE THESES

My first thesis is that a number of historically prominent skeptical arguments make no obvious mistake. On the contrary, such arguments begin
with assumptions about knowledge and evidence that seem eminently
plausible, outside the context of philosophical inquiry. Often they are
assumptions that we ourselves accept either explicitly or implicitly. But
by reasoning that is seemingly cogent, such arguments "prove" a conclusion that is outrageously implausible, even incredible in the literal
sense. Accordingly, skeptical arguments are powerful in the following
sense: it is not at all easy to see where they go wrong, and rejecting
them requires one to adopt substantive and controversial theses about
the nature of knowledge and evidence. This is not to say that they are
powerful in a psychological sense that they have the power to persuade. In this respect skeptical arguments are like arguments for God's
existence: it is doubtful that any has ever produced a convert.
My second thesis is methodological and is closely tied to the first.
Specifically, I argue that the analysis of skeptical arguments is philosophically useful and important. This is not because skepticism might be true
and we need to assure ourselves that we know what we think we know.
Neither is it because we need to persuade some other poor soul out of
her skepticism. Rather, skeptical arguments are useful and important
1

Some philosophers do take skeptical arguments seriously, giving them pride of place in their
own methodology. For recent discussions that endorse my first two theses, see Peter Klein,
Certainty: A Refutation of Scepticism (Minneapolis: University of Minnesota Press, 1981); Barry
Stroud, The Significance of Philosophical Scepticism (Oxford: Clarendon Press, 1984); Robert
Fogelin, Pyrrhonian Reflections on Knowledge andJustification (Oxford: Oxford University Press,

1994); and Michael Williams, Unnatural Doubts (Princeton: Princeton University Press,
1996). None of these authors endorses my third thesis, however. Williams argues that skeptical arguments mistakenly assume a thesis he calls "epistemic realism," while Klein argues
for a defeasibility theory of knowledge. Stroud and Fogelin suggest that certain skeptical
arguments cannot be answered in any satisfactory manner. In the context of the more usual
attitudes regarding the nature and usefulness of skeptical arguments, however, my disagreements with these authors come fairly late in the day. Much of what I say in the book, in
fact, overlaps with one or another of them. Finally, a good number of analytic epistemologists
- reliabilists and externalists among them - engage in the analysis of skeptical arguments
episodically as part of their methodology, and it is fairly clear that their own accounts of
knowledge and evidence are motivated by this. In what follows I try to give this common
practice an explicit articulation and systematic defense. I also recommend a more consistent
application of it.

because they drive progress in philosophy. They do this by highlighting

plausible but mistaken assumptions about knowledge and evidence, and
by showing us that those assumptions have consequences that are unacceptable. As a result we are forced to develop substantive and controversial positions in their place. On this view skeptical arguments are important not because they might show that we do not have knowledge, but
because they drive us to a better understanding of the knowledge we do
have. There is another side to this coin: the price of ignoring skeptical
arguments, or of rejecting them for the wrong reasons, is that we miss
the lessons that the arguments can teach us.
Again, if these two theses are correct, then philosophers from a wide
range of traditions have badly misconceived the nature and place of
skeptical arguments. As we have already noted, many philosophers think
that skeptical arguments make some obvious mistake and may therefore
be easily dismissed. One example of this position is the view that skeptical arguments require some high level of certainty for knowledge
perhaps infallible premises, or incorrigible ones. Another is the view that
skeptical arguments allow only deductive inferences. According to these
diagnoses, to reject skepticism we need only reject its implausible standards for knowledge. Yet another widely held view is that skepticism is
self-refuting, either because the skeptic makes incompatible claims (she
knows that no one knows), or because skepticism cannot be lived and
so the skeptic is caught in a "performative" contradiction.
If any one of these diagnoses were correct, then the analysis of
skeptical arguments would not be philosophically useful or important.
Such easy refutations would teach no lesson at all, or perhaps only one
learned long ago that rationalism is false. But against these views I
argue that historically prominent skeptical arguments for example, ones
from Aristotle, Descartes, and Hume make no obvious mistake and
therefore cannot be easily dismissed. These skeptical arguments can be
interpreted so that they involve no contradiction, performative or otherwise. Moreover, a close investigation of such arguments reveals that
they run on assumptions much more dear to us than infallibilism or
deductivism. Their lesson is not that some vestige of rationalism is false,
but that something much more plausible will have to be given up.
There is another view of skepticism that is widely popular among
philosophers, and that is incompatible with the one I am defending here.
Philosophers from a surprising range of traditions claim that skeptical
arguments presuppose a Cartesian ontology of internal subjects and external objects. Once the offending dualism is given up, these philoso-

phers claim, skeptical arguments cannot even get off the ground. A
strong version of this diagnosis implicates not only skepticism but epistemology as well. The idea is that epistemology is essentially the activity
of constructing solutions to skeptical problems. But if skeptical problems
cannot arise in a post-modern world, then epistemology is robbed of its
purpose, and therefore of its existence.
Other philosophers think that skepticism is grounded not in a bad
ontology but in a bad philosophy of mind. These philosophers make
representationalism the root of all skeptical arguments. Still other philosophers think that a bad theory of reference is the problem, and still
others think that the skeptic makes some kind of linguistic mistake.
Against all of these positions I want to argue that skeptical arguments
run on mistaken assumptions about the nature of knowledge and evidence. A close analysis of skeptical arguments drives positive epistemology, not ontology, or philosophy of mind, or philosophy of language.
My third thesis is that taking skeptical arguments seriously pushes us
in a particular direction in epistemology. Specifically, it pushes us toward
externalism and reliabilism. Even more specifically, it pushes us toward
agent reliabilism. The idea is this: reconstructed in their most plausible
form, a number of skeptical arguments show, quite correctly, that there
is no necessary relation between our beliefs and their evidential grounds.
It is now a commonplace to recognize that there is no such deductive
relation. The more interesting point, however, is that there is no necessary inductive relation either; it is not a necessary truth that the grounds
for our beliefs make them even probable. For many philosophers this
contention would be enough to entail skepticism. If there is no logical
or quasi-logical relation between our evidence and our beliefs, as some
would require for our cognition to be "within the logical space of
reasons," then a fundamental condition of knowledge goes unfulfilled.
There is in fact no such relation, however. This is one of the most
important lessons that skeptical arguments teach us. A necessary condition for avoiding skepticism, therefore, is to rethink what it is for the
grounds of our beliefs to be good evidence. Put another way, it is
necessary to rethink what it is to be within the space of reasons. As it
turns out, that space is neither logical nor quasi-logical. It is at most a
contingent fact that the grounds for our beliefs are reliable indications of
their truth, and any adequate epistemology must account for this.
The relevance of all of this to reliabilism is now easy to see. Taking
skeptical arguments seriously provides a powerful motivation for reliabilism in epistemology, insofar as reliabilism can explain why evidence

need not be logical or quasi-logical. According to reliabilism, a belief

has positive epistemic status (roughly) just in case it is in fact reliably
formed. Put in terms of evidence, the grounds on which a person forms
her belief amount to good evidence (again roughly) just in case those
grounds are in fact a reliable indication that the belief is true. There is
no requirement that the person know that her grounds are reliable, or
even that she could know this on reflection. The latter requirements are
plausibly fulfilled if we think that evidential relations are necessary. In
that case, one might expect a kind of a priori insight into the fact that
one's evidence indicates the truth of one's belief, either by entailing it
or by necessarily making it probable. But there is no possibility of
fulfilling such a requirement if the relation between our evidence and
our beliefs is merely contingent. On the contrary, a requirement in that
direction leads straight into skepticism. Reliabilism makes no such requirement and gives us a general approach to knowledge and evidence
that explains why none is needed.
There are serious problems with generic reliabilism, however serious enough to cause some philosophers to reject the position out of
hand. One problem is that beliefs can be reliably formed by accident for example, by arbitrarily adopting a method which, unknown to the
believer, happens to be reliable. This would seem to violate a "no
accident" condition on knowledge. A second problem is that beliefs can
be reliably formed and yet subjectively inappropriate. This seemingly
violates a "subjective justification" condition on knowledge, and reliabilism has been widely criticized on this point. Agent reliabilism addresses both problems by drawing on the resources of virtue theory. The
main idea is to define knowledge in terms of virtuous cognitive character, and to define virtuous character in terms of proper motivation and
reliable success. This takes care of the "no accident" condition on
knowledge, in that true belief which is formed through an agent's
reliable character is not an accident in any relevant sense. It takes care of
the "subjective justification" condition as well, since there is proper
motivation, and, as Aristotle would say, "the moving principle is within
the agent." Roughly, a belief is both subjectively and objectively justified, in the sense required for knowledge, when it is produced by a
properly motivated, reliable cognitive character.
Agent reliabilism is therefore a kind of virtue epistemology, in the
sense that it makes cognitive or intellectual virtue central in the analysis
of important epistemic concepts. As such, it is an improvement over
previous versions of reliabilism, including process reliabilism, method

reliabilism, and evidence reliabilism. All of these are subject to one or

both of the two problems that I have mentioned, precisely because they
fail to ground knowledge in the virtuous character of the knower. In
defending agent reliabilism I do not pretend to offer a position that is
either wholly original or fully worked out in its details. Rather, I am
defending a general direction in epistemology. This direction has been
taken by others, both historically and more recently. I am arguing that
it is necessary in the context of well-known but underappreciated skeptical considerations.2
2. STRATEGY AND OUTLINE OF THE BOOK

So I have three major theses: one about the structure and content of
skeptical arguments, one about their methodological role in philosophical inquiry, and one about where this methodology leads us. My strategy
for establishing these is to engage in five tasks.
One thing I will have to do is to consider and reject dismissive
responses to skepticism. I define a "dismissive response" as one that
either (a) does not engage skeptical arguments at all or (b) engages them
only superficially. Such responses are "dismissive" because they reject
the skeptical conclusion without seriously considering the reasoning that
leads to it. I include charges of self-refutation under "type-a" dismissive
responses. Type-a responses do not consider skeptical arguments at all
but rather react to the implausibility of the skeptical conclusion. But
skepticism is not self-refuting in any philosophically interesting sense.
To think that it is blinds one to the more subtle mistakes that skeptical
arguments make, and that many non-skeptical philosophers make as
well.
Under type-b dismissive responses I include the charges that skepticism assumes infallibilism or deductivism. We will see that these responses depend on uncharitable readings of the skeptical arguments and
so are rightly classified as dismissive responses; if they engage the arguments at all, they do so only superficially.
My second task is to consider and reject non-epistemological diag2 Agent reliabilism has its historical roots in Aristotle, Aquinas, and Thomas Reid, among
others. More recently, versions of the position have been defended by Ernest Sosa, Knowledge
in Perspective (Cambridge: Cambridge University Press, 1991); Alvin Goldman, Liaisons: Philosophy Meets the Cognitive and Social Sciences (Cambridge, MA: MIT Press, 1992) and Alvin
Plantinga, Warrant and Proper Function (Oxford: Oxford University Press, 1993).

noses of skepticism. These diagnoses often engage skeptical arguments

quite seriously ultimately fail to understand them correctly. For example,
skeptical arguments can be reconstructed so that they do not presuppose
a modern ontology. Neither do such arguments depend on representationalism, or some version of a traditional theory of ideas. Neither do
they depend on realism. Again, the lessons that skeptical arguments teach
are about the nature of knowledge and evidence, rather than ontology
or the philosophy of mind.
These first two tasks constitute indirect defenses of my thesis about
the nature of skeptical arguments the thesis that such arguments run
on plausible assumptions about the nature of knowledge and evidence.
My third task is to make good on this claim by analyzing some historically prominent skeptical arguments, and by showing exactly what epistemological mistakes they do in fact make. Here I focus on four arguments in particular: the argument from an infinite regress of reasons,
from Aristotle's Posterior Analytics; the argument for skepticism about the
world, from Descartes' Meditations on First Philosophy; the argument for
skepticism about the world, from Section XII of Hume's Enquiry concerning Human Understanding; and the argument for skepticism about unobserved matters of fact, from Section IV of the same work.
The task here is to look at these arguments as they actually appear in
their texts, but also to reconstruct them so as to bring out their real
force. The methodology I defend instructs us to put the arguments in
their most powerful form rather than rely on historical particularities to
score hollow victories. For example, Descartes begins his skeptical inquiry by doubting the validity of all of his knowledge at once. But is
there anything essential in Descartes' reasoning that requires him to
present it that way? I argue that there is not, and so rejecting it on that
basis does not give us insight into the real force of Descartes' skeptical
considerations. Likewise, Hume's skeptical arguments are couched in
terms of his empiricist theory of ideas. But his reasoning does not
essentially depend on that theory, and therefore Hume's arguments
cannot be rejected on that basis.
This puts me in position to undertake a fourth task. Having identified
several plausible but mistaken assumptions about the nature of knowledge and evidence, I offer a theory that shows why these are mistakes.
The idea is that a theory of knowledge and evidence that explains why
important skeptical arguments go wrong, and that therefore preserves
our common sense intuitions about what we do and do not know, is

made plausible by virtue of having those features. This makes good on

my second thesis: that the analysis of skeptical arguments is philosophically useful because it drives positive epistemology.
My fifth task is to argue that the methodology I defend can be
extended to moral and religious epistemology. In this way I continue to
defend the thesis that skeptical arguments repay analysis, now by driving
us to deeper understanding of moral and religious knowledge. Here the
methodology can be extended in two ways. First, it can be extended
directly, by applying it exactly as we do in the investigation of empirical
knowledge. When doing empirical epistemology, we assume that we
have knowledge, and we use skeptical arguments to root out assumptions that entail that we do not. In this way we uncover mistaken
assumptions about the nature of empirical knowledge and evidence. This
same methodology can be applied in moral and religious epistemology
as well. For example, we can start with the assumption that we do have
moral knowledge. We can then use skeptical arguments that conclude
otherwise as heuristic devices for rooting out mistaken assumptions
about the epistemology of moral beliefs. The methodology is as legitimate here as it is in the empirical realm. It is implausible to claim that I
do not know that here is a hand, and any argument that concludes that
I do not is almost certainly mistaken somewhere. But it is equally
implausible to claim that I do not know that killing innocent children is
wrong. If a moral epistemology entails that I do not, then we have good
reason to think that there is something wrong with that moral epistemology.
The methodology can also be extended indirectly. In this case we do
not assume that we have moral or religious knowledge, or even that
some of these beliefs are more reasonable than others. Rather, we examine arguments for moral and religious skepticism, and we look for
assumptions that, if true, entail that there is no empirical knowledge. If
an argument for skepticism in the moral or religious realm can be shown
to involve such an assumption, then we are warranted in rejecting it for
that reason. Some arguments against the rationality of religious belief
have exactly this character. Such arguments are meant to show that
rational religious belief is impossible. But if these same arguments entail
that I am not rational in believing that here is a hand, then something
in them is almost certainly wrong. Accordingly, they do not have force
against the rationality of religious belief. Some arguments against moral
perception have an analogous character: if they were correct, they would
show that moral perception is impossible, but they would also show that

empirical perception is impossible. Since the latter claim is implausible,

we have reason for rejecting the relevant objection to moral perception.
That is my five-point strategy for defending the three major theses of
the book. The outline of the book is as follows: In Chapter 2 I look
closely at two important arguments for skepticism about the world, one
from Descartes and the other from Hume. The purpose here is to
reconstruct the arguments in their most powerful forms. I conclude that
the strongest version of Descartes' argument is an inarticulate version of
Hume's. The main idea is that there is no good inference (deductive or
inductive) from the way things appear to the way things are. Put another
way, there is no good argument, not even an inductive one, from
appearance to reality. Far from relying on implausibly high standards for
knowledge, this argument offers powerful considerations for the conclusion that we lack even inductive knowledge of the world.
Another version of Descartes' argument, less powerful but still formidable, trades on the plausible assumption that knowledge must discriminate truth from alternative possibilities. For example, it is claimed
that I cannot know that I am sitting by the fire if I cannot discriminate
this state of affairs from the possibility that I am a disembodied spirit
deceived by an evil demon. I suggest that a "relevant possibilities"
approach is promising here. The central idea is that, intuitively, knowledge requires only that we can discriminate among some possibilities,
while others can be ignored as irrelevant. But such a response requires
development. We would like a principled account of what makes some
possibilities relevant and others irrelevant. Such development is undertaken in Chapter 8.
In Chapter 3, I consider and reject several dismissive responses to the
reconstructed arguments from Chapter 2. These include charges of selfrefutation, as well as several other responses based on pragmatic and
rhetorical considerations. I claim that all of these responses miss the
mark, since the skeptical arguments retain their force even if there are
no skeptics to put the arguments forward. What gives skeptical arguments their force is not that some other person is willing to defend them.
Rather, it is that they begin from premises that we are inclined to accept,
and that seemingly entail conclusions that we do not accept. If this is
right, then skeptical arguments are a problem for us, and whether or not
there are any "real" skeptics to defend them. Pragmatic and rhetorical
considerations are therefore irrelevant for adequately answering skeptical
arguments. They would be relevant if the problem of skepticism involved some skeptic, at whom such considerations could be directed. But

if the problem of skepticism is the problem of analyzing skeptical arguments, such considerations are simply irrelevant.
Another kind of dismissive response considered in Chapter 3 charges
that the standards for knowledge assumed by skeptical arguments are too
high, requiring, for example, deduction from evidence or absolute certainty for premises. Here I argue that the arguments reconstructed from
Hume and Descartes require no such thing, and that a close analysis of
the arguments reveals this. Finally, several versions of transcendental
arguments are considered as dismissive responses to skepticism, and all
are rejected as inadequate.
In Chapter 4, I consider some non-epistemological responses to "no
good inference arguments" for skepticism about the world. These include the diagnosis that skepticism requires a dualism between knowing
mind and material object of knowledge. Alternatively, some philosophers see representationalism as the driving force behind skeptical arguments, while others claim that realism is the problem. I reject all of these
diagnoses, arguing that "no good inference" arguments can be reconstructed without dualism, representationalism, or realism. In fact, even
Berkeley's radical idealism is consistent with a charitable reconstruction
of the skeptical argument.
These various non-epistemological diagnoses of skepticism are rejected in favor of an epistemological one. Specifically, I contend that
"no good inference arguments" misunderstand the way that sensory
appearances act as evidence for beliefs about the world. What these
arguments assume, and what many non-skeptical philosophers assume
with them, is that all evidential relations are inferential relations. In
other words, they assume that sensory appearances can act as evidence
for beliefs about the world only if the latter are inferred from the
former. But since no such inference is forthcoming, the arguments
conclude that appearances cannot give rise to knowledge of the world.
My position is that skeptical arguments are correct in claiming that
there is no good inference from appearance to reality and are therefore
wrong in claiming that beliefs about the world must be inferred from
sensory appearances. The latter is the plausible but ultimately disastrous
assumption that many skeptical arguments and many non-skeptics mistakenly share.
In Chapter 5, I consider the ancient skeptical argument from an
infinite regress of reasons. This argument contends that all knowledge
requires justification by adequate evidence, and that all such justification
involves inference from good reasons. But since all good reasons require

10

further reasons for their evidence, knowledge requires an infinite regress

ofjustifying reasons, and therefore no knowledge is possible. I begin by
revisiting some dismissive responses to skepticism and showing that these
miss the mark against the regress argument as badly as they do against
Cartesian and Humean arguments. For example, the argument does not
trade on a requirement for infallible reasons, or irrevisable ones. Second,
I look at the three most popular non-dismissive responses to the regress
argument: foundationalism, coherentism, and contextualism. Here I argue that coherentism is psychologically implausible, and that plausible
versions of contextualism reduce to foundationalism. Accordingly, I
defend a contextualist version of foundationalism one that is not open
to the usual objections to the foundationalist position.
"Contextualist foundationalism" might sound like an oxymoron, but
only because insufficient attention has been given to what foundationalism requires. Here again a close analysis of skeptical arguments proves
instructive, in this case showing what foundationalism does and does not
require to stop the infinite regress of justifying reasons. For example, it
does not require that foundational beliefs be infallible, incorrigible, devoid of contextual or social features, or even irrevisable. What it does
require is, once again, that not all evidential relations are inferential.
Perhaps all knowledge must be grounded in evidence, but some knowledge is not inferred from other beliefs. This is the defining characteristic
of foundationalism and is what is needed to answer the age-old infinite
regress argument.
In Chapter 6 I consider Hume's skeptical arguments regarding unobserved matters of fact. Here I distinguish two arguments one from
Section IV of the Enquiry and one from Section VII. I argue that the
former is the most powerful and is immune to dismissive responses that
are, at best, effective only against the latter. For example, the argument
is in no respect dependent on Hume's empiricist theory of ideas. I also
argue against the most popular interpretation of Hume, namely, that he
is a deductivist regarding knowledge of the unobserved. On my interpretation Hume's claim is that our beliefs about unobserved matters of
fact are not even inductively supported by past observations. On this
view Hume does not require that our evidence be deductive, but that
there be some necessary inductive relationship between our evidence
and our conclusions. This makes Hume's argument far more powerful
than is usually supposed. No one thinks nowadays that all evidence must
be deductive, and if Hume's argument ran on that assumption, then it
could be easily refuted. But it is seemingly obvious that evidence must

11

be at least inductively relevant to give rise to knowledge. This is in fact

false and is the real lesson of Hume's skepticism about unobserved
matters of fact. It is also a good illustration of the claim that skeptical
arguments drive positive epistemology. If we did not see its skeptical
consequences, we would hardly be inclined to reject such a commonplace, seemingly innocent assumption. But once focused on the real
force of Hume's argument, we see that the assumption must go in favor
of some better understanding of the nature of evidence.
Chapters 4 through 6 constitute a kind of negative epistemology:
rather than saying what knowledge is, negative epistemology largely
restricts itself to saying what knowledge is not. This is a worthwhile
activity, in that it disabuses us of plausible but mistaken assumptions
about the nature of knowledge and evidence. But the ultimate goal of
the methodology I defend is to construct a positive epistemology, or a
positive account of what knowledge is and how evidence works. This
happens when we construct a theory of knowledge that explains the
largely negative conclusions drawn from our analyses of skeptical arguments. This project is undertaken in Chapter 7.
Notice that the progress from negative to positive epistemology corresponds to three degrees of success we might have in refuting a skeptical
argument. The first and least satisfying degree of success is to find an
assumption in the argument that we need not accept. If we see that an
assumption leads to unacceptable consequences, and if there is no overwhelming reason to accept the assumption, then we are warranted in
giving it up as mistaken. For example, suppose we identify as disastrous
the assumption that knowledge of objects must be inferred, deductively
or inductively, from knowledge of how objects appear. This assumption
is something we need not accept, and so we should give it up once we
see where it leads.
In the next degree of success we arrive at some reason why the
skeptical assumption is mistaken. For example, we conclude that not all
evidential relations are inferential, and that this explains why sensory
appearances can be evidence for a belief even if the latter is not inferred
from the former. This seems to be a plausible position, and we might
look for further confirmations of it in cases unrelated to the skeptical
argument at hand. But even at this second degree we do not have an
explanation of a particular sort. What we really want is to have a theory
of knowledge that explains why the skeptical assumption is false. In
other words, we want an account that tells us what knowledge is and

12

how evidence works, and which thereby provides a theoretical explanation of why the assumption in question is mistaken.
In Chapter 7, I defend a virtue theory of knowledge that does just
this. Again, by a "virtue theory" I mean one that makes the cognitive
faculties and habits of persons central in the analysis of important epistemic concepts. As we saw, the theory is a version of reliabilism, in that a
stable disposition of a person counts as a virtue only if it is reliably
successful in achieving its end. In the case of the cognitive virtues, this
means that the faculty or habit makes the person reliable in forming true
beliefs of the kind relevant to the virtue in question. Agent reliabilism
explains why the skeptical assumptions rejected in earlier chapters are
false. Namely, it explains (a) why not all evidence is inferential, (b) how
sensory appearances can function as evidence without functioning as
premises in an inference, (c) how some knowledge can be foundational,
and (d) how propositional evidence that is neither logical nor quasilogical can give rise to knowledge.
We saw earlier that one of Descartes' skeptical arguments trades on
our inability to discriminate among various alternative possibilities. For
example, if one's evidence does not discriminate between being in front
of the fire and being a disembodied spirit deceived by an evil demon,
then one cannot know that one is sitting in front of the fire. A promising
strategy in response to this kind of skeptical reasoning is to distinguish
between relevant and irrelevant alternative possibilities, and to claim that
knowledge requires only that we discriminate among the relevant ones.
The problem, then, is to give a theoretical account of what makes an
alternative possibility relevant or irrelevant. In Chapter 8, I argue that
agent reliabilism can do just this.
The main idea is that virtues in general are abilities to achieve some
result, and abilities in general are functions of success in relevantly close
possible worlds. In other words, to say that someone has an ability to
achieve X is to say that she would be successful in achieving X in a
range of situations relevantly similar to those in which she typically finds
herself. But then possibilities that do not occur in typical situations are
irrelevant for determining whether a person has an ability in question.
For example, it does not count against Babe Ruth's ability to hit baseballs that he cannot hit them in the dark. Likewise, it does not count
against our perceptual abilities that we cannot discriminate real tables
and fires from demon-induced hallucinations. But then our inability to
rule out hypothetical demon scenarios is irrelevant to whether we have

13

knowledge, and the skeptical scenario is not a relevant possibility in that

sense. This account of "relevant possibility" confirms the plausibility of
agent reiiabilism and deepens our understanding of what we must mean
by a cognitive virtue or ability.
Finally, in Chapter 9, I argue that the methodology I am defending
can be extended to religious and moral epistemology. Many arguments
against the rationality of religious belief trade on assumptions about
knowledge and evidence that, if true, would count against rational belief
and knowledge in the empirical realm. By exposing such assumptions
and rejecting objections to religious belief on that basis, the epistemology
of religious belief is advanced. In the first part of Chapter 9 I endorse
recent work by Alvin Plantinga and William Alston as instances of
exactly this methodology, and I offer some suggestions about how their
views might be further defended along this line.
In the second part of the chapter I consider the possibility of moral
perception. Here I argue that recent work in the epistemology and
psychology of empirical perception opens up possibilities for moral perception. Specifically, to avoid skepticism about the natural world we
must understand empirical perception as a non-inferential cognitive faculty, but one that is nevertheless influenced by background beliefs,
special training, and the like. Second, we must have an account of how
complex, dispositional properties can be objects of empirical perception.
Accounts of how these features are possible for empirical perception
suggest promising extensions to moral perception as well. For example,
recent theory concerning the roles of personae and scripts in empirical
perception suggests interesting applications to moral perception.
Chapters 4 through 8 suggest a moderate foundationalism. The main
idea is that knowledge and justified belief arise from the cognitive
abilities of reliable believers, and that some of those abilities must be
characterized as non-inferential. The combined results of Chapter 9
suggest that the foundations of knowledge are broad as well. In other
words, they suggest that there exists a wide variety of non-inferential
sources of evidence, including evidence for moral and religious beliefs.
What emerges is a broad and moderate foundationalism in which much
evidence is non-inferential, and where even inferential evidence is seldom deductive or inductive in a sense that is commonly supposed. What
matters for knowledge and justified belief is not the infallibility or incorrigibility of our premises, nor even the logical or quasi-logical relations
among our premises and conclusions. Rather, knowledge and justified

14

belief arise from virtuous belief formation, where the notion of "virtue"
must be understood in terms of the contingent causal and motivational
features of our cognition, rather than the necessary or intrinsic features
of propositions, evidential relations, or the like.
3. THREE CRITERIA FOR AN ADEQUATE THEORY OF
KNOWLEDGE

Before closing this first chapter I want to talk about one more methodological issue. Specifically, I want to suggest three criteria for an adequate theory of knowledge, and to talk about how they are related to
the methodology for epistemology that I have been proposing.
First, an adequate theory of knowledge should do a good job of
organizing our pre-theoretical intuitions about what cases count as
knowledge. In other words, the theory should count as knowledge those
cases that intuitively seem to be knowledge, and it should count as not
knowledge those cases that intuitively seem not to be. "But whose
intuitions count as the right ones?", our suspicious friends will ask. The
answer is the intuitions of us all in our non-philosophical lives. An
adequate theory of knowledge should explain why normal people, not
people caught in the grip of a philosophical theory, count particular
cases as knowledge and other cases as not. "Normal people" includes
most non-philosophers, and most philosophers when they are not philosophizing. Among such people there is in fact very wide agreement
about which cases do and do not count as knowledge.
Moreover, universal agreement is not required for the methodology
being proposed. What is necessary is that there is a wide range of cases
that most people would find intuitively obvious. If a theory of knowledge does a good job with these, then that is a strong consideration in
its favor. However, vague and contested cases are important as well; a
good theory of knowledge should explain why certain cases are vague
and why certain ones are contested. Consider vagueness first. There will
be cases where we are not sure what to say where we have no strong
intuition about whether the case is one of knowledge or not. If our
theory can identify some aspect of the case that is vague as described and
can tell us that just this aspect is important for knowledge, then that will
count in favor of the theory. For example, we saw that agent reliabilism
requires that knowers be reliable in forming true beliefs and avoiding
error in a relevant domain of inquiry. But how reliable must one be to

15

have knowledge? This might be left vague by the theory, or might be

left vague in the description of a particular case. Either way, this could
account for the vagueness of an intuition.
Finally, a good theory of knowledge should account for disagreement
among intuitions as well. For example, if knowledge requires reliability,
and if we disagree over whether someone is reliable in a particular case,
then we might disagree over whether the case is one of knowledge. To
say that a theory of knowledge should account for our intuitions, then,
does not require that our intuitions lack vagueness or enjoy unanimity.
A good theory should account for agreement where we agree, disagreement where we disagree, and should explain why vagueness arises when
it does.
The second criterion for an adequate theory of knowledge is that it
be immune to skeptical arguments. This criterion is related to the first,
because our pre-theoretical intuitions are overwhelmingly non-skeptical.
Any theory that entails that there is no knowledge of objects in the
world, or no knowledge of unobserved facts, or no moral knowledge
does a horrible job with our pre-theoretical intuitions about what cases
count as knowledge. This is so for all of us, since none of us are skeptics
in our everyday lives.
For this reason, a theory that has radically skeptical consequences does
not capture the concept of knowledge that is actually in use - or at least
there is a very strong presumption that it does not. It is, we may suppose,
possible in principle that our ordinary concept of knowledge has widely
unrecognized skeptical consequences. But if there are competitor accounts that do not have such consequences, then that is an almost
insurmountable advantage of those accounts. Again, this follows from
the first criterion of an adequate epistemology: that it capture our pretheoretical intuitions about which cases count as knowledge.
We must take the qualification of the previous paragraph seriously,
however, in light of a sophisticated version of skepticism recently suggested by Robert Fogelin.3 Sophisticated skepticism claims that it can
explain our non-skeptical intuitions, by virtue of certain aspects of our
linguistic practices. The idea is that practical purposes make it appropriate to assert knowledge-claims in a wide range of cases, and that this
linguistic fact is behind our pre-reflective intuitions that such claims are
true. However, at least many of our knowledge-claims are literally false.
Moreover, in contexts where practical considerations are put aside and
3

See Fogelin, Pyrrhonian Reflections on Knowledge and Justification.

16

our level of scrutiny regarding knowledge-claims is raised, our intuitions

become skeptical, as when skeptical arguments are made or skeptical
considerations pushed. In light of this Fogelin-type sophisticated skepticism, we cannot claim that skeptical arguments, by their very nature,
cannot account for our pre-theoretical intuitions about which cases
count as knowledge. Rather, we have to put emphasis on the claim that
a non-skeptical theory of knowledge is strongly preferable, other things
being roughly equal. That is, if a non-skeptical theory of knowledge is
available that explains why the majority of our intuitions are true, this
will be preferable to a skeptical theory, and even if that theory comes
with an explanation as to why what seems obviously true is in fact false.4
One reason that the non-skeptical theory will be preferable, other
things being equal, is that it does not need the extra explanation. Another reason is that our non-skeptical intuitions are both strong and
persistent. Fogelin is correct that our intuitions are to some extent
unstable that under the pressure of skeptical arguments it can seem to
us that we do not know such things as that here is a hand or this is a
pencil. But these fleeting moments of doubt should not be overemphasized. On the other side are the persistent and overwhelming
intuitions of common sense, even among those philosophers who are
convinced by skeptical arguments in the study.
For example, consider G. E. Moore's famous statement that he knows
that here is a hand, made when his hand was held up in clear view. Any
theory that entails that I do not know such a thing, even with an
explanation of why it seems obvious that I do know, has its work cut
out for it. In this respect, any skeptical theory will be in the position
Russell's was in when Moore wrote the following:
What I want, however, finally to emphasize is this: Russell's view that I do not
know for certain that this is a pencil or that you are conscious rests, if I am
right, on no less than four distinct assumptions. . . . And what I can't help asking
myself is this: Is it, in fact, as certain that all these four assumptions are true, as
that I do know that this is a pencil and that you are conscious?5

I say "Fogelin-type" skepticism because Fogelin's Pyrrhonian principles do not allow him
to actually endorse any philosophical theory, including skepticism with respect to ordinary
knowledge claims. Therefore, the skeptical position I have just described is not literally
Fogelin's, although his discussion strongly suggests it as a possible alternative to non-skeptical
theories.
G. E. Moore, "Four Forms of Scepticism," in Philosophical Papers (London: Allen and Unwin, 1959), p. 222.

17

The more radical the skeptical consequences of a position are, the more
strongly this point will hold.6
The third criterion for an adequate epistemology is that it be psychologically plausible. What I mean by this is that an adequate theory of
knowledge ought to be consistent with our common sense judgments
about our own cognitive abilities, and with our best cognitive science as
well. This too is related to the first two criteria, because an account that
is not psychologically plausible will generate skeptical arguments. We
can see this if we look at the most basic structure that any skeptical
argument must have.
Although skeptical arguments come in many shapes and sizes, all of
them can be boiled down to two essential premises: one stating that
knowledge requires that some condition or set of conditions be fulfilled,
and one stating that these conditions are in fact not fulfilled. More
formally, we have the following skeptical argument structure:
(SAS)

1. K=>C.
2. Not-C.
3. Therefore, not-K.

Any theory of knowledge that is psychologically implausible will generate an argument with this structure. First, any theory of knowledge
whatsoever will entail premises corresponding to premise (1) of (SAS),
since any such theory posits conditions that must be fulfilled in order to
have knowledge. But the fact that a theory is psychologically implausible
guarantees that it will generate a premise corresponding to (2): that is, a
premise stating that conditions laid down by the theory in question are
not satisfied by beings with our psychology. Accordingly, we will have
a skeptical argument amounting to a reductio ad absurdum of the theory
in question.
This points to an elaboration of the methodology I am defending.
Specifically, we are not restricted to the use of historically prominent
skeptical arguments and their reconstructions. We can make up new
arguments to demonstrate the mistaken assumptions of alternative accounts of knowledge, or even alternative solutions to skeptical problems.
This actually happens in contemporary epistemology it has happened,
6

For an extended argument that our intuitions are not better explained by Fogelin-type
warranted assertability maneuvers, see Keith DeRose, "Contextualism: An Explanation and
Defense," in John Greco and Ernest Sosa, eds., The Blackwell Guide to Epistemology (Oxford:
Blackwell, 1999).

18

for example, in recent arguments against coherentism.7 Coherentism was

originally proposed as a non-skeptical response to the argument from an
infinite regress of reasons. The idea is that reasons can be mutually
supportive, and so no infinite regress of reasons is necessary to ground
knowledge in adequate evidence. But a major problem with coherentism is that it fails to give a psychologically plausible account of the role
of experience in forming perceptual beliefs. Take, for example, my
current perceptual belief that it is raining. The coherentist contends that
all beliefs are supported by other beliefs that serve as its evidence. But
against this position, it is not psychologically plausible that I infer my
belief that it is raining from other beliefs that I have and that act as its
evidence. Surely I know that it is raining because I can see that it is.
The coherentist will have a story about perceptual knowledge, but it
will not be a psychologically plausible one. For example, the coherentist
might say that we unconsciously infer beliefs about objects from beliefs
about sensory appearances. But it is psychologically implausible that we
typically have beliefs about sensory appearances, much less infer beliefs
about objects from them. And even if we did typically have such beliefs,
what evidence do we have for the supposed unconscious inferences? I
certainly do not seem to make the relevant inferences in perception. On
the contrary, the empirical evidence suggests that such inferences are a
philosophical invention.
Where does the empirical evidence regarding our cognitive capacities
come from, and who gets to decide what that evidence makes psychologically plausible? On the view defended here, there are two principal
sources of empirical evidence: our own reflection and empirical psychology.
In many instances a philosophical position can be recognized as psychologically implausible as soon as the question is raised. For example,
it seems obvious upon reflection that we do not typically form beliefs
about sensory appearances; in the typical case, we form beliefs about
tables and trees, not about how tables and trees appear to us. Sometimes
a little experimentation can confirm what seems obvious upon reflection. As Thomas Reid points out, objects not in focus present a double
image. This is confirmed by placing your finger in front of your face
and then focusing on an object in the distance. Attention to appearances
will reveal that your finger presents a double image. Alternatively, if you
7

For several examples of this kind of critique of coherentism, see John Bender, ed., The
Current State of the Coherence Theory (Dordrecht: Kluwer, 1989).

19

focus on your finger, the object in the distance will present a double
image. Reid argues that we must be presented with double images
almost all of the time, since almost always some objects are out of focus
for us. And yet we do not notice this, confirming the point that we do
not typically form beliefs about sensory appearances.8 This fact constitutes a devastating objection to coherentism, or to any other epistemology on which empirical knowledge requires beliefs about appearances.
For since we do not typically have such beliefs, any such theory has the
consequence that we typically lack empirical knowledge.
Finally, we can learn about our cognitive abilities from more rigorous
empirical research. For example, some coherence theories have a "total
evidence" requirement for knowledge and justified belief, laying down
a requirement that rational belief acquisition must be sensitive to the
total evidence that the person has at the time. But empirical research
shows that people are sensitive only to a small number of their total
beliefs at any one time. Here again we see a devastating empirical
objection to an epistemological theory; if a theory requires sensitivity to
all of the beliefs we have, and if our cognition is not capable of that
kind of sensitivity, then the theory has unacceptable skeptical results. In
this case the theory results in total skepticism, since the psychologically
implausible requirement is a completely general one.9
The methodology that I am defending here is an extension of what
Roderick Chisholm calls ' 'particularism. "10Chisholm argues that we
should follow philosophers like Reid and Moore in testing philosophical
8

"Thus you may find a man that can say, with a good conscience, that he never saw things
double all his life; yet this very man, put in the situation above mentioned, with his finger
between him and the candle, and desired to attend to the appearance of the object which
he does not look at, will, upon the first trial, see the candle double, when he looks at his
finger; and his finger double, when he looks at the candle. Does he now see otherwise
than he saw before? No, surely; but he now attends to what he never attended to before.
The same double appearance of an object hath been a thousand times presented to his eye
before now, but he did not attend to it; and so it is as little an object of his reflection and
memory, as if it had never happened." Thomas Reid, Philosophical Works, ed. H. M.
Bracken, 2 vols. (Hildesheim: Georg Olms, 1983), vol. 1, p. 164b.
9 The point about total evidence requirements is made by Alvin Goldman in Epistemology
and Cognition (Cambridge, MA: Harvard University Press, 1986), pp. 204-207. There
Goldman cites John Anderson, The Architecture of Cognition (Cambridge, MA: Harvard
University Press, 1983), and Christopher Cherniak, "Rationality and the Structure of Human Memory," Synthese 57 (1985): 163-186. Goldman makes the point specifically against
coherentism in his essay "Bonjour's The Structure of Empirical Knowledge," his contribution
to Bender, The Current State of the Coherence Theory, p. 112.

10 Roderick Chisholm, "The Problem of the Criterion," in The Foundations of Knowing (Minneapolis: University of Minnesota Press, 1982).

20

theories against pre-philosophical intuitions about particular cases. If a

theory has terribly counterintuitive results, then so much for the theory.
I am following this basic formula but extending it in four ways. First, I
am following Chisholm in holding that a theory should account for
intuitions we find clear and obvious, but I am suggesting that a good
theory should account for disagreement and vagueness among our intuitions as well. Second, I am suggesting that we use skeptical arguments
both historical ones and ones we make up to bring out the counterintuitive results of philosophical theories and to focus our unblinking
attention on those results. Third, I am suggesting that facts about human
cognition as well as intuitions about particular cases can serve to counter
philosophical theories. If a theory runs up against our intuition that we
know that objects in the world exist, then that counts strongly against
the theory in question. But if a theory runs up against our best empirical
knowledge, either from science or from our own experience, then here
too we have a reason for giving up the theory. I have already pointed
out that these two extensions of the methodology are not unrelated; if a
theory of knowledge is psychologically implausible, it will generate a
skeptical argument with counterintuitive results.
Fourth, I am proposing that implicit assumptions about the nature of
knowledge and evidence, as well as explicitly defended theories, should
be tested in these ways. Such assumptions in fact amount to philosophical theories, even though they are not defended as such and even though
they are the common property of theoreticians and non-theoreticians
alike.
Because the methodology that I propose here is a natural extension
of what Chisholm calls "particularism," I will call it "strong particularism." This methodology will seem obviously right to some and obviously wrong to others. On the one hand, it is the methodology that
many contemporary analytic epistemologists take for granted. On the
other, this is one reason why analytic epistemology seems so strange to
many outside of it. Many philosophers will think that the methodology
is hopelessly question begging, both in its invocation of our nonskeptical intuitions and in its use of results from the empirical sciences.
Others will think that it is too conservative, guaranteeing, in effect, that
our theories do not challenge our pre-philosophical judgments. But one
burden of this book is to defend strong particularism. This will be done
in two ways.
First, as with any methodology, this one must be judged by its fruits.
In the following chapters I will try to show that real progress can be

21

(and has been) made in philosophy by employing it. Second, the adequacy and appropriateness of strong particularism largely depend on how
we conceive of skeptical arguments and how we conceive of the project
of engaging them.
If we think of skeptical arguments as coming from real people whom
we are to engage as opponents in a debate, then the use of our own
intuitions will be contested and the use of empirical psychology will
seem question begging. So will the use of these seem pointless if we
think that our project is to persuade someone out of her skepticism by
mounting a convincing argument. In such a case we would have to start
from premises already accepted, and a skeptic would accept neither our
non-skeptical intuitions nor the results of empirical sciences. But I will
argue in Chapter 3 that these conceptions of the epistemological project
are misconceived and misguided. There are no real skeptics, either to be
debated or to be persuaded out of their skepticism. In other words, there
is no one who actually lives out the skeptical position, or who even
believes it outside the study or the classroom. And even if there were,
debate and persuasion would be hopeless. A debate with a consistent
skeptic cannot be won, especially if we adopt the ground rules that
philosophers are wont to concede. Moreover, if we are to persuade
someone out of her skepticism, then argument and philosophy are the
last thing she needs. Here Reid had it right when he considered the
appropriate response to a friend who was found really to be a skeptic:
"[Would we] not hope for his cure from physic and good regimen,
rather than from metaphysic and logic?"11
In any case, that is not what skeptical arguments or their analysis are
about. There is no practical or "existential" problem of skepticism, since
the position is not one that any person can live. Rather, skeptical
arguments constitute theoretical problems; they start from assumptions
about knowledge and evidence that we ourselves find plausible or even
accept, and they show that such assumptions lead to consequences that
we cannot accept. As a result we see that some implicit or explicit
theory of knowledge is wrong, and we are pushed to develop new and
unexpected positions to replace the mistaken assumptions that skeptical
arguments expose. If this is the point of engaging skeptical arguments,
then it is perfectly reasonable to use our best intuitions and our best
science to construct and test alternative positions. We are trying to put
together the most plausible account of knowledge and evidence by our
11

Reid, Philosophical Works, p. 100b.

22

own lights. If that is our project, then we should use all of the resources
that are available to us.
If this is our understanding of skeptical arguments then begging the
question does not come into it there is no one to beg the question
against. However, our considered intuitions and science do come into
it, for they are the best that we have to go on. As Chisholm puts it,
"This may seem the wrong place to start. But where else could we
start?"12 Here is another way to gloss Chisholm's point: If a theory of
knowledge does not explain our considered intuitions about knowledge,
what else could it have going for it? What kind of evidence could we
bring to bear to show that the theory is correct?
I have been arguing that we must start with our intuitions. Is that
where we must end? In other words, does strong particularism guarantee
that skepticism is false? Or worse, does it guarantee that we can never
reach any conclusion but a conservative one that is, one that confirms
our pre-theoretical intuitions? Well, yes and no. First, there is no guarantee that we will reach any conclusion at all. In other words, strong
particularism does not guarantee that we can account for our pretheoretical intuitions with an adequate theory; it merely lays down
criteria for theoretical adequacy. On the other hand, if we do reach a
theoretical conclusion by using the method, then that theory must
preserve at least a great many of our pretheoretical intuitions. But that
does not mean that an adequate theory must preserve all of our intuitions, since some of these might best be revised along the way. What is
more, they might be revised in very important ways.
This consideration can be used to address another kind of objection
to strong particularism namely, that if a theory merely preserves our
intuitions about knowledge and evidence, then it cannot be used to
criticize them. I would reply by insisting that just the opposite is the
case: that a theory can be used to criticize our intuitions only if it does
preserve a great many of them. The point is that a theory can have a
critical function only if we think it is correct. And again, the only reason
we could have for thinking that a theory of knowledge is correct is that
it does a good job of explaining how we know the things that we seem
to know. Granting that a theory of knowledge should have a critical
function, it can successfully critique some of our intuitions only if it does
a good job of explaining others. In the last chapter of the book this
12 Roderick Chisholm, The Theory of Knowledge, 2nd ed. (Englewood Cliffs, NJ: Prentice
Hall, 1977), p. 16. Chisholm is commenting on the adequacy of particularism here.

23

point is illustrated with respect to our intuitions about moral and religious knowledge. Against the moral and religious skepticism of our
times, I argue that an adequate theory of empirical perception opens up
the possibility of moral and religious perception. It becomes possible to
explain how we might "see" that an action is wrong, or "feel" God's
love and presence. This might seem absurd at first, but it follows
straightforwardly from an adequate that is, non-skeptical empirical
epistemology. Strong particularism, then, does not merely preserve our
pre-theoretical intuitions.

24

Skepticism about the World: Part

One Reconstructions

In this chapter, I consider two historically prominent skeptical arguments: one from Section XII of Hume's Enquiry concerning Human Understanding, and the other from Meditation I of Descartes' Meditations on
First Philosophy. My purpose is to reconstruct the arguments so as to put
them in their most powerful form. But we should remember that by
"powerful" I do not mean psychologically convincing. I do not think
that either skeptical argument, even in its most powerful form, has the
power to convince or persuade. Rather, the arguments can be reconstructed so that they are powerful in another sense: it is not at all easy to
see where they go wrong, and rejecting them requires us to give up
something that otherwise would seem plausible or perhaps even obvious.
Such arguments are powerful heuristic devices in that they drive us to
give up plausible but mistaken assumptions about knowledge and evidence, and inspire us to put something substantive in their place.
In the next chapter we will consider how several dismissive responses
to skepticism fare against the reconstructed arguments. Remember that
dismissive responses are ones that either (a) do not engage skeptical
arguments at all or (b) engage them only superficially. Such responses
are dismissive because they reject the skeptical conclusion without paying serious attention to the reasoning that leads up to it. Accordingly,
dismissive responses miss the lessons that skeptical arguments can teach.
1. AN ARGUMENT FROM HUME
Descartes' argument from Meditation I has probably received more
attention than any other skeptical argument and surely has been the
target of the most abuse. However, I want to start with the argument

25

from Hume, which I think is in some respects the stronger of the two.
More exactly, I argue that the most powerful interpretation of Descartes'
argument makes it run on considerations brought out more straightforwardly by Hume's. In this sense the most powerful form of the argument
from Descartes reduces to the argument from Hume.
Consider the following passage from Section XII of Hume's Enquiry,
where he is considering our evidence for the existence of "external
objects" and "an external universe."1
By what argument can it be proved, that the perceptions of the mind must
be caused by external objects, entirely different from them, though resembling
them (if that be possible) and could not arise either from the energy of the mind
itself, or from the suggestion of some invisible and unknown spirit, or from
some other cause still more unknown to us? It is acknowledged, that, in fact,
many of these perceptions arise not from anything external, as in dreams,
madness, and other diseases. . . .
It is a question of fact, whether the perceptions of the sense be produced by
external objects, resembling them: how shall this question be determined? By
experience surely; as all other questions of a like nature. But here experience is,
and must be entirely silent. The mind has never anything present to it but the
perceptions, and cannot possibly reach any experience of their connexion with
objects. The supposition of such a connexion is, therefore, without any foundation in reasoning. (Enquiry, pp. 152153)
In the first paragraph it looks like we are getting the familiar "alternative
possibilities" reasoning: How can we know through perceptions that
external objects exist if those same perceptions could be caused by a
dream, or a spirit, or something else still more unknown? The last
sentence of the first paragraph suggests the old "sometimes we make
mistakes" argument: Sometimes people are deceived by dreams or madness or other diseases, and so how do I know I am not deceived now?
But I do not think that Hume has either of these arguments primarily in
mind. Rather, his reasoning is suggested by the cryptic second paragraph. The argument contained there is obscure, and we would probably
not guess Hume's meaning if we were not already familiar with his
reasoning in Section IV of the Enquiry regarding unobserved matters of
fact. But with that reasoning in mind we can pick out a forceful argument here, for the arguments in Section IV and Section XII have closely
analogous structures.
1

All references to Hume are from Enquiries concerning Human Understanding and concerning the

Principles of Morals, 3 ed. L. A. Selby-Bigge, 3rd ed. (Oxford: Clarendon Press, 1975).

26

The argument of Section IV goes roughly like this: All beliefs about
future matters of fact depend for their evidence on both (a) observations
of past cases and (b) the assumption that observed cases will be a good
indication of future cases, that is, on the assumption that the future will
resemble the past. Let us call this the "regularity principle," because it is
equivalent to saying that there is a regularity in nature. But now the
regularity principle is itself a belief about future matters of fact; it is an
assumption that in the future observed cases will continue to be a reliable
indication of future cases. As such, the only way that the principle could
be justified is by inference from (a) past observations and (b) the assumption that the future will resemble the past. In other words, we think that
observed cases will be a reliable indication of future cases because we
have observed in the past that nature has been regular in that way. But
this means that the only evidence we could have for the regularity
principle must include the principle itself. Such reasoning is blatantly
circular, however, and therefore cannot give rise to knowledge.
More formally, we have the following argument.
(HI)

1. All of our beliefs about future matters of fact depend for their
justification on past observations, together with the assumption
(Al) that the future will resemble the past.
2. But (Al) is itself a belief about a future matter of fact.
3. Therefore, assumption (Al) depends for its evidence on (Al). (1,2)
4. Circular reasoning cannot give rise to knowledge.
5. Therefore, (Al) is not known. (3,4)2

This conclusion is bad enough, but of course Hume's reasoning does

not stop there. Because the regularity principle is involved in all of our
reasoning concerning future matters of fact, the real conclusion is that
there is no knowledge of the future.3
6. All of our beliefs about future matters of fact depend on an assumption that is not known. (1,5)
2 My custom will be to set out various skeptical arguments with numbered premises. Numbers
in parentheses after a conclusion or subconclusion indicate the premises from which the
conclusion or subconclusion is supposed to be immediately entailed. Such entailments might
depend on the laws of logic, or might depend on the necessary relations of the concepts
involved in the premises and conclusions. In presenting the arguments in this manner I am
trying to capture the way that the skeptical reasoning is supposed to go. It is not my claim,
in other words, that conclusions and subconclusions are in fact entailed by the indicated
premises. I mean that to remain an open question.
3 Actually, the real conclusion is about all unobserved matters of fact, not just those in the
future. This point is made more carefully in Chapter 6, but for present purposes I treat
Hume's argument as if it were about our knowledge of future matters of fact specifically.

27

7. Beliefs that depend on an unknown assumption are themselves not

known.
8. Therefore, no one knows anything about the future. (6,7)
We will look at this argument more closely in Chapter 6. At present
my purpose is to use it to bring out the structure of Hume's reasoning
in Section XII. There Hume is concerned with our evidence for external
objects rather than unobserved matters of fact, but, as I have noted, the
two skeptical arguments have closely analogous structures.
Let us now look back at the passage that I quoted earlier from Section
XII. On the reading I am proposing Hume's first paragraph merely
points out that sensory appearances by themselves do not adequately
support our beliefs about objects in the world; the fact that objects
appear to be a certain way is consistent with their not being that way at
all. What is needed together with sensory appearances is the assumption
that such appearances are a reliable guide to external reality. In other
words, our evidence for our beliefs about objects in the world involves
both (a) the way things appear to us through the senses and (b) the
assumption that these sensory appearances are a reliable indication of
how objects in the world really are. We might not voice such an
assumption to ourselves explicitly, but this must be how we are thinking;
we are thinking not only that things appear a certain way, but that their
appearing that way is an indication that they are that way.
But now Hume notes, "It is a question of fact, whether the perceptions of the sense be produced by external objects." In other words, the
assumption on which all of our beliefs about external reality depend is
itself & belief about external reality. It is a belief that our sensory experience is caused by objects in the world that have the properties that our
experience presents them as having. But how could we know that this
assumption is true? "How shall this question be determined?" The
answer is clear: "By experience surely; as all other questions of a like
nature." But of course now we have another circularity problem. All of
our beliefs about the world depend on two things for their evidence:
sensory appearances and the assumption that sensory appearances are a
reliable guide to external reality. But that assumption is itself about
external reality, and so its evidence depends on itself.
Hume has identified another circle in our reasoning. Or perhaps his
point is just slightly different and goes like this: Either our beliefs about
the world are based on the assumption that sensory appearances are a
reliable guide to reality, or they are not. If they are not, then such beliefs

28

are unsupported, because appearances alone cannot do the job. If they

are, then such beliefs are based on circular reasoning. In either case, we
have no knowledge of the world.
Immediately following the second passage that I quoted, Hume considers an escape from the argument. It might be supposed, as Descartes
argued, that the reliability assumption is not in need of evidence from
sensory appearances but can rather be established a priori via an argument
about God's veracity. Hume responds to this line of reasoning as follows:
To have recourse to the veracity of the supreme Being, in order to prove the
veracity of our senses, is surely making a very unexpected circuit. If his veracity
were at all concerned in this matter, our senses would be entirely infallible;
because it is not possible that he can ever deceive. Not to mention, that, if the
external world be called in question, we shall be at a loss to find arguments, by
which we may prove the existence of that Being or any of his attributes.
{Enquiry, p. 153)

One point Hume is making here is that Descartes' strategy proves too
much; if God's veracity guaranteed the trustworthiness of our senses,
this would establish infallibility rather than a plausible reliability. But
two other points being made are more important for Hume's skeptical
reasoning. First, it is not in fact the case that we believe our senses are
reliable because we have knowledge of God's veracity. That would be
to make an "unexpected circuit." Second, even if we tried to mount
such an argument, we would fall back into circles; our knowledge of
God must presuppose our knowledge of the world and therefore cannot
be used to support it.
Let me take the opportunity to state Hume's argument more formally, in both of the versions I have suggested.
(H2)

1. All of our beliefs about the external world depend for their evidence on sensory appearances, together with the assumption (A2)
that sensory appearances are a reliable guide to external reality.
2. But (A2) is itself a belief about the external world.
3. Therefore, assumption (A2) depends for its evidence on (A2). (1,2)
4. Circular reasoning cannot give rise to knowledge.
5. Therefore, (A2) is not known. (3,4)
6. All of our beliefs about the world depend on an assumption that is
not known. (1,5)
7. Beliefs that depend on an unknown assumption are themselves not
known.
8. Therefore, no one knows anything about external reality. (6,7)

29

(H3)

1. Any belief about external reality either depends on (A2) for its
evidence or depends on sensory appearances alone.
2. If it depends on sensory appearances alone, then it is not adequately
supported.
3. If it depends on (A2), then it is supported by circular reasoning.
4. Evidence that is not adequately supporting cannot give rise to
knowledge.
5. Circular reasoning cannot give rise to knowledge.
6. Therefore, no belief about external reality amounts to knowledge.
No one knows anything about external reality. (1,2,3,4,5)

Before leaving Hume's argument I want to address one possible

objection to it. It might be suggested that the argument trades on a
particular understanding of sensory appearances. Namely, the argument
assumes that sensory appearances are merely causal antecedents to belief,
themselves devoid of any conceptual content. On this assumption, when
an object appears to a person through the senses, the person's sensory
experience does not represent the object as being of a particular kind or
as having particular properties. Rather, the experience is only a causal
antecedent to a belief that first represents the object that way. It is only
on this characterization of sensory appearances, the objection goes, that
premise (1) of (H2) seems plausible. If all seeing is "seeing as," however,
then it is plausible that sensory appearances alone make probable beliefs
about objects in the world. For example, a sensory appearance with the
conceptual content that some object is an apple tree makes probable the
perceptual belief that the object is an apple tree, and without needing
the additional assumption that sensory appearances are a reliable guide
to reality. Let us call sensory appearances conceived as having only
phenomenal content "thin experience," and appearances conceived as
having conceptual content "thick experience." The present objection is
that Hume's argument depends on characterizing sensory evidence as
thin experience.
This much should be granted to the objection: if we are thinking of
sensory appearances as devoid of conceptual content, then Hume's skeptical reasoning goes through straightforwardly. There is no good inference from something's appearing phenomenally a certain way, to a belief
that the thing is that way, if only because inferential support is a function
of conceptual content.4 What would be needed is a belief to the effect
4 This is not the only problem with this characterization of sensory evidence. There are also
well-known problems concerning the relationship between the phenomenal properties of

30

that something is appearing a certain way phenomenally, together with

the assumption that the way things appear phenomenally is a reliable
indication of the way things are. It is important to see, however, that
(H2) and (H3) run just as well on the assumption that sensory appearances do have conceptual content. Either way that we think of sensory
appearances, the argument goes through if we grant its other assumptions.
To see that this is so, assume that sensory appearances are thick.
When a person sees an apple tree as an apple tree, this is understood to
involve both a phenomenal aspect and a conceptual aspect, the latter
with the content that the object is an apple tree. On this assumption,
sensory appearances are always sensory "takings" or "seemings"; to have
a sensory experience of an apple tree is not only to be appeared to
phenomenally in a particular way, but to take the object appearing to be
an apple tree. Of course, not only apple trees can be seen as apple trees.
A cherry tree might seem to be an apple tree in this sense, especially if
one is not good at discriminating cherry trees from apple trees. Similarly,
an apple tree might seem to be something else - for example, seem to
be a man at the far end of a darkened field. But in any case, on this
view, sensory appearances always come interpreted they must be
understood as having a phenomenal content, but always a conceptual
content as well.
Even so, there appearing to be an apple tree (understanding sensory
appearances now as having both phenomenal and conceptual content)
does not make probable the belief that there is an apple tree, without
the assumption that, at least in general, the way things appear is a reliable
indication of the way things are. Alternatively, there "seeming visually"
to be an apple tree does not make probable the belief that there is an
apple tree, without the assumption that the way things seem visually is
generally a reliable indication of the way things are. Are such assumptions true? Every indication we have about our perceptual powers is that
the answer is yes. But Hume's point in (H2) and (H3) is not to challenge
the truth of these assumptions. Rather, his point is that if the assumptions
are truths, then they are contingent truths. If it is true that the way
things seem to be via the senses is a reliable indication of the way things
are, then this is a contingent truth about our cognition and the world
appearances and the external properties of objects. I am not concerned with this latter kind
of problem, since it does not arise for what I think is the more plausible account of sensory
evidence, i.e., the one in which sensory appearances are thick.

31

rather than a necessary truth about the concepts of sensory seemings and
reality. But if this kind of assumption is a contingent matter regarding
the way the world is, then our evidence for it must involve empirical
observation. In other words, our justification for this sort of assumption
must itself be grounded in sensory seemings, and therefore in the assumption that sensory seemings are a reliable indication of the way
things are.5
This suggests that Hume's arguments can be taken two ways. We can
interpret the arguments as being about sensory appearances understood
as having only phenomenal content. In this case the problem is that my
beliefs about the world depend for their justification on the way things
appear phenomenally, but appearances so conceived cannot support
those beliefs. But we can also interpret Hume's arguments as being about
sensory seemings or takings. Now the problem is that my beliefs about
the world depend for their justification on the way things seem to be
via the senses, but appearances so conceived still do not make my beliefs
probable by themselves. In both cases we need to add an assumption in
order to make the relevant appearances function as evidence for beliefs
about the world. And in both cases there seems to be no noncircular
justification for that assumption.
Here is Hume's reasoning reconstructed both ways.
(H2a)

1. All of our beliefs about the external world depend for their evidence on the way things appear phenomenally, together with the
assumption (A3) that something's appearing phenomenally a certain way is a reliable indication that it is that way.
2. But (A3) is itself a belief about the external world.
3. Therefore, assumption (A3) depends for its evidence on (A3).
(1.2)
4. Circular reasoning cannot give rise to knowledge.
5. Therefore, (A3) is not known. (3,4)
6. All of our beliefs about the world depend on an assumption that
is not known. (1,5)

It might be objected here that I am assuming an internalist theory of content. On externalist

theories, the objection goes, it is a necessary truth that the content of one's thick sensory
experience refers to an external reality. In reply, even on the strongest versions of externalist
theories of content, the necessary connection between content and world is only indirect
and tenuous. In other words, the connection does not guarantee the kind of reliability
between appearance and reality that is claimed in assumption (A2) of arguments (H2) and
(H3). Externalists about content do not dispute this. E.g., see Hilary Putnam, Reason, Truth
and History (Cambridge: Cambridge University Press, 1981).

32

7. Beliefs that depend on unknown assumption are themselves not

known.
8. Therefore, no one knows anything about external reality. (6,7)
(H2b)

1. All of our beliefs about the external world depend for their evidence on the way things seem to be via the senses, together with
the assumption (A4) that the way things seem to be is a reliable
indication of the way things are.
2. But (A4) is itself a belief about the external world.
3. Therefore, assumption (A4) depends for its evidence on (A4).
(1.2)
4. Circular reasoning cannot give rise to knowledge.
5. Therefore, (A4) is not known. (3,4)
6. All of our beliefs about the world depend on an assumption that
is not known. (1,5)
7. Beliefs that depend on unknown assumption are themselves not
known.
8. Therefore, no one knows anything about external reality. (6,7)

I end this section with a final remark regarding Hume's reasoning. A

number of philosophers have denied the plausibility of premise (1) of
(H2a). As we have just seen, their reasoning is that appearances conceived as having merely phenomenal content cannot be in the "logical
space of reasons" and therefore cannot function as evidence for beliefs,
although such appearances still might function as causal antecedents for
beliefs.6 But it seems to me that premise (1) of (H2a) is not wholly
implausible. For example, consider my perceptual belief that the cat is
sitting on the couch. Isn't my reason for believing this at least partly the
fact that things appear phenomenally a particular way? Clearly this is part
of the cause of my believing that the cat is on the couch, but isn't it at
least plausibly a reason in the epistemological sense? For example, isn't it
part of what allows me to know that the cat is on the couch? Of course
one might stipulate that "reasons" or "evidence" by definition involve
logical relations, but this would not affect the substantive point at issue.
We could still specify a broader sense of epistemic grounds, and it would
remain plausible that the way things appear phenomenally makes up at
least part of our epistemic grounds for perceptual beliefs.
Perhaps what these philosophers mean to claim is that appearances
6

This view is most famously defended by Wilfred Sellars for example, in "Empiricism and
the Philosophy of Mind," in Science, Perception and Reality (London: Routledge, 1963). A
somewhat similar view is defended in John McDowell, Mind and World (Cambridge, MA:
Harvard University Press, 1996).

33

conceived as having merely phenomenal content cannot be all of our

evidence for perceptual beliefs; that there must be some other aspect of
that evidence as well in order to generate knowledge or justified belief.
But that claim is consistent with Hume's reasoning in (H2a). In any
case, even if we reject (H2a) on the grounds that it involves an inadequate account of sensory appearances, which ultimately I think we
should, this still teaches an important lesson about the nature of sensory
evidence. Moreover, this line of objection does not offer an adequate
diagnosis of (H2b).
I turn next to Descartes' skeptical considerations in the Meditation I.
The argument presented there has been interpreted in many different
ways. I will argue that its most powerful version reduces to the reasoning
we have just seen from Hume.

2. AN ARGUMENT FROM DESCARTES

Although Descartes was no skeptic, he does present some powerful

skeptical considerations. Notoriously, he puts forward those considerations because he wants to discard all of his former opinions everything
that he has previously believed and even taken for granted. This is the
first step of a wider project in which Descartes tries to rebuild his
knowledge according to the "method for rightly conducting reason and
seeking truth in the sciences." Roughly, that method involves establishing a solid foundation of clear and distinct first truths and proceeding
from those only by demonstrations that are themselves clearly and distinctly valid. Many have thought that Descartes did a better job of
undermining his former opinions than he did of rebuilding his knowledge, and I concur with that judgment. But let us take a closer look at
how Descartes' skeptical reasoning is supposed to go. It is this first step
of his project that is of interest to our own investigation.
As I have just mentioned, Descartes raises doubts in Meditation I
because he wants to discard everything that he has formerly believed.
But he quickly notes that it would be impossible for him to find reasons
for doubting his beliefs one at a time. The more effective alternative
would be to find reasons for doubting the sources of his beliefs, thereby
implicating all of his beliefs at once, or at least large groups of them at

7 All quotations from Descartes are from The Philosophical Works of Descartes, vol. 2, trans.
Elizabeth S. Haldane and G. R. T. Ross (Cambridge: Cambridge University Press, 1979).

34

And for that end it will not be requisite that I should examine each in particular,
which would be an endless undertaking; for owing to the fact that the destruction of the foundations of necessity brings with it the downfall of the rest of the
edifice, I shall only in the first place attack those principles upon which all my
former opinions rested. (Meditations, p. 145)
The "principle" he hits upon, of course, is sensory experience. "All that
up to the present time I have accepted as most true and certain I have
learned either from the senses or through the senses; but it is sometimes
proved to me that these senses are deceptive" (p. 145).
This is the old "sometimes I make mistakes" reasoning again. But a
more important point is made next. Namely, it is possible to have the
same experience one does now even when one is only dreaming, and
when things are not the way they appear at all.
How often has it happened to me that in the night I dreamt that I found myself
in this particular place, that I was dressed and seated near the fire, whilst in
reality I was lying undressed in bed! At this moment it does indeed seem to me
that it is with eyes awake that I am looking at this paper; that this head which I
move is not asleep, that it is deliberately and of set purpose that I extend my
hand and perceive it; what happens in sleep does not appear so clear nor so
distinct as does all this. But in thinking this over I remind myself that on many
occasions I have in sleep been deceived by similar illusions, and in dwelling
carefully on this reflection I see so manifestly that there are no certain indications
by which we may clearly distinguish wakefulness from sleep that I am lost in
astonishment, (pp. 145-6)
At first Descartes considers the kind of dream that occurs in sleep, but
toward the end of the meditation he raises the possibility of something
much more extreme:
Nevertheless I have long had fixed in my mind the belief that an all-powerful
God existed by whom I have been created such as I am. But how do I know
that He has not brought it to pass that there is no earth, no heaven, no extended
body, no magnitude, no place, and that nevertheless [I possess the perceptions
of all these things, and that] they seem to me to exist just exactly as I now see
them? (p. 147)
It is possible for things to appear to me exactly as they do now even if
things are drastically different from what I believe them to be. It is
possible, for example, that things appear as they do, but that I am the
victim of a powerful and elaborate deception. Perhaps God deceives me,
or perhaps I am fooled by an "evil genius" a being "not less powerful

35

than deceitful," and who "has employed his whole energies in deceiving
me" (p. 148).
It is not obvious how the skeptical reasoning is supposed to go here,
but one point can be made right away: namely, Descartes never actually
believes that there might be an evil genius deceiving him at every turn.
Rather, he introduces the possibility of such a demon as a kind of
psychological tool; he wants to discard all of his former opinions but
finds this psychologically impossible. His solution is to pretend that there
is a powerful demon bent on deceiving him, so that he may keep in
mind his resolution to doubt his former beliefs and to use nothing that
is less than certain in his further deliberations.
But it is not sufficient to have made these remarks, we must also be careful
to keep them in mind. For these ancient and commonly held opinions still
revert frequently to my mind, long and familiar custom having given them the
right to occupy my mind against my inclination and rendered them almost
masters of my belief; nor will I ever lose the habit of deferring to them or of
placing my confidence in them, so long as I consider them as they really are,
i.e. opinions in some measure doubtful, as I have just shown, and at the same
time highly probable, so that there is much more reason to believe in than to
deny them. That is why I consider that I shall not be acting amiss, if, taking of
set purpose a contrary belief, I allow myself to be deceived, and for a certain
time pretend that all these opinions are entirely false and imaginary, until at last,
having thus balanced my former prejudices with my latter [so that they cannot
divert my opinions more to one side than to the other], my judgment will no
longer be dominated by bad usage or turned away from the right knowledge of
the truth.
I shall then suppose, not that God who is supremely good and the fountain
of truth, but some evil genius not less powerful than deceitful, has employed his
whole energies in deceiving me. (p. 148)
For this reason alone, much of the abuse that has been heaped on
Descartes' argument simply misses the mark. However misguided Descartes' project in the Meditations, he never seriously thought that his life
might be a dream orchestrated by an evil genius.
If Descartes never takes seriously the possibility that he is deceived,
then what is his line of reasoning in the passages we have been reviewing? I think it is essentially this: Descartes sets out to evaluate his evidence for the various beliefs he has long held, and he concludes that his
evidence is not very good. And the reason it is not very good is that it

36

fails to rule out other possibilities that are inconsistent with what he
believes. Let us take a closer look at this line of thought. 8
Consider Descartes' belief that he is sitting by the fire in a dressing
gown. Presumably he has this belief because this is how things are
presented to him by his senses. However, Descartes reasons, things could
appear to him just as they do even if he were in fact not sitting by the
fire but was instead sleeping, or mad, or the victim of an evil deceiver.
The point is not that these other things might well be true, or that they
ought to be taken seriously as real possibilities. Rather, it is that Descartes' evidence does not rule these possibilities out. And if it does not
rule them out, then it cannot be very good evidence for his belief that
he is sitting by the fire.
It should be emphasized that Descartes' reasoning follows from a
seemingly obvious principle about adequate evidence. Namely, a body
of evidence does not adequately support a conclusion unless that evidence effectively rules out other possibilities which are inconsistent with
that conclusion. For example, in murder cases the prosecutor must
present evidence that rules out suicide or another murderer. In scientific
investigations a hypothesis is confirmed only when the evidence effectively rules out competitors. What Descartes notices is that this general
principle has skeptical consequences when applied consistently to our
perceptual beliefs about objects in the world. Our evidence in such cases
is sensory experience, but that experience fails to rule out a host of
alternative possibilities.
Two points need to be emphasized here. First, alternative possibilities
to what I believe undermine my knowledge even if they are false. For
example, if I cannot rule out the possibility of suicide in a suspicious
death, then I cannot know that someone committed a murder. And that
is true even if there was no suicide even if the alternative possibility I
cannot rule out is in reality false. This is what gives the skeptical reasoning its force. For although there is no reason whatsoever to think that I
am in bed dreaming, or that I am the victim of a deceiving demon, it is
hard to deny that these are possibilities in some broad sense. That is to
say, there is nothing incoherent or logically impossible about them. But

Here and in the next two paragraphs I follow Barry Stroud, although I depart from him in
some respects. See Stroud, The Significance of Philosophical Scepticism. For a different reading
of Descartes' argument, see Frederick Schmitt, Knowledge and Belief (London: Routledge,
1992).

37

then, that is all that the skeptic needs. For knowledge demands that our
evidence rule out alternative possibilities, and these possibilities cannot
be ruled out. Second, Descartes' reasoning runs just as well whether we
are thinking of sensory appearances as thin or thick. The fact that things
appear phenomenally a certain way does not rule out the possibility that
I am dreaming, or that I am deceived by an evil demon. But neither
does the fact that things present themselves a certain way in thick sensory
seemings.
Here is Descartes' reasoning put more formally.
(Dl)

1. A person can know something on the basis of her evidence only if

that evidence rules out possibilities that are inconsistent with what
the person claims to know.
2. A person's evidence for her beliefs about the world is her sensory
experience.
3. It is a possibility that one is dreaming, and the possibility that one
is dreaming is inconsistent with one's beliefs about the world.
4. Therefore, a person can know things about the world only if her
sensory experience rules out the possibility that she is dreaming.
(1,2,3)
5. But a person's experience does not rule out the possibility that she
is dreaming.
6. Therefore, no one knows anything about the world. (4,5)

We now have a rough characterization of Descartes' skeptical argument.

But this characterization is still consistent with a number of interpretations, some of which do not have much force at all. What I want to do
now is to get the argument into its most powerful form. I will conclude
that there are two versions which are quite powerful in the sense defined
earlier: That is, there are two versions of the argument that make no
obvious mistake, and that therefore require us to say something philosophically substantive in refuting them. But the most powerful of these
two reduces to the argument we have already seen from Hume.
The most significant way in which we get different versions of (Dl)
is according to how we understand the idea of evidence "ruling out" an
alternative possibility. On some interpretations of this idea the argument
has little or no force. On others it is hard to see where the argument
goes wrong. Before we look at these different interpretations, however,
I want to address an objection to the argument that is widely endorsed
but seems to me to be off the mark.

38

3. NEED THE SKEPTIC CHALLENGE ALL OF OUR

KNOWLEDGE AT ONCE?

A common objection to Descartes' argument is that it challenges all of

our knowledge at once. Sometimes the charge is put somewhat differently: What is wrong with the argument is that it assumes (incorrectly)
that we can challenge all of our knowledge at once. It incorrectly assumes
that such general doubts make sense, or are intelligible. Sophisticated
versions of this objection occur in Wittgenstein, J. L. Austin, and Stanley
Cavell, and more recently in Barry Stroud and Michael Williams. 9 Now
to be fair, all of these philosophers say much more in their critiques of
skepticism than this, and none of them can be said to be dismissive in
the sense that they do not engage skeptical arguments seriously. But
then the frequency of the charge and the quality of its sources is all the
more puzzling. For although Descartes in fact presents himself as examining all of his knowledge at once, it seems clear that there is nothing in
his skeptical argument that requires him to do so. In other words, there
is nothing in his reasoning that requires him either to (a) challenge all of
his knowledge at once or (b) assume that such general doubts are intelligible. Even if Descartes wants to end that way, there is no reason why
he has to begin that way. Therefore his reasoning cannot be criticized
along those lines.
It will be interesting to look more closely at how the charge arises. I
want to argue that the objection becomes plausible only on misguided
conceptions of our engagement with skeptical arguments. For example,
one way to think of that engagement is in the context of a debate. In that
context we can imagine the skeptic making a particular rhetorical move
in response to criticisms. Specifically, the skeptic argues that we cannot
pursue any line of objection which assumes that we have knowledge of
the world, since that is the very thing in question. The skeptical argument
means to challenge all of our knowledge at once, or, more exactly, all of
our knowledge of the world at once, and so any line of objection that
assumes such knowledge begs the question against the skeptical position.
It begs the question, for example, to invoke empirical science as showing
that our cognitive faculties are reliable in some important respect.

See Ludwig Wittgenstein, On Certainty (Oxford: Blackwell, 1969); J. L. Austin, Sense and
Sensibilia (Oxford: Oxford University Press, 1962); Stanley Cavell, The Claims of Reason
(Oxford: Oxford University Press, 1979); Stroud, The Significance of Philosophical Scepticism;
and Williams, Unnatural Doubts.

39

Now it seems clear that this charge of question begging is relevant if

we think of skeptical arguments as moments in a debate. If the debate is
over whether empirical knowledge is possible, we cannot invoke empirical science to show that it is. The question-begging charge is not
relevant, however, if we think of skeptical arguments as objects of
analysis. I have been suggesting that we should use skeptical arguments
as heuristic devices that can drive us to better theories of knowledge and
evidence. Skeptical arguments are philosophically interesting because
they start from premises that seem plausible to us, and yet lead to
conclusions that we find implausible. The philosophical task becomes to
find something that we can reject, and to replace it with something
better. In this context it is irrelevant that an imaginary skeptic could
invoke ground rules in an imaginary debate. So far as we find a line of
critique to be plausible, it does not matter that it begs a question against
an imaginary someone else.
Let us make a distinction between (a) assumptions that play the role
of premises in an argument and (b) ground rules that play the role of
rhetorical restrictions on a conversation. If we are thinking of skeptical
arguments as objects of analysis rather than as moments in a debate, then
rhetorical ground rules cannot be used to block a critique of an argument's premises. And for exactly the same reasons, a critique of rhetorical ground rules cannot replace a critique of an argument's premises. We
need to look at the premises of a skeptical argument, and at the logic
that takes us from those premises to its conclusion, and to find something
that we can reject there.
We can now return to Descartes' argument with these issues in mind.
Does anything in Descartes' skeptical reasoning require that he challenge
all of our knowledge at once? Does any premise in his reasoning involve
the assumption that we can do this? Put another way, does Descartes
assume anywhere in his argument's premises that a general or total
assessment of our knowledge is possible, or that completely general
doubts are possible? It is true that Descartes wants to challenge all of our
knowledge at once. But the reason he does is to be found in the wider
project he is undertaking rather than in anything essential to the skeptical
argument he puts forward. Remember that Descartes wants to clear
away all of his former opinions so that he can build back his knowledge
according to the right method for conducting his reason. Only after he
has cleared away all of his old beliefs can he proceed to establish a certain
foundation on which to build back his knowledge. But nothing in the
skeptical argument he raises assumes that he can challenge all of his

40

beliefs at once. Neither does his argument assume, in any essential

respect, that Descartes can raise completely general doubts. That is
where Descartes wants to get to, but it is not where he has to start.
We can see this by looking at Stroud's own reconstruction of Descartes' reasoning. On Stroud's account Descartes considers his belief that
he is sitting by the fire as a "best-possible case" from which he can infer
a more general conclusion.
The fire and the piece of paper are not too small or too far away to be seen
properly, they are right there before his eyes; it seems to be the best kind of
position someone could be in for getting reliable beliefs or knowledge by means
of the senses about what is going on around him. That is just how Descartes
regards it. Its being a best-possible case of that kind is precisely what he thinks
enables him to investigate or assess at one fell swoop all our sensory knowledge
of the world around us.10
On this conception of Descartes' reasoning, the skeptic does not begin
by questioning all of our knowledge at once; rather, he begins by
looking at our evidence for a best-possible case and concludes that our
evidence even in that case is not very good. The skeptic then infers from
this that in general our evidence for beliefs about the world is not very
good. Stroud implies that the inference is made in something like the
following way: I do not know that I am sitting by the fire because my
sensory experience does not adequately support that belief over alternative possibilities. But if I do not know that, then I do not know anything
about the world on the basis of my sensory experience this, after all,
was a best-possible case. Another way to make the inference is to notice
that there is nothing special about the best-possible case that would
block the same conclusion in other cases; the reasoning would go pretty
much the same way for any belief about the world grounded in sensory
appearances. But that includes all of my beliefs about the world, because
all of them are grounded, directly or indirectly, on the way things appear
to me through my senses.
Here is a reconstruction of the skeptical reasoning beginning from a
single case rather than from all of our knowledge at once, with both
ways of making the generalization included:
(D2)

1. A person can know something on the basis of her evidence only

if that evidence rules out every possibility that is inconsistent with
what the person claims to know.
10 Stroud, p. 9.

41

2. My evidence for my belief that I am sitting by the fire is my

sensory experience.
3. It is a possibility that I am in bed dreaming, and this is inconsistent
with my belief that I am sitting by the fire.
4. Therefore, I can know that I am sitting by the fire only if my
sensory experience rules out the possibility that I am dreaming.
(1,2,3)
5. But my sensory experience does not rule out that possibility.
6. Therefore, I do not know that I am sitting by the fire. (4,5)
7a. If I do not know that, then I don't know anything about the
world.

Or,
7b. The belief that I am sitting by the fire is arbitrary; if I plug in any
belief about the world, the argument goes through with respect
to that belief as well.
8. Therefore, I have no knowledge of the world. (6,7)

Two comments are in order. First, (D2) presents a formidable problem

even if we stop at subconclusion (6). In other words, the argument
reaches an unbelievable and unacceptable conclusion even before it makes
any general skeptical claim, because it is unbelievable that Descartes does
not know that he is sitting by the fire. Of course he knows this, and so
there must be some mistake in the argument that has nothing to do with
its rejection of knowledge of the world in general.
Second, suppose we do follow the argument to its general conclusion
in (8). Is there anything about the reasoning presented this way that
questions our knowledge of the world "all at once"? Of course the
argument's conclusion does, but that is the result of the skeptical reasoning, not a presupposition of it. We cannot say that the reasoning presupposes that our knowledge can be questioned all at once, or that what is
wrong with the argument is that it assumes that it can be. This kind of
critique of Descartes' argument, therefore, misses the mark. Even if it
presents a problem for the argument as it is presented in (Dl), Descartes'
reasoning is easily revised along the lines of (D2).11
So why does Stroud think that the skeptical argument must challenge
all of our knowledge at once, or must assume that we can raise com11 It is not obvious to me that challenging all of our knowledge at once, or assuming that we
can, does create a problem for the skeptical reasoning in (Dl). For example, what premise
or step in (Dl) is thereby made objectionable? But since these aspects of the reasoning are
inessential, there is no need to pursue the question.

42

pletely general doubts, even when Stroud's own reconstruction shows

that the argument need not proceed in either of these ways? As I said,
the claim becomes plausible if we think of ourselves as debating a
skeptic. In that case we can imagine the skeptic making certain rhetorical
moves in an effort to head off certain kinds of responses to his skeptical
reasoning. The skeptic insists that he is challenging all of our knowledge
at once, and that that is why we cannot invoke some aspect of our
knowledge to critique the skeptical argument. This rhetorical move
amounts to challenging all of our knowledge at once and, in effect,
involves the assumption that to do so is intelligible.
Another way that Stroud's claim becomes plausible is if we think of
ourselves as engaged in a special kind of philosophical question or
project. As Stroud describes it, the special project of traditional epistemology is to answer the skeptic's question about knowledge of the
world, but within the context of certain severe restrictions. The traditional epistemologist's task is to show that knowledge of the world is
possible, but without assuming any information about the world in
doing so. We cannot assume, for example, as Moore did, that we know
that here is a hand, and then use this knowledge to show that we have
knowledge in general. In the following passage, Stroud makes exactly
this point on behalf of the skeptic:
If we have the feeling that Moore nevertheless fails to answer the philosophical
question about our knowledge of external things, as we do, it is because we
understand that question as requiring a certain withdrawal or detachment from
the whole body of our knowledge of the world. We recognize that when I ask
in that detached philosophical way whether I know that there are external
things, I am not supposed to be allowed to appeal to other things I think I
know about external things in order to help me settle the question. All of my
knowledge of the external world is supposed to have been brought into question
at one fell swoop; no particular piece of it is to be available as unquestioned
knowledge to help me decide whether or not another particular candidate is
true.12
Stroud does not stop at ruling out Moore-type refutations of skepticism. He uses the same reasoning to block responses to skepticism by
naturalized epistemology and, by extension, responses by any epistemology that invokes empirical knowledge. In the next passage Stroud is
considering whether W. V. O. Quine's epistemology could answer the
12 Stroud, p. 118.

43

traditional question about knowledge of the world in general. He concludes that it could not, because Quine assumes the very knowledge that
the traditional epistemologist calls into question.
Another apparent difference is that Quine's question about our knowledge is to
be answered by making use of any scientific information we happen to possess
or can discover, whereas the traditional epistemologist's question was meant to
put all that alleged information into jeopardy and hence to render it unavailable
for such explanatory purposes. Any question empirical science can answer could
not be the traditional philosopher's question.13
It is not obvious that Stroud has accurately described the traditional
epistemologist's question. Stroud's description of this question does not
seem to fit the questions of Plato, Aristotle, or Aquinas, for example.
But putting that aside, the essential point for our purposes is that it is
rhetorical considerations that have once again imported the issue of total
assessment. In other words, even on Stroud's account there is nothing
in the skeptical argument that challenges all of our knowledge at once, or
that raises completely general doubts, or that assumes that it is possible
to do these things. By Stroud's own lights, it is only the purpose of the
skeptical argument, not the argument itself, that involves such assumptions. And this means that rejecting such assumptions does not touch
the argument. Even if we reject the traditional epistemologist's question
as Stroud describes it, as surely we should, the skeptical argument remains standing and in need of further analysis.
Michael Williams is another philosopher who recognizes that Descartes' argument can be reconstructed along the lines of (D2) but who
nevertheless insists that it assumes, at least implicitly, that all of our
knowledge of the world can be challenged at once. Williams endorses
Stroud's claim that the assumption can be traced back to traditional
epistemology's special project:
In trying to explain how what might otherwise seem to be truisms take on a
surprising significance, it is natural to look first to the traditional epistemologist's
aim of assessing the totality of our knowledge of the world. Because he wants to
explain how we are able to know anything at all about the external world, his
plan is to assess all such knowledge, all at once. But surely, the argument now
goes, if we are to understand how it is possible for us to know anything at all
13 Ibid., p. 221.

44

about external reality, we must trace that knowledge to knowledge we should

still have even if we knew nothing about the world.14
Williams claims that assumptions about total assessment also show up
in skeptical arguments. This is because epistemology's project and skeptical arguments share a deeper assumption: what Williams calls "epistemological realism." According to Williams, it is this concealed theoretical commitment that both (a) makes the project of total assessment
plausible and (b) allows the skeptic to draw general conclusions from
specific cases along the lines of (D2).
In Williams' sense, epistemological realism is not a metaphysical thesis
about the mind independence of reality. Rather, it is a constellation of
related epistemological theses about the nature of knowledge. First,
epistemological realism treats knowledge of the world like a natural kind.
The epistemological realist thinks of knowledge in very much the way the
scientific realist thinks of heat: beneath the surface diversity there is structural
unity. Not everything we call knowledge need be knowledge properly so called.
But there is a way of bringing together the genuine cases into a coherent
theoretical kind. By doing so - and only by so doing - we make such things as
"knowledge of the external world" the objects of a distinctive form of theoretical investigation. We make it possible to investigate knowledge, or knowledge
of the world, as such. (pp. 108-109)
Second, and seemingly following from this, the epistemic status of a
belief is fully objective and radically non-contextual. For this reason the
epistemological realist can evaluate abstract propositions as well as particular judgments: the contextual aspects of a particular judgment can be
ignored because they are irrelevant to the judgment's epistemic status,
which is determined by its propositional content alone. For the realist,
"a judgment derives its epistemological status from some highly abstract
feature of the content of the proposition it contains" (p. 67; see also
pp. 199200). More specifically, relations of epistemic priority are objective and non-contextual; for an epistemological realist, "there are relations of epistemological priority that hold between propositions
independently of the circumstances in which those propositions are
advanced, the interests that govern their assessment, or any other such
'contextual' features" (p. 67).
Williams argues that it is exactly these presuppositions of epistemo14 Williams, Unnatural Doubts, p. 89; cited hereafter by page number in the text.

45

logical realism that allow the skeptic to derive general conclusions from
specific examples of knowledge.
When we have a natural kind, we can learn general things about the kind
by investigating the properties of appropriate samples. The procedure is legitimate because of the supposed underlying hidden structure. Where there is such
a structure, what goes for one sample will go for all; where there isn't, no
general lessons can be drawn. . . .
We see, then, that the strategy of arguing from a representative case, far from
sidestepping the traditional epistemologist's realist presuppositions with respect
to epistemic kinds and relations, brings us right back to them. His procedure is
legitimate only if it is reasonable to treat terms like "empirical knowledge" as
natural kind terms. There must be something in the epistemic realm analogous
to the hidden structure or essential characteristics of naturally occurring substances such as gold. There has to be a microstructure, a hidden essence, of
empirical knowledge. We have already seen what this is: context invariant, fully
objective and autonomous epistemological constraints: in particular, natural relations of epistemic priority, (pp. 164165)

In these passages Williams has given up the idea that skeptical arguments must begin by questioning all of our knowledge at once, or that
they must involve the assumption that total assessment is possible. His
considered position is that skeptical arguments presuppose epistemological realism, and that it is this presupposition which allows the skeptic to
arrive at total assessment by examining specific cases of knowledge. In
later chapters, however, Williams returns to his earlier diagnosis: "When
we see that the sceptic's truth talk is really a way of asking his characteristically general questions about knowledge and justification, we see that
the totality condition on a philosophical understanding of knowledge,
and the controversial and implausible ideas it embodies, is the proper
target for the theoretical diagnostician"(pp. 246247). Presumably what
Williams should say here is that skepticism depends on epistemological
realism, and that epistemological realism makes sense of the skeptical
project of total assessment. But as Williams has already agreed, skeptical
arguments need not begin by challenging all of our knowledge at once
and need not start with the assumption that total assessment is possible.
Is Williams correct that skeptical arguments depend on the assumption of epistemological realism? If they do, and if epistemological realism
entails that total assessment is possible, then there is a sense in which
skeptical arguments do depend on the possibility of total assessment.
Namely, if total assessment is not possible, then, by modus tollens, something assumed in the skeptical argument is false namely, epistemolog-

46

ical realism. Perhaps this is all that Williams ever meant by claiming that
skeptical arguments assume that total assessment is possible; they do not
assume it as a premise, but they are committed to it in the sense that
they assume something else (epistemological realism) that entails it.
But putting all of that aside, Williams' claim that skeptical arguments
do depend on epistemological realism seems incorrect. We may see this
by distinguishing a weak and a strong version of epistemological realism.
In the strong sense, epistemological realism claims that the epistemic
status of a belief in entirely non-contextual. This seems to be the thesis
that Williams intends in his own characterizations of epistemological
realism. Let us define a weaker version of the thesis, however. Weak
epistemological realism claims that the epistemic status of a belief is not
entirely a function of context. In other words, weak epistemological
realism claims that some conditions relevant to epistemic status are noncontextual. And now the point is this: Williams claims that skeptical
arguments can draw general conclusions from specific cases only by
assuming strong epistemological realism. But in fact, all that is needed is
weak epistemological realism. So long as some conditions relevant to
epistemic status are not context dependent, and so long as the skeptical
argument exploits only those conditions, the argument need not assume
that no other conditions relevant to epistemic status are context dependent.
Put differently, to make the inference to a general conclusion the
skeptic need only deny that he has wrongly exploited any contextual
feature of knowledge; he must deny that the skeptical argument invokes
in the specific case some contextual feature that does not carry over to
cases of knowledge in general. But denying this does not commit the
skeptic to epistemological realism in the strong sense. Strong epistemological realism is the thesis that no conditions on knowledge are sensitive
to context, whereas the skeptic need only deny that all conditions on
knowledge are. In particular, the skeptical arguments from Hume and
Descartes assume that all beliefs about the world depend, directly or
indirectly, on the way things appear to us through the senses. But to
affirm that this aspect of empirical knowledge is context invariant does
not commit one to holding that all features of empirical knowledge are
context invariant.
Suppose that Williams grants this but continues to charge that skeptical arguments assume weak epistemological realism. Now Williams'
objection must be characterized as follows: "In order to draw general
skeptical conclusions from specific cases, the skeptical argument must

47

assume that some feature of epistemic status is context invariant. But no

features of epistemic status are invariant even weak epistemological
realism is false." The problem with this objection is that weak epistemological realism seems true. In other words, it seems that some conditions relevant to knowledge are context invariant. Specifically, the assumption made by Hume and Descartes that all knowledge of the world
depends on sensory appearances seems correct, even platitudinous.
Here Williams would disagree. According to him, not all knowledge
of the world depends on the evidence of sensory appearances, since
often enough knowledge of the world depends on other knowledge of
the world, and sometimes on no evidence at all. In effect, Williams
would reject the move from (6) to (8) in (D2) and would reject premise
(1) of (H2). Even if some well-chosen belief depends for its evidence on
how things appear through the senses, it does not follow that all knowledge of the world depends on that kind of evidence.
As I have said, my own assessment is that it is true that all beliefs
about the world depend for their evidence on the way things appear,
and that this claim is a truism rather than something we must derive
from epistemic realism. A number of points can be made to bring this
out. First, we should not build into the claim any particular theory of
sensory appearances, or any particular theory about how sensory appearances act as evidence for knowledge about the world. The principle of
charity requires that we do not load philosophical theories into the claim,
especially not implausible or false ones. Second, the claim is not that all
beliefs about the world depend directly on the evidence of sensory appearances. It might be that our most immediate reasons for believing
one thing about the world is that we believe other things about the
world. But if these other things eventually rest on the evidence of
sensory appearances (on how things look, feel, sound, etc.), then this
dependence transfers to the belief about the world in question. It is
consistent with Williams' point that some knowledge of the world is
directly supported by other knowledge of the world, to also claim that
all knowledge of the world is indirectly supported by evidence involving
the way things appear. Third, the claim need not be interpreted as stating
that all of our evidence for all of our beliefs about the world must
ultimately rest on evidence involving sensory appearances. It is enough
to make the case that some of our evidence for each belief does. The
skeptical argument need not assume, therefore, a foundationalism in
which all empirical claims are entirely grounded in a foundation of
sensory appearances.

48

But now isn't it plausible that all of our beliefs about the world
depend on evidence involving how things appear, at least partly and
indirectly? Can we think of any beliefs that do not? Williams' preferred
example is Moore's belief that here is a hand. In some contexts, he
admits, one might need evidence for such a belief. In others, however,
one knows it on no evidence at all. But this seems implausible to me. In
some cases where I know that here is a hand, I know it because I can
see it or feel it. In other cases I know it because I can reason from other
things, including other things that I can see or feel. For example, I can
reason that I would have a hand here unless something drastic had
happened, and if something drastic had happened I would have seen it
or felt it, but I did not. In the first case my knowledge is directly based
on the way things look or feel, and in the second case it is indirectly
based on the way things look or feel. So even in Williams' preferred
example, knowledge about the world is either directly or indirectly
grounded in how things appear.
We therefore have no good reason for denying what otherwise seems
obviously true: that all beliefs about the world depend for their evidence,
in one way or another, on how things look or feel (or taste, or smell, or
sound). Moreover, this is the only claim about epistemic priority that
the arguments from Hume and Descartes need. At least partly, and at
least indirectly, all beliefs about the world depend for their evidence on
sensory appearances. If we read that claim charitably and in a common
sense manner, then there is no reason for thinking that it is anything less
than a truism.
My own diagnosis of (H2) and (D2) concedes that this claim is a
truism. However, the skeptical arguments from Descartes and Hume
make a different mistake. Those arguments also assume that beliefs about
the world must be inferredfromthe way things appear. In other words,
they assume a specific account of how sensory appearances act as evidence
for beliefs about the world. Put differently, the arguments assume that
all evidential relations are inferential relations. This is in fact a natural
and widely shared assumption. Williams himself seems to accept it, and
this is why he must look for a mistake in the skeptical argument elsewhere.
If our knowledge of the external world really does need to be derived, in some
general way, from prior experiential data, we ought to be able to explain how.
But it is difficult to see how we could give an account of a warrant-conferring
form of inference connecting how things appear with claims about how they
are in the external world which would be adequate to meeting the sceptic's

49

challenge while doing justice to our sense of the objectivity of the world. The
doctrine of the epistemic priority of experiential knowledge over knowledge of
the world seems to disconnect our beliefs about the external world from the
only evidence available to support them. (p. 56)
In this passage Williams concedes that an adequate inference from appearance to reality is impossible, and for this reason goes on to deny that
our evidence for beliefs about the world is how things appear. He is
assuming that if sensory appearances are to function as evidence, this
must be via good inferences from them. Or consider Williams' endorsement of Ayers' classic analysis of the pattern of skeptical arguments: First
it is argued that some kind of knowledge-claim is in need of inferential
support from some kind of evidence, and then it is argued that there is
no deductive inference and no inductive inference forthcoming. Again,
Williams' diagnosis accepts the assumption that good evidence must
ground deductive or inductive inferences. What he rejects is the assumption that there are epistemic kinds, carrying with them relationships of
epistemic priority.
In later chapters I argue that it is the assumption that evidence is
always inferential that is mistaken. To be sure, some good evidence is
inferential, to be understood in terms of deductive and inductive relations among propositions. But not all evidential relations are inferential
relations. Moreover, I argue that an adequate theory of knowledge and
evidence can explain why this is so and, specifically, how knowledge
grounded directly in sensory appearances can be non-inferential.
It should be noted, however, that Williams' diagnosis is fully consistent with two of my major theses in this book namely, that traditional
skeptical arguments cannot be easily dismissed as making some obvious
mistake, and that the close analysis of skeptical arguments reveals substantive but mistaken epistemological assumptions, the rejection of
which drives progress in epistemology. Williams himself has defended
these theses as persuasively as anyone. In fact, my agreement with Williams goes even deeper than this. Both of us endorse accounts of knowledge which, in broad outline, are contextualist, externalist, and foundationalist, and which require both objective reliability and subjective
justification for knowledge. At this level of generality the difference
between us is this: Williams identifies epistemological realism as the
mistaken assumption of skeptical arguments and thus replaces it with
contextualism. I deny that traditional skeptical arguments presuppose
epistemological realism, at least in the strong sense that Williams seems
to intend. In addition, I argue that agent reliabilism provides a theoretical

50

explanation for the various diagnoses of skeptical arguments that I do

think are correct. Put a different way, agent reliabilism provides a deeper
theoretical explanation for the contextualist, externalist foundationalism
that Williams and I share.15
4. TWO MORE WAYS NOT TO UNDERSTAND DESCARTES'
ARGUMENT

In the preceding section I have been arguing against a particular interpretation of Descartes' skeptical argument namely, I have been arguing
against the claim that Descartes' reasoning must (a) challenge all of our
knowledge at once or (b) assume that completely general doubts are
intelligible. I now look at two more ways not to understand Descartes'
argument. Remember that we are trying to reconstruct the argument in
its most powerful form. In other words, we are trying to reconstruct it
employing only assumptions that seem natural and plausible to us. This
is in the context of a methodology which uses skeptical arguments as
heuristic devices for driving positive epistemology. Reconstructed in
their most powerful form, skeptical arguments highlight mistaken assumptions about the nature of knowledge and evidence, the rejection of
which must drive us to substantive epistemological positions. The more
powerful the argument in terms of its plausibility, the more substantive
the position that is required to reject it.
I said earlier that various reconstructions of Descartes' argument are
generated by different ways of understanding the idea of evidence "ruling out" alternative possibilities. One way that idea is commonly understood is in terms of entailment. On this reading, premise (1) of (D2) says
that one's evidence can give rise to knowledge only if it entails that all
alternatives to one's belief are false. That amounts to saying that evidence
gives rise to knowledge only if it entails the belief that is based on it.
This interpretation makes premise (1) of (D2) implausible, however.
To understand the skeptical argument that way robs it of any force that
it might have. It is perhaps a fair reading of Descartes, in that he
sometimes seems to accept only demonstration as an acceptable way of
building one's knowledge from indubitable first truths. But if this is
what he intended, then, as before, it has more to do with his wider
project than it does with the skeptical reasoning he puts forward in the
first step of that project. Therefore, even if this is a fair reading of the
15 See Williams, Unnatural Doubts, and "Skepticism," in Greco and Sosa (1999).

51

argument as it appears in Descartes' text, there is no reason that we have

to understand the skeptical argument that way.
Perhaps the most common way to understand the idea of ruling out
alternative possibilities is epistemically, so that to rule out a possibility is,
for example, to know that it is false, or to have good reason for believing
that it is false, or to be certain that it is false. But this way of understanding the phrase also robs (D2) of its force, for on this understanding it
seems wrong that I cannot rule out alternative possibilities. Don't I
know, for example, that I am not dreaming? Most of us have the strong
intuition that we do know that we are not dreaming in the typical case,
and so no skeptical argument will have force if it begins from the
assumption that we do not know this. What we are looking for, we
should remember, is a version of the argument that employs assumptions
that seem plausible to us. But on epistemic interpretations of "ruling
out," premise (5) of (D2) will seem false to anyone who is not already a
skeptic, and so will seem false to us.
Someone wishing to defend the skeptic might concede that these are
our intuitions regarding dreams in the normal sense that is, regarding
the kind of dreams we experience when we are asleep in bed, and yet
challenge the claim that we have similar intuitions regarding philosophical dreams. Do I know that my whole life is not a dream, as it would
be if I were the victim of an evil deceiver? Or consider the possibility
that I am the victim of a psychology experiment in the year 3000. How
do I know that I am in fact sitting at a table writing on a computer?
Perhaps I am a disembodied brain, hooked up to a computer that is
thousands of times more powerful and programmed to stimulate my
severed nerve endings so as to create exactly the experiences I am having
now. If we think of the dream possibility in either of these more radical
ways, the argument goes, then it no longer seems obvious to us that we
know that we are not dreaming.
But doesn't it? Don't I know that both of these radical dream scenarios are false, because I know that I am sitting at my desk writing on my
computer? I must confess, I think it is obvious that I do know these
things. But even if some would disagree, it surely is not obvious that I
do not know that these dream scenarios are false. And so interpreted this
way, the skeptical argument still depends on a premise that does not
seem obvious or compelling.16
16 A number of philosophers have thought that premise (5) is intuitively plausible on the

52

Of course, premise (5) does not have to be intuitively plausible if the

skeptic can give independent grounds for it. Stroud puts forward an
argument that is supposed to do just this, so we should take a look at
that argument. According to Stroud, Descartes thinks that he can know
that he is sitting by the fire only if he knows that he is not dreaming.
But Descartes cannot know that he is not dreaming, for the following
reason. If it really is a condition of knowing anything about the world
that one know that one is not dreaming, then one can have knowledge
of the world only if some state of affairs or some test establishes that one
is not dreaming. But whether such a test exists or is employed successfully is itself a fact about the way things are in the world, and so one
must already have knowledge of the world for the test to do its work.
Let us suppose that there is in fact some test which a person can perform
successfully only if he is not dreaming, or some set of circumstances or state of
affairs which obtains only if that person is not dreaming. Of course for that test
or state of affairs to be of any use to him Descartes would have to know about
it. He would have to know that there is such a test or that there is such a state
of affairs that shows that he is not dreaming. . . . Now strictly speaking if it is a
condition of knowing anything about the world beyond one's sensory experience
that one know that one is not dreaming, there is an obvious obstacle to
Descartes's ever having got the information he needs about that test or state of
affairs. He would have to have known he is not dreaming to get the information
he needs to tell at any time that he is not dreaming - and that cannot be done.17
This argument is supposed to provide independent grounds for premise
(5) of (D2), where that premise is understood to be the claim that I
cannot know that I am not dreaming. But if we look closely, we can
see that the argument is not a very good one. The argument can be
reconstructed as follows:
1. I can know that I am not dreaming only if I can know something about
the world for example, that some test guarantees that I am not dreaming.
present reading, but it seems to me that their intuitions have been tutored here. Most nonphilosophers, I would think, find it implausible that they do not know they are brains in
vats or victims of evil demons. For the contrary view see Keith DeRose, "Solving the
Skeptical Problem," Philosophical Review 104, (1995):l-52. See also Stewart Cohen, "How
To Be a Fallibilist," Philosophical Perspectives 2 (1988): 91-123; and Robert Nozick, Philosophical Explanations (Cambridge, MA: Harvard University Press, 1981).
17 Stroud, p. 21.

53

2. But I can know something about the world only if I can know I am not
dreaming.
3. Therefore, I cannot know that I am not dreaming.
Here premises (1) and (2) look plausible, but (3) does not follow from
them. All that follows from (1) and (2) is that I can know that I am not
dreaming only if I can know that I am not dreaming. That is true
enough, but of little skeptical force.
Perhaps Stroud means his premises to be temporalized. In that case
we get the following reconstruction.
1. I can know that I am not dreaming only if I already know at some earlier
time something about the world.
2. I can know something about the world only if I already know at some
earlier time that I am not dreaming.
3. Therefore, I can know that I am not dreaming only if I already know at
some earlier time that I am not dreaming. (1,2)
4. Therefore, I cannot know that I am not dreaming. (4)
Here (3) follows from premises (1) and (2), and (4) follows from (3). But
the problem with this version of the argument is that premise (2) no
longer seems plausible. Most of us think that we know things about the
world at the same time that we know we are not dreaming. No one
thinks that we must first find out that we are not dreaming and then
infer beliefs about the world, or vice versa, for that matter. So on either
understanding of Stroud's argument for the claim that we cannot know
we are not dreaming, we do not have good independent grounds for
that claim. The lesson we should draw from this is that we should not
understand the concept of "ruling out" in (D2) in an epistemic sense.

5. DISCRIMINATING EVIDENCE

There are at least two other interpretations of "ruling out" available,

and both generate powerful skeptical arguments. The first is that we
think of evidence ruling out alternatives in terms of discriminating those
alternatives from the state of affairs that is believed to hold. 18 For example, often my evidence for the belief that my wife has come home from
work is that I hear her voice announcing her arrival. This auditory
18 For a similar understanding of the concept, see Alvin Goldman, "Discrimination and Perceptual Knowledge," Journal of Philosophy 73 (1976): 771791, reprinted in Goldman, Liaisons: Philosophy Meets the Cognitive and Social Sciences.

54

experience discriminates her being home from my neighbor's yelling for

his children and from my cat's whining for her dinner. It also discriminates that state of affairs from the television set's being on, but not from
my wife having set up a tape recorder so as to fool me into thinking she
has come home. If I were to go into the other room and add my visual
experience to the evidence that I already have, then my evidence would
discriminate my belief from this last possibility as well.
The current idea is that this is how we should understand the skeptical argument in (D2). A set of evidence gives me knowledge only if it
discriminates the state of affairs I believe to hold from others that are
inconsistent with it. If I believe that my wife is home but on the basis
of evidence that does not discriminate her being home from her not
being home, then I cannot know that she is home on the basis of that
evidence.
More exactly, the current proposal is this:
(Df.l)

A set of evidence E rules out alternatives to a belief p for a person

S if and only if E discriminates p's being true from possibilities
that are inconsistent with p's being true.

And,
(Df.2) E discriminates p's being true from possibilities qx . . . qn for S if
and only if E would cause S to believe that p is true when p is
true, and E would not cause S to believe that p is true when any
of q1 . . . qn is true.
We then have the following interpretation of key premises in argument
(D2):
(D3)

1. S can know that p is true on the basis of evidence E only if E

discriminates p's being true from possibilities that are inconsistent
with p's being true.
2. My evidence for my belief that I am sitting by the fire is my
sensory experience.
3. It is a possibility that I am not sitting by the fire but only dreaming
that I am.
4. Therefore, I can know that I am sitting by the fire only if my
sensory experience discriminates my sitting by the fire from my
only dreaming that I am. (1,2,3)
5. But my sensory experience does not discriminate these possibilities
for me.
6. Therefore, I do not know that I am sitting by the fire. (4,5)

55

At this point the argument is starting to look good. What premise,

for example, is it plausible to deny? One might deny premise (1), and
I think that ultimately this is the right move to make. But no one
should think that (1) jumps out as obviously false, or that denying (1)
should be considered uncontroversial. Consider that (1) is equivalent to
the following claim: One's evidence can give rise to knowledge only if
it discriminates situations where one's belief is true from situations
where it is not. That claim is not implausible, and many would consider it to be obviously true. Especially when considered outside the
context of the skeptical argument, the proposition might pass as a
platitude.
But in fact premise (1) of (D3) is false, or so I argue. This diagnosis
of the argument presents itself when we consider that (1) treats all
alternative possibilities the same way. In other words, (1) requires that
our evidence discriminate the truth of our belief from every alternative
possibility whatsoever. But it is questionable whether our ordinary concept of knowledge in fact requires that our evidence do this. Knowing
that my wife is home from work requires that I have evidence that
discriminates this situation from the one in which only my cat is home
and is whining loudly for her dinner. But it does not require, or so I
would think, that I can discriminate my wife's being home from her
having been abducted by aliens and replaced with a convincing lookalike. It had better not require that, for by hypothesis the imposter is
convincing, and so my evidence could not discriminate this possibility
from the one in which my wife is home. This possibility, like the evil
genius and brain-in-a-vat possibilities, is designed so that my evidence
cannot discriminate between it and the usual things that I believe on the
basis of that evidence.
These last considerations suggest that not all alternative possibilities to
a knowledge-claim carry the same epistemic weight. To know that
someone has committed murder we need to rule out suicide. And to
know that my wife is home I must rule out the possibility that it is only
my cat or a neighbor that is making some noise. I do not need to rule
out other things that are nevertheless clearly possibilities in some broad
sense. Some possibilities seem to be relevant, in the sense that they do
need to be ruled out in order to have knowledge, whereas other possibilities seem not to be relevant in that sense.
This suggests that we revise premise (1) of (D3) as follows, requiring
an analogous revision in premise (3) to keep the argument valid.

56

(D4)

1. S can know that p is true on the basis of evidence E only if E

discriminates p's being true from all relevant possibilities that are
inconsistent with p's being true.
2. My evidence for my belief that I am sitting by the fire is my
sensory experience.
3. It is a relevant possibility that I am not sitting by the fire but only
dreaming that I am.
4. Therefore, I can know that I am sitting by the fire only if my
sensory experience discriminates my sitting by the fire from my
only dreaming that I am. (1,2,3)
5. But my sensory experience does not discriminate these possibilities
for me.
6. Therefore, I do not know that I am sitting by the fire. (4,5)

On this interpretation, premise (1) now seems true. In fact, if what

we mean by "relevant possibility" is a possibility that one's evidence
must rule out in order to generate knowledge, then (1) becomes a
tautology. But now a different diagnosis of the argument suggests itself.
First, we can make a distinction between normal dreams that occur in
sleep and "philosophical" dreams such as those caused by evil demons
and supercomputers in futuristic psychology experiments. We can then
say that the possibility that I am fooled by a normal dream is relevant in
some cases, but my sensory experience can usually rule it out. In other
words, my sensory experience can usually discriminate waking life from
a normal dream. Philosophical dreams cannot be so discriminated, but
they are not relevant possibilities. So if by dreams we mean normal
dreams, then premise (5) of (D4) is false. But if by dreams we mean
philosophical dreams, then premise (3) of (D4) is false. Either way, the
skeptical argument is unsound.
This approach to (D4) is promising, but we should admit that it is
only the beginning of an answer. It is a beginning, because it is already
a substantive position to hold that one's evidence need not discriminate
the truth of a knowledge-claim from all possibilities that are inconsistent
with it. But it is only a beginning, because we would like an account of
what makes some possibilities irrelevant in this sense and others not.
Furthermore, we would like an account of knowledge and evidence that
explains why all of this is so.
In Chapter 8, I offer an account of "relevant possibility" that captures
our pre-theoretical intuitions about which possibilities are relevant ones
and which are not, and I argue that a virtue theory of knowledge and

57

evidence explains why the account has the content that it does. In other
words, I argue that a virtue theory explains why our evidence must
discriminate the truth of knowledge-claims from some possibilities and
not others.
6. SUPPORTING EVIDENCE
Another way we can interpret the idea of evidence ruling out alternatives is in terms of deductive and inductive support. Specifically,
A body of evidence E rules out a possibility q if and only if E supports
(deductively or inductively) not-q.
On this interpretation it is not clear that premise (1) of (D2) needs to be
revised so as to be restricted to relevant possibilities. Or, put differently,
it is not clear that a distinction between relevant and irrelevant possibilities needs to be made. For on the present interpretation premise (1) of
(D2) says that one's evidence ought to support the negation of all
possibilities that are inconsistent with what is believed on the basis of
that evidence. But that is equivalent to saying that one's evidence ought
to support the beliefs that are based on that evidence.
More formally, we have the following argument.
(D5)

1. S can know that p is true on the basis of her evidence E only if,
for all q such that q is inconsistent with p, E supports not-q.

Or equivalently,
S can know that p is true on the basis of E only if E supports p.
2. My evidence for my belief that I am sitting by the fire is my
sensory experience.
3. It is a possibility that I am not sitting by the fire but only dreaming
that I am.
4. Therefore, I can know that I am sitting by the fire only if my
sensory experience supports the proposition, deductively or inductively, that I am not merely dreaming this. (1,2,3)
5. But my sensory experience does not support the proposition that I
am not dreaming.
6. Therefore, I do not know that I am sitting by the fire. (4,5)
Again, premise (1) of (D5) looks plausible and might even be considered
obvious. Moreover, premises (3) and (5) also look good on this interpretation. Since we are no longer required to restrict (1) to relevant possibilities, there is no need to restrict (3) or (5) that way either. Accord-

58

ingly, premise (3) of (D5) need only claim that my dreaming is possible
in some broad sense of possibility, and that claim is hard to deny.
Premise (5) might seem to be the weak point of the argument, but
here considerations from Hume can be brought forward to give the
premise a strong defense. That my sensory experience does not deductively support the proposition that I am not dreaming is obvious, since
it is possible that I have exactly that experience even when I am dreaming. This should be conceded even in the case of normal dreams, but
certainly it is true in the case of philosophical dreams. By hypothesis,
such dreams replicate the way things appear in waking life.
But more importantly, my sensory experience alone does not inductively support the proposition that I am not dreaming. The point here is
the same that we saw in the discussion of (H2) and (H3): my sensory
appearances do not even make likely the proposition that I am not
dreaming, unless I am also assuming, at least implicitly, that sensory
appearances are a reliable guide to reality. Without this assumption, or
something very much like it, sensory appearances do not make likely
anything about the way the world is. But that includes the proposition
that I am not dreaming, since that proposition is itself about the way the
world is.
And, of course, merely assuming that my experience is a reliable
guide to reality is not enough. If my evidence for the proposition that I
am not dreaming is to be any good to me, I must know, or at least be
justified in believing, that the assumption in question is true. But what
is my evidence for the proposition that my experience is a reliable guide
to reality? As Hume has pointed out, this is itself a matter of fact and
therefore could be established only by empirical evidence. But since
such evidence must already include the assumption in question, any
attempt at justification here would be circular.
I believe that this is the most powerful version of the argument from
Descartes' Meditation I. It does not seem susceptible to the relevant
possibilities approach that makes a distinction between normal dreams
and philosophical dreams, since premise (1) of (D5) is plausible just as it
is stated. Who would deny, outside the context of threatening skeptical
conclusions, that one's evidence must support the conclusions that are
based on it?
But neither do any of the other premises of the argument seem
implausible. Consider premise (2) of (D5). When I am sitting by the
fire, the reason I believe this is that this is how the world is presented to
me in my sensory experience. And isn't this just obvious? Some philos-

59

ophers have denied it, but usually what they are objecting to is some
particular account of sensory experience, or some account of how sensory
appearance acts as a reason for my belief. Reading the premise in a
common sense way, however, who could deny that, in the typical case,
the reason that I believe I am sitting in some location is that this is how
things appear to me, either visually or by some other sensory modality?
Perhaps someone will deny it if they are in the grip of a philosophical
theory, or if they think they are pushed to deny it in order to stave off
unacceptable skeptical consequences. But no one would deny it pretheoretically, and we probably should not deny it even posttheoretically.
Next consider premise (3), which claims only that it is a possibility
that I am dreaming. Premise (3) need not claim that I have some good
reason to think that I am dreaming, or even that I should take the
possibility seriously as something that might turn out to be true. Rather,
the premise claims only that my dreaming is a possibility in a much
broader sense that is, it is not incoherent, or inconceivable, or logically
impossible. And again, this seems hard to deny.
The remaining independent premise is (5), and we have already seen
an argument from Hume to support the claim that is being made there.
Accordingly, this version of Descartes' argument ultimately depends on
the argument from Hume. Both arguments essentially make the claim
that beliefs about the world are inadequately supported by their evidence. Descartes' version makes the claim in terms of ruling out alternative possibilities, whereas Hume makes the point more straightforwardly by considering what evidence we have for beliefs about the
world. But both arguments ultimately make the point that there is no
good inference from sensory appearances to beliefs about the world.
Put another way, both arguments claim that there is no good inference from the way things appear to the way things are. Furthermore,
both claim that this is so because even an inductive inference would
require the assumption that sensory experience is a reliable guide to the
way things are, and there is no non-circular justification for that assumption. But since knowledge requires good inferences from one's evidence,
there is no knowledge of the way things are.

60

Skepticism about the World: Part

Two Dismissive Responses

We are finally in a position to consider dismissive responses to skepticism

about the world. In Chapter 1 I defined "dismissive responses" as responses that either (a) do not engage skeptical arguments at all, or (b)
engage them only superficially. Rather, such responses dismiss skeptical
conclusions without paying serious attention to the reasoning that leads
up to them. Because such responses fail to engage skeptical arguments
seriously, they fail to locate the mistakes that the arguments actually
make. This insures, in turn, that they miss the lessons that skeptical
arguments can teach.
By definition, dismissive responses fail to engage skeptical arguments
seriously. Type-a dismissive responses do not engage skeptical arguments
at all but instead focus on some aspect, or alleged aspect, of the skeptical
conclusion. Type-a responses fall into three categories: charges of selfrefutation, pragmatic responses, and rhetorical responses. After considering these I turn to type-b dismissive responses, or responses that engage
skeptical arguments only superficially. Here the focus is on two versions:
(i) the charge that skeptical arguments assume that knowledge requires
absolute certainty and (ii) the charge that skeptical arguments assume
that knowledge requires deductive evidence. Finally, I consider transcendental arguments as dismissive responses to skepticism. Transcendental arguments can be type-a or type-b responses, depending on how
they are understood.
1. CHARGES OF SELF-REFUTATION
Some philosophers have claimed that the skeptical conclusion is inconsistent, and that therefore skepticism is self-refuting. The idea is that in

61

claiming that no one knows, the skeptic is himself making a knowledgeclaim, and an inconsistent one at that. The skeptic falls into contradiction, claiming that he knows that no one knows.
A question naturally arises concerning the status of skepticism itself as a claim
about the human condition. Can it be known to be true? Is it itself a truth?
One can easily see that no coherent answer is possible. If one answers "yes>"
the original claim is denied. If one answers "no," the original claim is again
denied. In effect, the only consistent thing an absolute skeptic can do is to keep
silent altogether. As soon as he opens his mouth in speech, he has refuted
himself. . . .
The problem, of course, lies in the fact that absolute skepticism refers to
itself. As an assertion of a position, it affirms its own truth and hence denies the
absolute universality of the skeptical position.1
A second version of this objection claims that the skeptic's conclusion is
inconsistent with his premises. The idea is that the skeptic puts forward
knowledge-claims in his premises that contradict his denial of knowledge
in his conclusion. Accordingly, the skeptic must either fall into contradiction or must admit that he does not know his premises. If the former,
then skepticism is self-refuting. If the latter, then the premises have no
power to establish their conclusion.
This kind of response is dismissive because it does not engage any
particular skeptical argument; it raises objections that are completely
general, and that are supposed to implicate the skeptical conclusion
independently of any particular reasoning for it. On the other hand, the
objections cannot be sustained on even a superficial consideration of
actual skeptical arguments. The arguments from Hume and Descartes,
for example, are easily seen not to be self-refuting in the ways that the
objections charge.
We can see that skeptical reasoning does not have to be self-refuting
by considering two aspects of any skeptical argument.2 First, let us say
that the "scope" of a proposition is its subject matter, so that the scope
of the conclusion in argument (D5) in Chapter 2 is knowledge of the
world. The scope of Hume's conclusion in Section IV of the Enquiry is
1

Vincent G. Potter, On Understanding Understanding (New York: Fordham University Press,

1994), p. 17.
2 Nicholas Rescher makes the same distinctions in Scepticism: A Critical Reappraisal (Totowa,
NJ: Rowman and Littlefield, 1980), p. 4. Somewhat similar distinctions are treated by
George Pappas in "Some Forms of Epistemological Scepticism," in George Pappas and
Marshall Swain, eds., Essays on Knowledge and Justification (Ithaca: Cornell University Press,
1978), pp. 309-316.

62

knowledge of the future. But then a skeptical argument avoids selfrefutation of the second kind contradiction between premises and
conclusion so long as the scope of its premises falls outside of the scope
of its conclusion. For example, the skeptic putting forward arguments
(H2) or (D5) in Chapter 2 can consistently claim to know his premises
while at the same time claiming that no one knows anything about the
world, for he can claim that his premises are about our concepts of
knowledge and evidence rather than about objects. In other words, the
skeptic can affirm that we have conceptual knowledge, and so can know
that her premises are true, while denying that we have empirical knowledge, thus avoiding any inconsistency between premises and conclusion.
The first kind of self-refutation, involving a conclusion that contradicts
itself, can be avoided in essentially the same manner. So long as the
conclusion of a skeptical argument is not itself an empirical claim, the
skeptic avoids inconsistency when he concludes that he knows (nonempirically) that we have no empirical knowledge.
Second, let us say that the "degree" of a skeptical conclusion is the
quality of knowledge or justification that the argument denies we have.
For example, a skeptical argument might conclude that, within some
scope, we fail to have absolute certainty, or knowledge, or moral certainty, or probable belief, or reasonable belief. Again, the skeptic can
avoid self-refutation of the second kind simply by being careful about
degree, making sure that he claims a lesser degree of justification for his
premises than he denies that we have in his conclusion. For example, it
is not inconsistent to claim that reasonable premises entail that nothing
is certain. In the same way, the skeptic can avoid self-refutation of the
first kind as well. So long as a lesser degree of justification is claimed for
the conclusion than is denied in that conclusion, the conclusion does
not contradict itself. For example, it is not contradictory to say that it is
reasonable to believe that no one has knowledge. Therefore not even a
skepticism that is universal in scope must be self-refuting.3
We have seen that a skeptic can avoid self-refutation by being careful
about the degree ofjustification claimed for his premises and conclusion.
A special case of this is when the skeptic makes no claims at all about
the epistemic status of his premises. In this case, the skeptic merely
points out that we non-skeptics hold those premises to be true and so by
our own lights should admit that we have no knowledge. This skeptic
makes no claims about knowledge at all but merely calls attention to the
3

This is contra Potter, p. 17, and Rescher, pp. 14-15.

63

fact that our own assumptions entail that we do not know what we
claim to know. On the one hand, we find several seemingly innocent
assumptions to be plausible. On the other hand, we think that we know
many things about how the world is. But then it is our position that is
inconsistent, because those seemingly innocent assumptions entail that
we do not know what we think we know. That is not a problem for
the skeptic, since he does not claim to know anything. But it is a
problem for us, because we do claim to know things that our own
assumptions entail we do not.
By paying attention to either scope or degree, the skeptic need not
make contradictory claims. In the limiting case the skeptic makes no
claims about knowledge at all but instead calls our attention to an
inconsistency in our own assumptions about the nature and existence of
knowledge.

2. PRAGMATIC AND RHETORICAL RESPONSES

Pragmatic responses claim that skepticism can be refuted by calling

attention to the skeptic's actions. One version of this response claims
that the skeptic's actions outside of the study are inconsistent with her
skeptical conclusion. In acting as we all do to avoid an oncoming bus or
secure good tickets to a game, the skeptic demonstrates that her skeptical
conclusion is false. Another version of this response claims that the
skeptic's actions demonstrate that she does not believe her own conclusion. By acting like the rest of us, or even by offering the skeptical
argument to us for our consideration, the skeptic reveals that not even
she believes that there is no knowledge. She believes, for example, that
she knows that we exist to consider her arguments.
Rhetorical responses imply that skepticism can be refuted by making
rhetorical points about the skeptic's position in the debate about knowledge. Thus Locke writes, "[I]f all be a dream, then he doth but dream,
that he makes the question; and so it is not much matter, that a waking
man should answer him." 4 In a similar vein, Reid says of Descartes'
skeptic that "[a] man who disbelieves his own existence, is surely as
unfit to be reasoned with as a man that believes he is made of glass."5
4 John Locke, An Essay concerning Human Understanding, ed. Peter Nidditch (Oxford: Oxford
University Press, 1975), p. 634.
5 Reid, Philosophical Works, p. 100a. Reid has much more to say in response to skepticism,
and taken in its entirety his response is not at all dismissive. For a largely sympathetic treat-

64

This kind of response can overlap with pragmatic ones, as when Reid
says of Hume, "He believed, against his principles, that he should be
read, and that he should retain his personal identity, till he reaped the
honour and reputation justly due to his metaphysical acumen."6
Pragmatic and rhetorical responses are dismissive because they fail to
engage the arguments for skeptical conclusions. For the same reason they
are completely ineffective. Notice that the premises of Descartes' and
Hume's arguments do not mention anything about how one ought to
act or how one ought to speak. Rather, they amount to claims about
what knowledge requires, together with claims that our own beliefs fail
to meet those requirements. No rhetorical or pragmatic point will touch
those premises, since the premises do not make rhetorical or pragmatic
claims.
This assessment is supported by the following consideration. The
skeptical arguments that we have been considering would not lose any
of their force if we found no one willing to defend their conclusions.
What makes them interesting philosophical problems is that we ourselves
find their premises plausible, and yet, by seemingly cogent reasoning,
they lead to a conclusion that we find unbelievable. If there never
existed a skeptic to be debated, the skeptical arguments would lose none
of their interest in this respect. But if the existence of skeptics is irrelevant to the force of skeptical arguments, then observations about either
the skeptic's actions or her rhetorical position are irrelevant as well.
Pragmatic and rhetorical responses can be relevant when you are
trying to win a debate against a contentious opponent. If, in that situation, you can show that not even she can believe the position she is
defending, then that is an effective blow to strike. So might such responses be relevant if we were trying to persuade someone out of her
skepticism. In that case it might help to show the poor soul that not
even she believes what she is saying, or that no one can consistently act
on such a belief. But the point of analyzing skeptical arguments is neither
to win a debate nor to persuade someone out of their skepticism. It
could not be, since there are no skeptics to be debated or persuaded. At
least there are none outside of the study or the classroom, as Hume
himself pointed out. In other words, there are no skeptics who need to

ment of Reid's critique of skepticism, see my "Reid's Critique of Berkeley and Hume:
What's the Big Idea?", Philosophy and Phenomenological Research 55 (1995): 279-296.

Reid, p. 102a.

65

be persuaded from a life of skepticism, or whom we need to debate to

achieve some other useful purpose.
This last point is noteworthy. Although some very few people are
skeptics in the sense that they endorse skepticism or say that they are
skeptics, no one is a skeptic in the sense of living out the position. Hume
famously insisted on this, and contemporary philosophers who incline
toward skepticism consistently endorse the point. Thus Michael Williams writes that recent years "have seen a remarkable revival of this
Humean attitude" and cites Barry Stroud, Thomas Nagel, and P. F.
Strawson as "New Sceptics" who, in one way or another, rehearse
Hume's point that skepticism cannot be maintained in everyday life.7
To say that skepticism cannot be lived is not merely to insist that no
one acts as a skeptic. It is not merely to say, for example, that no one
fails to avoid an oncoming bus. The more important point is that, in the
context of everyday life, no one even believes that skepticism is true.
Even people who profess to be skeptics in their books still make judgments about who ought to be believed, discriminate good from bad
evidence, and make other epistemic judgments in the course of their
everyday lives. Hume insisted on this as well, as do present-day representatives of the skeptical position.8 But if no one acts out the skeptical
position, and if no one even believes it outside of the classroom or study,
then no one needs to be saved from it. There is no practical or "existential" problem of skepticism. And therefore strategies that aim at practical
or existential solutions are out of place.
These last considerations raise the question of epistemology's project.
What is the question that epistemology is supposed to answer or the
problem that it is supposed to solve? Historically, and I think to this day,
the project of epistemology has been conceived in five principal ways:
(a) to beat the skeptic in a debate; (b) to persuade someone out of his or
her skepticism; (c) to prove to someone else that we have knowledge;
(d) to prove to ourselves that we have knowledge; and (e) to provide an
adequate account of what knowledge is. My claim is that all of these
conceptions are misguided except for the last, and that this is rather
obvious once the question of epistemology's project is raised explicitly.
We have already seen that the first two conceptions are misguided
because there are no skeptics to be debated or persuaded. That is, no

7 Williams, Unnatural Doubts, p. xiii.

See Stroud, The Significance of Philosophical Scepticism., esp. Ch. 2, and Fogelin, Pyrrhonian
Reflections on Knowledge and Justification, e.g., pp. 910.

66

one exists who lives his life as a skeptic and who therefore needs our
help to be brought out of his skepticism. And even if there were such a
person, Reid is correct that "physic and good regimen" would be a
more likely cure than "metaphysic and logic." Moreover, there is no
one whom we need to defeat in a debate over the issue. If we come
across a pretend skeptic who wants to play that game, as we sometimes
do in a bar or an undergraduate classroom, the better policy is to redirect
attention to some more worthwhile activity. But if one likes to play
strange games, consider that the pretend skeptic cannot lose, provided
that she has the least bit of wit about her. All she need do is refuse to
allow any premise against her, and her skepticism will be completely
unassailable. Real skeptics are not to be found, and if they are found
they are not to be engaged with philosophy. Pretend skeptics should be
avoided altogether.
The third and fourth conceptions of epistemology's project either
reduce to the first two or are too easy. If the purpose of proving that we
have knowledge is to persuade or to win a debate, then we are back in
the misguided projects we have already considered. On the other hand,
if the point is to prove to ourselves or to some other non-skeptic that we
have knowledge, then the project is pointless, because it is too easy. To
prove something is to show that it follows from premises that are already
known. But if we are non-skeptics, then it is easy to prove that we have
knowledge from such premises. Being non-skeptics, we already know
that we know lots of things, and these particular cases vacuously establish
that we have knowledge of the kind in question for example, knowledge of the world. Someone might complain that this kind of proof begs
the question. But against whom does it beg the question? Remember
that we are not debating with a skeptic in the case being considered.
So the project of epistemology is not to persuade lost souls, or to win
debates with contentious opponents, or to prove that we know what
we think we know. Rather, it is to construct accurate accounts of
knowledge and evidence. In other words, the project of epistemology is
to answer Plato's age-old question in the Theaetetus, "What is knowledge?" And the point of analyzing skeptical arguments is that they help
us to do this. Skeptical arguments highlight mistaken assumptions about
knowledge that might otherwise go unnoticed, and they drive us to
replace those assumptions with something more adequate. Moreover, to
arrive at adequate accounts of knowledge and evidence we must consider our best judgments about which particular cases count as knowledge and which count as good evidence. Here again the analysis of

67

skeptical arguments helps us, because those arguments bring home the
counterintuitive results of proposals that inadequately capture those
judgments. Of course, skeptical arguments cannot help us if they are
dismissed with irrelevant considerations about how imaginary people
speak and act. Skeptical arguments can serve as powerful heuristic devices for driving positive epistemology, but only in the context of a
project properly conceived.
I have been arguing that epistemology's project is to give an adequate
account of knowledge that is, to answer Plato's question, "What is
knowledge?" Some might think that this conception of epistemology is
an outdated one, insulated from recent appreciation in philosophy of the
social, political, and other contextual factors involved in knowledge and
knowledge relations. But this is to misunderstand the project of giving
an account of knowledge. Nothing about that project disallows that
knowledge has social or political dimensions, or that gender has epistemological significance, or that attention to particulars is necessary, or
that knowledge depends on context in some other way. In fact, all of
these claims constitute competing epistemologies. They claim that an
adequate account of knowledge reveals that it has the contextual dimensions in question.
To drive home the point that recent contextualisms are not antithetical to epistemology, consider that all of them are consistent with what
is perhaps the most common position in epistemology today. According
to generic reliabilism, knowledge is true belief that arises from reliable
cognitive processes, or cognitive processes that reliably produce true
rather than false beliefs. Different versions of reliabilism give different
accounts of the processes involved, but all agree that knowledge has
more to do with the abilities of cognizers to "hook up" with the world
than it does with the properties of individual beliefs for example,
properties such as indubitability, incorrigibility, or irrevisability. But note
that reliabilism in this broad sense is perfectly consistent with the claims
that knowledge is social, political, gendered, or otherwise contextual.
All of these questions are in fact wide open within the context of generic
reliabilism, since it is an empirical matter how cognitive processes are
made reliable or unreliable by social, political or other factors. Moreover,
once the question is raised this way, it would be amazing if at least some
of these factors were not importantly involved in the production of
knowledge.
For this reason, recent self-proclaimed opponents of epistemology are
better seen as doing epistemology than as rejecting it. They are offering

68

accounts that emphasize aspects of knowledge and evidence that have

been neglected historically, but that are perfectly consistent with more
recent epistemological theorizing. Accordingly, what these thinkers reject is not the general project of giving an account of knowledge, but
particular outdated accounts that do not recognize aspects of knowledge
that contextualists deem to be central. And when I say they are rejecting
accounts that are outdated, I mean outdated for the better part of a
century. Accordingly, these thinkers need to catch up to epistemology
rather than reject it.

3. MORE DISMISSIVE RESPONSES

We are now in a position to consider type-b dismissive responses

to skepticism. Such responses also traffic in outdated theories of knowledge, accusing skeptical arguments of presupposing some version of
seventeenth- or eighteenth-century rationalism. For example, some philosophers have diagnosed skepticism as requiring absolute certainty for
knowledge. Absolute certainty has been conceived variously as requiring infallibility or incorrigibility, for example. But the general theme is
that skeptical arguments set the standards for knowledge outrageously
high, and so rejecting the skeptical conclusion requires no more than
rejecting implausible skeptical standards.9
Another look at our reconstructions of Descartes and Hume shows
that this charge is without any force. The point of Hume's argument is
that our beliefs about the world are either wholly unsupported by their
evidence or else are supported only by reasoning that is logically circular.
Such defects in one's evidence do not undermine only absolute certainty. If Hume is right, then our beliefs about the world lack even the
lowest degree of positive epistemic status. Of course, that is not to say
that Hume is right. The point here is that Hume's argument does not
run on a requirement of absolute certainty.
It might be claimed, however, that Hume is requiring that our evidence for our beliefs be absolutely certain, and that this is why he
conceives of our evidence for objects in the world as incorrigible beliefs
about our sensory experience. But although Hume in fact conceived our
evidence for objects in this manner, there is nothing about his skeptical
reasoning that requires him to do so. The arguments from Hume as
9

This kind of diagnosis is quite common. Rescher is one example (see esp. p. 56); see also
John Dewey, The Quest for Certainty (New York: Minton, Balch, 1929).

69

reconstructed in (H2) and (H3) in Chapter 2 do not make the claim

that beliefs about sensory experience are incorrigible, or that they need
to be to give rise to knowledge. Rather, premise (1) of each argument
can be read as claiming something seemingly obvious: that our evidence
for objects is that things appear a particular way, together with the
assumption that the way things appear is a reliable indication of how
they are. We do not have to say that sensory appearances are an infallible
indication of the way things are, and we do not have to say that
appearances are incorrigible. The argument runs just as well on more
sane characterizations of sensory appearances and their relation to objects
in the world.
Similar things can be said about the charge that skeptical arguments
require deductive inferences for knowledge. A close look at Hume's
reasoning reveals that he thinks our evidence for objects in the world is
neither deductive nor inductive. That it is not deductive is widely
recognized. But Hume claims that our evidence is not even inductive,
if it does not include some assumption to the effect that appearances are
a reliable guide to reality, and he argues quite powerfully that no such
assumption can be justified in a non-circular fashion. Again, whether
Hume is correct in all of this is not the question; I take it for granted
that he is not correct, and that skepticism about the world is false. But
Hume's mistake is not that he requires that evidence be deductive, or
that knowledge of the world be deduced from the way things appear.
The charge of deductivism might seem to hold against Descartes'
reasoning in Meditation I, but only if we read the argument there
uncharitably. As we saw in Chapter 2, one way to understand the idea
of evidence ruling out alternative possibilities is in terms of deduction,
so that one's evidence rules out the possibility that one is dreaming only
if one's sensory experience deductively entails that one is not dreaming.
But there are at least two other ways to understand this idea that do not
make the argument assume an implausible deductivism about evidence.
First, it is a plausible claim about evidence that it must discriminate
alternative possibilities from what is believed, where discrimination is
understood not in terms of deduction but in terms of the causal properties of one's evidence. The way we defined the notion in Chapter 2,
evidence discriminates the truth of a belief p from alternatives if it would
cause you to believe that p is true when p is true, but would not cause
you to believe that p is true when the alternatives are. Consideration of
everyday cases brings out the plausibility of Descartes' principle under-

70

stood in this manner, but whether the principle is true or false, it clearly
does not involve the thesis that evidence must be deductive.
Second, we can read Descartes' argument as making essentially the
same point as Hume's. On this reading the claim that sensory experience
does not rule out alternative possibilities is understood as the claim that
such evidence does not support the negation of those possibilities either
deductively or inductively. It does not support their negation deductively, because it is possible to have exactly the same experience when
the possibilities are true. But it does not support them inductively either,
because such support would require the assumption that sensory experience is a reliable guide to reality, which assumption cannot itself be
justified in a non-circular fashion.
To see that deductivism is not the real issue here, consider this: even
if we did have some adequate justification for the assumption that experience is a reliable guide to reality, that still would not make our
evidence deductive. For that assumption claims only that sensory experience is a reliable indication of how things are, not an infallible indication. So even if the assumption could be used as part of our evidence
for beliefs about the world it would provide only inductive or probable
support. Accordingly, the point of the skeptical argument is not that
deductive support is required for knowledge. The real point is that
inductive support is required, and we do not have even that for our
beliefs about the world.
I said that type-b responses to skepticism are dismissive because they
do not consider skeptical arguments seriously; rather, they depend on
uncharitable readings of such arguments, failing to consider that more
powerful reconstructions are easily available. Accordingly, these responses reject the arguments from Descartes and Hume for the wrong
reasons, and so fail to recognize the more interesting lessons that those
arguments can teach.
4. TRANSCENDENTAL ARGUMENTS
Before closing this chapter I want to consider one more kind of response
to skepticism about the world. Transcendental arguments are a kind of
dismissive response, because they fail to consider skeptical arguments or
only consider them superficially. As we will see, transcendental arguments are type-a or type-b depending on how we interpret the way
they are being employed against the skeptical conclusion.

71

By "transcendental arguments" I mean arguments of a certain form.

Such arguments start by considering the conditions for the possibility of
some undisputed phenomenon and argue to the conclusion that some
disputed object exists or that some disputed state of affairs is actual. For
example, Kant argued that a condition for the possibility of experience
is that spatialtemporal objects exist, and Heidegger argued that a condition for the possibility of inquiry is that we already dwell in the world
at which inquiry is directed. More recently Hilary Putnam has argued
that a condition for inquiring about knowledge of the empirical world
is that such a world exists,10 and Donald Davidson has argued that a
condition of our having any beliefs at all is that most of them are true.11
These arguments vary in their details, but their general strategy is the
same and puts them in a common class. All of them start with some
phenomenon that even the skeptic acknowledges, or that he must acknowledge as a presupposition of his own inquiry. And from there each
argues that a necessary condition for the possibility of that phenomenon
is that some other thing the skeptic disputes is true. By this line of
argument the skeptic is refuted by making explicit the conditions of his
own skeptical inquiry.
Accordingly, transcendental arguments share the following structure:
(TAS)

1. Some undisputed phenomenon is possible only if some disputed

phenomenon is actual.
2. The undisputed phenomenon is actual.
3. Therefore, the disputed phenomenon is actual.

Whether transcendental arguments are sound depends nearly entirely on

their first premise. This is because their argument form is clearly valid,
and the examples I have given do trade on undisputed phenomena in
their second premises. So, for example, it is undisputed that we have
sensory experience, or that we are inquirers, or that we have beliefs.
On the other hand, anyone familiar with these arguments knows that
establishing their first premise is no walk in the park. The arguments
here are long and complex, and depend in several places on claims that
10

Putnam, Reason, Truth and History.

11 Donald Davidson, "A Coherence Theory of Truth and Knowledge," in Ernest LePore,
ed., Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson (Oxford: Black-

well, 1992), reprinted from D. Henrich, ed., Kant oder Hegel (Stuttgart: Klett-Cotta, 1983).
See also, Donald Davidson, "The Method of Truth in Metaphysics," in P. A. French, T. E.
Ueling, Jr., and H. K. Wettstein, eds., Midwest Studies in Philosophy 2: Studies in the Philos-

ophy of Language (Morris: University of Minnesota Press, 1977) and reprinted in Donald
Davidson, Inquiries into Truth and Interpretation (Oxford: Clarendon Press, 1985).

72

are, at the very least, controversial. However, none of my criticisms of

transcendental arguments depends on the controversial nature of their
major premises. On the contrary, I think such arguments are ineffective
responses to skepticism even if they are exactly right from start to finish.
Neither do any of my criticisms depend on charges of question begging.
Since I am not conceiving the present context as a debate with the
skeptic, concerns about begging questions are not really relevant. To
beg a question there has to be someone to beg it against, but, as we
have seen, there are no skeptics that we ought to be debating.
Perhaps the most natural way to interpret transcendental arguments is
to view them as type-a responses to skepticism. On this interpretation
they do not engage skeptical arguments at all but instead try to show
that the skeptical conclusion must be false and so can be rejected independently of how one gets there. But on this interpretation there are
several reasons why transcendental arguments are ineffective responses to
skepticism.
The first reason they are ineffective is that the conclusions they try to
establish are extremely general. For example, Kant concludes that a
condition for the possibility of experience is that spatialtemporal objects
exist. But he does not claim which objects must exist, or when, or
where. And so even if Kant's argument is correct, it leaves untouched
any skepticism about knowledge-claims that are more particular. The
same is true of Heidegger's argument as well as Putnam's and Davidson's. As a rule, the more general the conditions that a transcendental
argument establishes, the less effective it is against skepticism about
particular knowledge-claims. On the other hand, the less general the
conditions established by the argument, the less plausible the argument
becomes. For example, no one would even think to argue that a condition of the possibility of my current sensory experience is that I am
here and now in front of a computer, or even that there are any
computers at all.
The second reason that transcendental arguments fail to refute skepticism is that they do not touch even the most general skeptical conclusions. This is because skeptical arguments are about knowledge rather
than truth. For example, the arguments reconstructed from Descartes
and Hume conclude that we lack knowledge of the world, not that the
world does not exist. Therefore no argument that establishes that the
world exists, or that we dwell in it, or that most of our beliefs are true
touches the conclusion that skeptical arguments put forward.
Even a superficial analysis of skeptical arguments shows that this is

73

the case. In Chapter 1 we saw that all skeptical arguments can be boiled
down to two essential premises, one stating that knowledge requires that
some condition be fulfilled, and one stating that the condition in question is not fulfilled. Looking at the arguments from Descartes and Hume,
we see that they do not claim that the truth-condition on knowledge is
not fulfilled. Rather, they claim that we lack appropriate evidence for
our beliefs. According to these arguments, even if our beliefs are true,
they do not amount to knowledge.
Skeptical arguments do not claim that the world does not exist, or
that we do not dwell in the world, or that most of our beliefs are not
true. For this reason, arguments with the purpose of establishing such
things cannot be effective responses against them or at least such
arguments cannot be effective as direct responses. Is there some other
way in which transcendental arguments are supposed to refute the skeptical conclusion? One possibility presents itself, but it can quickly be
rejected.
It might be thought that transcendental arguments refute skepticism
indirectly, because they can be used to give us knowledge that would
otherwise be lacking. In other words, even if skeptical arguments are
right that we typically lack evidence for our beliefs, transcendental arguments can provide evidence for at least some very general claims about
the world. We can prove, for example, that such a world exists. Or we
can prove that most of our beliefs are true, even if we do not know
which. But this line of thought is highly problematic. If the idea is that
only these very general truths about the world can be known, and that
only a small group of philosophers have ever known them, then the
proposal is highly counterintuitive. On the other hand, if the idea is that
this is how Jane and Joe in the street get knowledge in typical cases,
then the claim is still highly problematic, because it is psychologically
implausible. Any argument, transcendental or otherwise, can give one
knowledge only if one understands the argument and uses it. But the
suggestion that the relevant transcendental arguments are actually used
in non-philosophical contexts is perhaps too implausible ever to have
been suggested.
So far I have been interpreting transcendental arguments as type-a
responses to skepticism; such arguments do not engage skeptical arguments at all but rather try to show that the skeptical conclusion cannot
be true. Another interpretation is to view them as type-b responses. On
this understanding, transcendental arguments are not supposed to prove
that skepticism is false but instead are used to reject something in the

74

skeptical reasoning. Specifically, they show that the possibilities that

skeptical arguments raise as alternatives to knowledge-claims are not real
possibilities, and that therefore an essential assumption of skeptical arguments is mistaken. For example, Descartes considers the possibility that
he is the bodiless victim of an evil genius, who makes it appear that
there are earth and stars and the like when in fact all of these things are
false. But this is not in fact a possibility, transcendental arguments show,
and so an assumption of Descartes' reasoning is mistaken.
On this interpretation transcendental arguments constitute type-b
responses to skepticism that is, they engage skeptical arguments, but
only superficially. Each of several considerations shows that this is the
case.
First, not all skeptical arguments invoke alternative possibilities to
knowledge-claims. For example, Hume's argument does not. Second,
not even Descartes' argument must assume that total illusions or nearly
total illusions are possibilities. For example, argument (D5) from Chapter
2 runs as well on the possibility of normal dreams as on the possibility
of philosophical ones. Third, even demon scenarios can be constructed
so that deception is highly restricted. Consider the possibility that the
world is for the most part as it appears to be but a deceiving demon
steps in from time to time to make one of our most confident judgments
false. The problem in this kind of scenario is not that all of our beliefs
might be false, but that any of our beliefs might be false. For any
particular knowledge-claim that one might make, it might be the case
that the demon makes that claim false while keeping many other claims
true. This last possibility is not ruled out by the argument from Davidson
or by any other transcendental argument we have considered.
This point echoes one that was made in Chapter 2: namely, skeptical
arguments need not start out by questioning all of our knowledge at
once. Another possibility, and one that is reflected in our reconstructions
of Descartes' argument, is that the skeptic may begin by raising doubts
about an individual knowledge-claim. But if there is nothing special
about that claim in particular, then the same doubts can be generalized
so as to undermine other knowledge-claims as well.
In conclusion, dismissive responses to skepticism fail to touch the
skeptical arguments we have reconstructed from Descartes and Hume.
One version of Descartes' argument seems to be open to a relevant
possibilities approach, although there remains work to be done there.
But argument (D5) does not seem open to that approach, and neither
do the arguments from Hume. These latter claim that there is no good

75

inference from appearance to reality, and so far we have seen nothing to

challenge that basic claim.
In Chapter 4 we turn to a different kind of response to skepticism
about the world. Specifically, we investigate the claim that the arguments from Descartes and Hume depend on a faulty ontology, or perhaps a faulty philosophy of mind. This kind of response is not dismissive,
since it engages the skeptical arguments in question seriously. It is mistaken, however or so I argue in the next chapter.

76

Skepticism about the World: Part

Three Dualism, Realism, and
Representationalism

At present, Western philosophy is divided among various schools. For

reasons that are partly philosophical and partly historical, philosophy as
it is practiced in the Anglo-American or "analytic" tradition often looks
very different from what goes on in traditional continental philosophy,
the tradition of American pragmatism, or the so-called post-modern
school. But across these various factions one discovers a point on which
there is odd agreement. A great many philosophers, including representatives from each of the schools mentioned, have defended essentially
the same diagnosis of skepticism about the world. Namely, they claim
that skepticism is not an epistemological problem at all, but is rather the
necessary consequence of a misguided modern ontology. Once that
ontology is given up, these philosophers agree, the problem of skepticism cannot even get off the ground.
Richard Rorty calls this characterization of the relationship between
skepticism and the modern ontology "the usual story," citing Etienne
Gilson and J. H. Randall as two of the story's prominent proponents. 1
Here are some characteristic statements of the position, beginning with
two from a founding father of Anglo-American philosophy, George
Berkeley:
[W]e have been led into very dangerous errors, by supposing a two-fold existence of the objects of sense - the one intelligible or in the mind; the other real
and without the mind, whereby unthinking things are thought to have a natural
subsistence of their own, distinct from being perceived by spirits. This, which,
if I mistake not, hath been shown to be the most groundless and absurd notion,
1

See Richard Rorty, Philosophy and the Mirror of Nature (Princeton: Princeton University Press,
1979), p. 49, n. 19, and pp. 51-52, n. 21.

77

is the very root of Scepticism. (A Treatise concerning the Principles of Human

Knowledge, pp. 107-108)
[S]o long as men thought that real things subsisted without the mind, and that
their knowledge was only so far forth real as it was conformable to real things, it
follows they could not be certain that they had any real knowledge at all. . . .
All this sceptical cant follows from our supposing a difference between things
and ideas, and that the former had a subsistence without the mind or unperceived. (Ibid., p. 108.)
In the next passage another great of analytic philosophy, J. L. Austin,
takes a swipe at Berkeley but makes essentially the same point in different terms.
But what is generally, and most importantly, wrong with [the anti-skeptic's]
argument is simply that he has got into (perhaps has let Berkeley lead him into)
the position of swallowing the two-languages doctrine - temporarily, at least,
appearing to swallow the two-entities doctrine on the way. And the resulting
question about how the evidence-language fidea'-language) is related to the
material-object-language, which he tries to answer, is a question that has no
answer, it's a quite unreal question. The main thing is not to get bamboozled
into asking it at all. {Sense and Sensibilia, p. 142)

While this diagnosis of skepticism can be found in analytic philosophy, it is almost ubiquitous in the Continental tradition. In this regard it
is helpful to understand that much of Continental philosophy is essentially Kantian, beginning from the premise of Kant's Copernican revolution.2 Roughly, the line of thinking is that Hume's skepticism presupposes a distinction between mind and mind-independent reality. To
avoid skepticism we must give up empirical realism in favor of transcendental idealism. In other words, we must give up the idea of a knowable
mind-independent reality. On the basis of roughly this line of thought,
many Continental philosophers now take it for granted that the object
of knowledge is created (or at least shaped) by the mind that knows it,
and Continental philosophy is largely devoted to exploring the implications of this idea.
Martin Heidegger's analysis differs in critical respects from Kant's, but
it is clear that Heidegger agrees at least with this much: the problem of
skepticism is rooted in inadequate ontology and more specifically, in a
dualism of internal knowing subject and external object of knowledge.
2 A notable exception is the realist tradition in phenomenology.

78

Now the more unequivocally one maintains that knowing is proximally and
really 'inside' and indeed has by no means the same kind of Being as entities
which are both physical and psychical, the less one presupposes when one
believes that one is making headway in the question of the essence of knowledge
and in the clarification of the relationship between Subject and Object. For only
then can the problem arise of how this knowing subject comes out of its inner
'sphere' into one which is 'other and external,' of how knowing can have any
object at all, and of how one must think of the object itself so that eventually
the subject knows it without needing to venture a leap into another sphere.
(Being and Time, p. 87)

In Continental philosophy the KantianHeideggerian diagnosis of skepticism is found to be so convincing that it is not an exaggeration to say
that it is taken for granted. It is largely for this reason that there is no
recognizable subdiscipline of epistemology in Continental philosophy,
and it is the reason that so many Continental philosophers accept some
form of anti-realism.3
Unfortunately, a similar line of thinking has been convincing to some
in America as well. In the following passages John Dewey joins Kant
and Heidegger in blaming bad ontology for skepticism, and with the
same anti-realist results:
For these questions [of epistemology] all spring from the assumption of a merely
beholding mind on one side and a foreign and remote object to be viewed and
noted on the other. They ask how a mind and world, subject and object, so
separate and independent can by any possibility come into such a relationship
to each other as to make true knowledge possible. (Reconstruction in Philosophy,
p. 123)
Those who have followed the previous discussions will not be surprised to hear
that, from the standpoint of experimental knowing, all of the rivalries and
connected problems [of epistemology] grow from a single root. They spring
from the assumption that the true and valid object of knowledge is that which
has being prior to and independent of the operations of knowing. (The Questfor
Certainty, p. 196)

Finally, essentially the same diagnosis of skepticism is embraced by

the post-modern school. Influenced by Heidegger and Dewey, post3 Dreyfus argues that Heidegger is not an anti-realist. See Hubert L. Dreyfus, Being-in-theWorld (Cambridge, MA: MIT Press, 1991), esp. pp. 252-265. This is controversial, but in
any case my purpose here is not to contribute to Heidegger scholarship; rather, I am interested in critiquing a general line of thought that has been influential in Continental philosophy, whether or not it can be correctly attributed to Heidegger.

79

modernists like Richard Rorty view epistemology as the activity of

working out solutions to the problem of skepticism. But that activity
and the problem it tries to solve both depend on the modern ontology
for their existence. Once that ontology is given up, there is no place left
for either epistemology or its central concern:
The idea of a discipline devoted to "the nature, origin, and limits of human
knowledge" the textbook definition of "epistemology" required a field of
study called "the human mind," and that field of study was what Descartes had
created. The Cartesian mind simultaneously made possible veil-of-ideas skepticism and a discipline devoted to circumventing such skepticism.4
These passages show that philosophers representing a wide range of
traditions can nevertheless share common ground. Let us take a closer
look at what that common ground is.
First, all of these philosophers call attention to a recognizably "modern" ontology of the self and the world. By this I mean one that is
Cartesian in its inspiration, and that distinguishes between an internal
knowing subject and an external object of knowledge. Here "internal"
is essentially a designation for mental or spiritual stuff, whereas "external" is used to designate a different ontological kind. We can use the
terms "physical" or "material" here, but they do not make explicit what
is perhaps the most important point at issue. That is, whereas subjects
are by nature a kind of mental substance, objects are mind independent in
the following sense. According to the modern ontology, objects in the
world exist and have many of the properties they do, independently of
whether any mind exists or how any mind thinks.
This is opposed to mental things such as minds, sensations, thoughts,
and thinkings, which depend for their existence on the existence and
activity of minds. There are no thoughts without a thinker and (trivially)
no thinkers without a thinker. Mind-independent reality is also opposed
to socially constructed things such as marriages and home runs. Marriages and home runs are not mental things, but they depend for their
existence on mental things. Unlike rocks and trees, they are mind dependent in the sense that they would not exist, or have many of the
properties they do, if minds did not think in one way rather than
another.
So understood, the modern ontology combines a dualism of mind
and material world with a realism about the latter. Another thesis often
4 Rorty, p. 140.

80

associated with the modern ontology, but only hinted at in the quoted
passages, is representationalism. According to this last doctrine, thought
about objects requires thought about ideas that represent them. A subthesis of this position is that perception is representational, requiring
thought about the sensations that material objects cause in perception.
As we shall see, representationalism is not really an ontological thesis at
all but is rather a position in the philosophy of mind that is sometimes
confused with the modern ontology.
A second shared theme in these passages is that the problem of
skepticism is created by the modern ontology. More specifically, the
problem arises as to how internal subjects can come to know external
objects, or to even think such objects in the first place. This second
theme can be broken down into two claims. First, it is claimed that the
modern ontology is necessary for generating skepticism about the world.
Thus Heidegger claims that "only then can the problem arise of how
this knowing subject comes out of its inner 'sphere' into one which is
'other and external'." Second, the claim is made that the modern ontology is sufficient for generating skepticism. Thus Berkeley suggests that
accepting the modern ontology inevitably results in skepticism about the
objects of sense. A corollary of the first claim is that rejecting the modern
ontology is sufficient for resolving the problem of skepticism. A corollary
of the second claim is that rejecting the modern ontology is necessary
for resolving that problem.
A third theme in the quoted passages is that the modern ontology
should be given up in favor of some version of anti-realism. With the
exception of Austin, all of the authors cited suggest that the objects we
know are created or shaped by our thinking. Of course, none of our
authors would call himself an anti-realist. All of them would say that to
accuse them of anti-realism is to remain in the false ontology that makes
the distinction between realism and anti-realism possible. But on the
standard meaning of the term, "anti-realism" is the position that the
object of knowledge is not independent of our knowing it. Put differently, anti-realism claims that the reality we know is created by, or at
least shaped by, the way we know it, or more generally, the way we
think about it. This position is in opposition to the view that we know
things as they are that is, as they are independently of our knowing
them or thinking them.
Now if this is what is meant by anti-realism, then Berkeley, Heidegger, Dewey, and Rorty are endorsing some form of anti-realism over
the modern ontology. They all think that the ontological distinction

81

between mental subject and mind-independent object of knowledge is

at the root of skepticism about the world, and they all recommend that
we erase the distinction by denying that the object of knowledge is
independent of the way we know it or think about it. Another way to
understand their common position is to say that for them all objects of
knowledge are like marriages and home runs, in that they are somehow
constructed by our thinking. On Berkeley's idealism all objects are
literally mental objects. On less radical forms of anti-realism there is a
distinction between the mental and the non-mental, but even nonmental objects depend for their existence and properties on the ways we
think about them.
My thesis in this chapter is that all of this is wrong-headed. I argue
that the modern ontology is not necessary for generating skepticism,
since powerful arguments for skepticism can be constructed without it.
Such arguments depend on an innocent appearancereality distinction
rather than any specific way of cashing out that distinction, modern or
otherwise. Neither do skeptical arguments depend on representationalism, or the thesis that thinking about objects requires thinking about
ideas that represent them. Representationalism has been closely associated with the modern ontology and is sometimes confused with it. But
skeptical arguments depend on neither the modern ontology nor a
representationalist theory of ideas.
Second, neither the modern ontology nor representationalism is sufficient for generating skepticism, since the best skeptical arguments depend on incorrect assumptions about knowledge and evidence. It is
these assumptions that really drive the skeptical arguments, and that
should be the focus of our inquiry. On the view that I am defending,
different ontologies and theories of ideas amount to so much window
dressing, sprucing up the arguments but playing no substantive role in
their reasoning.
Finally, because skepticism is an epistemological problem and not an
ontological one, the most common motivation for anti-realism is just
misguided. On the one hand, it is not necessary to embrace anti-realism
to refute skeptical arguments. On the other, embracing anti-realism does
not do the job; the best skeptical arguments from Descartes and Hume
run just as nicely on anti-realist ontologies. So, for example, even if we
adopted Berkeley's radical idealism, the skeptical argument from Hume
would still be a powerful one.
In Part I of this chapter, I consider whether the modern ontology or
representationalism is necessary for generating powerful skeptical argu-

82

ments about our knowledge of the world. In Part II, I consider whether
either of these doctrines is sufficient. But before proceeding it should be
stressed that my purpose is not to defend either the modern ontology or
representationalism. Rather, it is to show that the relationship between
these theses and skepticism is not what it is commonly and widely
claimed to be. Accordingly, skeptical arguments teach epistemological
lessons, not lessons in ontology or the philosophy of mind.

/. What Is Necessary for Skepticism about the World

It is now time to take a closer look at how the modern ontology and
representationalism are supposed to give rise to skepticism. In order to
do so we should look at the best skeptical arguments available and ask
where they depend on the doctrines in question. Of course, I argue that
they do not depend on the modern ontology or on any particular
ontology whatsoever. Neither do they depend on a representationalist
theory of ideas. To the contrary, such arguments run nicely on any
ontology or theory of ideas that is even minimally plausible.
What I mean by this last point is the following: To be even minimally
plausible, any ontology and any theory of ideas must recognize certain
distinctions and be consistent with certain assumptions that are in themselves platitudinous. So, for example, any minimally plausible ontology
must recognize a distinction between appearance and reality, since it is a
platitude that things are not always as they appear. Although different
theories cash out the distinctions and assumptions in different ways, all
must include them as part of the data to be explained. Once these
distinctions and assumptions are in place, however, the skeptical arguments we have been considering are off and running.
Skeptical arguments are mistaken somewhere - that much is a working assumption of the present methodology. But they need not make
any mistake about ontology, or about how the way things appear represents the way things are. This, in any case, is the thesis of Part I.
1. THE MODERN ONTOLOGY AND SKEPTICISM

I argued in Chapter 2 that the best arguments for skepticism from

Descartes and Hume are versions of the "no good inference" argument.
Both of those arguments begin with the assumption that, in some broad
sense and at least in part, our beliefs about the world depend on the way

83

things appear to us. The second claim they share is that this dependence
is broadly evidential. If we give these two claims charitable readings,
then they are platitudes - no one who is not in the grip of a philosophical theory would think to deny them. But if we do accept them, then
we have most of the materials for a powerful line of skeptical reasoning.
Consider the following, which I think captures the common structure
of "no good inference" arguments for skepticism about the world:
(NGI)

1. Our beliefs about objects in the world depend, at least in part, on

the way things appear to us via the senses.
2. The nature of this dependency is broadly evidential the fact that
objects in the world appear a certain way is often our reason for
thinking that they are that way.
3. Therefore, if I am to know how objects in the world are, it must
be via some good inference from how things appear to me. (2,3)
4. But there is no good inference from the way things appear to the
way things are.
5. Therefore, I cannot know how objects in the world are. (3,4)

Let us consider the initial force of this argument. As I said, premise (1)
is properly considered to be a platitude. It states only that our beliefs
about objects in the world partly depend on the way those objects appear
to us. But then premise (2) is eminently plausible as well. It seems clear
that, at least in typical cases, something's appearing to be a certain way
constitutes a reason for thinking that it is that way. In Chapter 2 I said
that this is plausible even if we are characterizing sensory appearances
merely phenomenally. But certainly it seems right if we are thinking of
sensory appearances as already having conceptual content. In that case,
premise (2) says that something's seeming to be a certain way via the
senses is often one's reason for thinking that it is that way.
Moving to premise (4), which is the remaining independent premise
of the argument, we have what looks to be another plausible claim.
Initial support for (4) is provided by the uncontroversial claim that things
are not always as they appear. But we have seen that Hume gives an
independent argument for the premise. In a nutshell, any inference from
appearance to reality would have to be circular, since it must depend on
an empirical premise to the effect that the way things appear is a reliable
indication of the way things are. So even if (4) is not as obvious as (1)
and (2), there are good reasons to think it is true.
Now to the question of the present section namely, how is argument (NGI) supposed to involve the modern ontology? I can think of

84

only two ways in which it might be thought to do so. First, the

argument talks of appearances being our evidence for objects in the
world, and it might be thought that this in itself commits the argument
to the modern ontology of internal subjects and external objects. Second, it might be thought that premise (4) implicitly assumes the modern
ontology, since only such an ontology would make (4) seem plausible. I
now consider each of these suggestions in turn.
a. The AppearanceReality Distinction

As we have seen, argument (NGI) talks about appearances being our

evidence for objects in the world. And certainly one interpretation of
this talk is to think of ourselves as internal subjects, and to think of
objects in the world as having a fundamentally different and independent
ontological status. Appearances are then the mental intermediaries by
which internal subjects are allowed to think the external objects of
knowledge. On this interpretation, which is perhaps the interpretation
that both Descartes and Hume would endorse, the argument claims that
our knowledge of internal appearances cannot provide sufficient evidence for knowledge of external objects in the world.
But is there any reason why we must understand the skeptical argument in this manner? There seem to be at least two other options. First,
we could refrain from further cashing out the language of argument
(NGI) in any way. In other words, we could attempt to critique (NGI)
without committing ourselves to any deeper ontology of sensory appearances and reality. Certainly we understand the thrust of the argument as it stands, and so we could simply opt to take it at face value and
assess it accordingly.
Our second option is to cash out the skeptical argument in terms of
a different ontology, one that does not involve the offending distinction
between internal knower and external object of knowledge. For example, we could employ Chisholm's adverbial theory of perception, which
takes sensory appearances to be modes of the perceiver rather than
mental objects.5 On Chisholm's view, to have a blue sensation is to be
appeared to bluely, thus avoiding ontological commitment to intermediate objects between the knower and the thing known. If we are
thinking of sensory appearances as having conceptual content, to have a
sensory appearance of a tree is to be appeared to in a characteristic way
5

Roderick Chisholm, Perceiving: A Philosophical Study (Ithaca: Cornell University Press, 1957).

85

phenomenally (to be appeared to treely), and to take the object appearing to be a tree. Alternatively, we could understand the skeptical argument as materialists, considering appearances to be material (or perhaps
functional) states of the knower, and understanding beliefs in a similar
manner.
On any of these options we get a skeptical argument that is at least as
forceful as any involving a modern ontology. More specifically, what
drives the skeptical argument is (a) an innocent distinction between the
way things appear and the way things are and (b) some assumptions
about the nature of our evidence for objects in the world. But both the
distinction and the assumptions about evidence are independent of the
modern ontology. I have more to say regarding the assumptions about
evidence in Section Lib. But let us consider the appearancereality
distinction now.
First, the distinction between the way things appear and the way
things are is independent of any specific ontology. The modern ontology
of external objects and internal ideas represents one way to cash out the
distinction, but it is not the only way, or even the most plausible way.
Second, it is a mere platitude that things are not always as they appear,
and so any minimally plausible ontology will have to cash out the
distinction in some fashion or another. Third, it is the distinction itself,
not the modern understanding of it, that is driving the skeptical argument. If anything, the argument is stronger without the modern interpretation, since it is thereby relieved of what is only excess baggage.
But perhaps this last point is incorrect. For even if the language of
(NGI) can be cashed out independently of the modern ontology, perhaps the argument depends on that ontology in a different way. Specifically, one might think that premise (4) becomes plausible only by
adopting the modern version of the appearancereality distinction. In
other words, one might think that only the modern ontology makes it
plausible that there is no good inference from how things appear to how
things are. In order to decide this point we must look more closely at
the considerations that can be given in support of premise (4) and at
whether these considerations involve the modern ontology in any essential respect.
b. Inference from Appearance to Reality

It seems to me that there are three main considerations that can be

brought in favor of premise (4) of (NGI). The first is the Humean

86

argument about the impossibility of a non-circular inference from appearances to reality. As we have seen, the main idea of this argument is
that premises about how things appear cannot give support to conclusions about how things are, unless such premises include an assumption
about the relationship of appearances to reality. In other words, there
will have to be some assumption to the effect that the way things appear
via the senses is a good indication of the way things are. But then the
problem arises as to how that kind of assumption could be justified. As
Hume points out, any such justification would itself require the evidence
of appearances, and so would require the very assumption whose justification is in question. Therefore, Hume concludes, a non-circular inference from appearances to reality is impossible.
My own view is that this argument from Hume in favor of premise
(4) is correct. Or, to qualify, it is correct on an important understanding
of "support" that we will consider more closely in subsequent chapters.
What is relevant for present purposes is that the Humean argument,
whatever its merit, is completely independent of any particular reading
of the appearancereality distinction. Once again, it is the appearancereality distinction itself that is driving the argument, not any modern
interpretation of the distinction. The problem would not go away, for
example, if we were to think of appearances as the adverbialist does or
as the materialist does. For on any interpretation of the appearancereality distinction it will seem that we need an assumption to the effect
that appearances are a reliable guide to reality. And on any remotely
plausible interpretation of that distinction, such an assumption will not
itself be a necessary truth but rather a contingent truth about the way
things in fact are. But then the central premises of the Humean argument
are in place, and the problem of circularity arises.
Even if these Humean considerations could be avoided, there is
another argument in favor of premise (4) that is, in my opinion, unanswerable. Even if a non-circular inference from appearances to reality
were in principle possible, no such inference would be psychologically
plausible. In other words, it would not be plausible that such an inference is actually used when we form beliefs about objects on the basis of
sensory appearances. This is because an inference takes us from belief to
belief, but we do not typically have beliefs about appearances. In the
typical case, we form our beliefs about objects in the world without
forming beliefs about appearances at all, much less inferring beliefs about
the world from beliefs about appearances. And notice that this point
holds independently of how we are thinking of the ontology of appear-

87

ances. For example, we do not typically have beliefs about the ways in
which we are appeared to, and we certainly do not typically form beliefs
about what functional states we are in.
A third argument in favor of premise (4) also involves considerations
of psychological plausibility and would hold even if we could remove
the difficulty regarding beliefs about appearances. The argument is made
by Hume, but it has not been adequately appreciated. Specifically, any
inference that is particularly clever or sophisticated is unlikely to play a
role in perceptual knowledge, because even brutes and small children
learn from experience. And so invoking any complex inference here is,
once again, psychologically implausible. It is not plausible, for example,
that we actually make some complex inference to the best explanation
when we see that there is a cup on the table. Introspection certainly
does not reveal any such inference. And a small child knows that there
is a cup on the table before she is capable of much simpler reasoning.
This is a place where the empirical facts of human cognition trump
philosophical theorizing. And once again, these points do not depend
on our understanding appearances in any specific way, modern or otherwise.
We may conclude that the skeptical argument represented in (NGI)
does not involve the modern ontology in any essential respect. A quick
look at arguments (H2), (H3), and (D5) from Chapter 2, all versions of
the "no good inference" argument, confirms the points that have been
made with regard to (NGI). All of these arguments are driven by an
appearancereality distinction and some assumptions about the nature of
our evidence for objects in the world, but neither that distinction nor
those assumptions require a modern ontology for their interpretation or
their support. Therefore, the modern ontology of internal subjects and
external objects is not necessary for generating powerful arguments for
skepticism about the world. For the same reasons, rejecting the modern
ontology is not sufficient for refuting these arguments. Philosophers who
have thought that it is have taken false refuge in that position.
2. REALISM AND SKEPTICISM
I have been arguing that skepticism about the world does not depend
on a dualistic modern ontology. The strategy I have used to make the
argument is rather simple: I have shown that the best skeptical arguments
run just as well on other ontologies. More specifically, I have shown
that the arguments can run on a monistic materialist ontology, and that

88

they can run on an adverbialist ontology that is consistent with both

dualism and monism. But here someone might object that I have been
missing the point. What is wrong with the modern ontology, the objection goes, is not its dualism but its realism. So long as we think that
objects in the world are mind independent we will have the skeptical
problem, and that is why skeptical arguments run on materialism and
adverbialism as well as on Descartes' dualism.
It is fairly easy to see, however, that this objection to the reasoning
of Section LI is incorrect. It is incorrect because adverbialism is not
committed to realism; although an adverbialist understanding of appearances is consistent with realism about objects in the world, it is also
consistent with anti-realism about objects in the world. Therefore, the
reasoning in Section I.I shows that the best skeptical arguments do not
depend on realism. They can run on adverbialism combined with antirealism.
To drive home the point, consider that even a radical idealist like
Berkeley should be worried by "no good inference" arguments. That is,
even on Berkeley's extreme anti-realism those arguments run just as
well. The reason this is so is that even Berkeley must maintain a distinction between appearance and reality - any minimally plausible ontology
must do so. But this is enough to get "no good inference" arguments
going, and nothing about Berkeley's ontology blocks the reasoning to
their skeptical conclusion.
Let us take a closer look. In Berkeley's idealism all that exists are
minds and ideas. Leaving aside the niceties, objects such as tigers and
trees are essentially well-ordered bundles of ideas. But even here we
have a distinction between "real" objects and mere appearances. For
example, Berkeley's ontology makes a distinction between actual tigers
and hallucinations of tigers. The latter, of course, are not well ordered
they lack the coherence and stability that bundles of ideas constituting
"real" tigers possess. But then what is our evidence, on Berkeley's view,
that there is a real tiger before us? In cases of perception, it has to be the
sensory appearances that we have at the moment. And of course momentary appearances, even if they are over several moments, can be deceiving. What appears to be a real tiger need not be, since initial stability
and coherence might later give way to the incoherence of an illusion.
All of this is as it would have to be. For of course our evidence for
objects in the world is the way that things appear to us, and of course
appearances can be deceiving. Berkeley accepts these as the platitudes
that they are and tries to account for them within the framework of his

89

idealism. But since he accepts these platitudes, the skeptical argument is

off and running. Certainly premises (1) and (2) of (NGI) remain true on
Berkeley's anti-realism. And the reasoning for premise (4) will be the
same as well.
Therefore, "no good inference" arguments run as well on radical
idealism as they do on realism. This is an important result. The most
common motivation for anti-realism is the attempt to avoid skepticism,
but anti-realism does not do the job.

3. REPRESENTATIONALISM AND SKEPTICISM

Some philosophers, however, will charge that I am still missing the
point. It is not the mindbody dualism of the modern ontology that
leads to skepticism, or even its realism about objects in the world.
Rather, the problem with the modern ontology is its representationalist
theory of ideas. A necessary assumption of the skeptical arguments, the
objection goes, is that we think about objects by first thinking about
ideas. This in turn sets up the requirement that our ideas "agree" with
their objects, or that our ideas accurately "represent" them. But once
we conceive things this way, it becomes impossible to know that our
representations do in fact agree with their objects. That is the real point
of the Heideggerian and post-modern critiques.
The following passages from Heidegger and Rorty suggest that this is
indeed what these philosophers have in mind:
What is thus perceived and made determinate can be expressed in propositions,
and can be retained and preserved as what has thus been asserted. This perceptive retention of an assertion about something is itself a way of Being-in-theworld; it is not to be Interpreted as a 'procedure' by which a subject provides
itself with representations of something which remain stored up 'inside' as
having been thus appropriated, and with regard to which the question of how
they 'agree' with actuality can occasionally arise. [Heidegger, Being and Time,
p. 89]
In Descartes's conception the one which became the basis for "modern"
epistemology it is representations which are in the "mind." The Inner Eye
surveys these representations hoping to find some mark which will testify to
their fidelity. Whereas skepticism in the ancient world had been a matter of a
moral attitude, a style of life, a reaction to the pretensions of the intellectual
fashions of the day, skepticism in the manner of Descartes's First Meditations [sic]
was a perfectly definite, precise, "professional" question: How do we know

90

that anything which is mental represents anything which is not mental? (Rorty,

Mirror of Nature, pp. 4546)

If this is the point of Heidegger's and Rorty's critiques, then it is badly
put when in other places they blame ontology for skepticism.6 For
representationalism is not an ontological thesis at all but a thesis about
how objects are thought. In other words, it is a position in the philosophy of mind. To see that this is so, consider that the modern ontology
does not entail representationalism, and representationalism does not
entail the modern ontology.
First, representationalism does not entail the modern ontology. R e member that representationalism is the thesis that one can think about
objects in the world only by thinking about ideas that represent those
objects. But nothing about this thesis suggests that either objects or ideas
must have any particular ontology. So, for example, an adverbialist about
ideas can be a representationalist, holding that one can think about
objects in the world only by thinking about the modes of thought that
represent them. Appearances or ideas considered as modes of thought
can serve just as nicely to represent objects as appearances or ideas
considered as mental objects. Neither does representationalism entail
anything about the ontology of objects in the world. Even an idealist
can be a representationalist, and Berkeley's theory of perception bears
this out.
Second, the modern ontology does not entail representationalism. As
we saw, that ontology is best characterized as a dualism between mind
and body, combined with a realism about the latter. Minds are mental
or spiritual substances while bodies are material substances, and the two
are considered to be of independent ontological kinds, meaning that
neither depends for its existence on the other. But nothing in that
position entails that thinking about material objects takes place by thinking about mental objects that represent them. An alternative position is
that minds can think about material objects directly. In other words, a
modernist can hold that although we think about objects with ideas, we
do not do this by thinking about the ideas. As the medievals put it, ideas
are the medium quo rather than the medium quod of thought about material
objects.
To make the point a different way, consider that there are two senses
of the term "object." In an ontological sense an object is a relatively

Dreyfus (p. 50) claims that it is not Heidegger's point.

91

stable and independent entity. In this sense of "object," the modern

ontology posits two kinds: mental and material. The second sense of
object is intentional. In this sense, to be an object of thought is to be
thought about. And of course something does not have to be an object
in the ontological sense to be an object in the intentional sense. Thus
one can think about acts, modes, and events as well as material objects
and persons. And now the point is this: the modern ontologist can hold
that ideas and appearances are objects in the ontological sense without
holding that they must always be objects in the intentional sense. Ideas
qua ontological objects might function in our thinking without being
intentional objects.
So the modern ontology is logically independent of a representationalist theory of ideas. One can be a representationalist without being a
modern ontologist, and one can be a modern ontologist without being
a representationalist. But that question aside, is representationalism necessary for generating skepticism? The answer is no, since nothing in the
skeptical arguments presupposes representationalism. Let us focus again
on argument (NGI).
Premise (1) of (NGI) says that our beliefs about objects in the world
depend on the way things appear to us. But it does not say how our
beliefs depend on appearances it does not say, for example, that
appearances must be objects of our thought before we can think about
objects in the world. On the contrary, (1) is consistent with the thesis
that appearances are the medium quo of thought about objects.
Premise (2) does say something about the way beliefs depend on
appearances; it says that the dependence is in some sense evidential. But
once again, this is consistent with the thesis that appearances are the
medium quo of thought. One might hold that the way things appear
serves as our evidence for the way things are but hold that one need not
actually think about that evidence for it to play its epistemic role. This
might sound odd, since it seems odd that something could serve as
evidence without our thinking about it. But we can see that the position
is not odd if we are careful about an important distinction. Namely,
there is a distinction between thinking about something and being
conscious of it. Put differently, there is a difference between thinking
about a thing and its being within one's consciousness.
To illustrate the difference, consider that you might drive on a highway for miles while absorbed in conversation. During this time you
might not think about the road at all a kind of automatic pilot kicks
in until either your interest in the conversation wanes or something on

92

the highway requires your attention. But although you do not think
about the highway, presumably you are conscious of it all the time.
Otherwise you would crash. Another example that illustrates the distinction is proprioception. At nearly every waking moment we are conscious of the position of our bodies, owing to receptors in our muscles.
This is why we do not bump into things, fall down, or knock things
over more often than we do. But we hardly ever think about the
position of our bodies. Our body position, in other words, is not constantly the object of our thought. Again, we can be conscious of things
without thinking about them.
It is plausible that this is the case with sensory appearances as well:
we are at every moment conscious of appearances, but we hardly ever
think about them. What we think about are the objects doing the
appearing, and that is why we are often ignorant of, or even wrong
about, their phenomenal properties. Of course, there are exceptions to
this. While an artist is painting, she might think hard about how a thing
appears, and any of us can turn our attention to the many shades of
color presented by a single facing wall. But these exceptions prove the
rule; we do not typically think about appearances, although we are
conscious of them at nearly every waking moment.
If this is right, then we can make good sense out of the claim that
sensory appearances serve as evidence without our thinking about them.
For although we do not typically think about appearances, we are at
every waking moment conscious of them. And our being conscious of
them is what allows them to serve as our evidence for beliefs about the
world. This position not only makes sense - it is almost certainly correct.
On the one hand, we do not typically think about appearances. On the
other hand, it seems obvious that our evidence for beliefs about the
world involves the way things appear to us through the senses. Consider
again my perceptual belief that the cat is on the couch. As I have already
said, it seems obvious to me that the reason I believe this is that this is
how things visually appear. And I mean that this is my reason in the
epistemic sense of "reason" this is how I know that my cat is on the
couch. How else would I know?
Finally, premise (4) of (NGI) does not presuppose representationalism. In fact, one of the reasons I gave in favor of premise (4) was that
we do not typically have beliefs about how things appear to us, and so
an inference from appearance to reality would be lacking in premises.
Even if such an inference were possible in principle, I argued, the claim
that we actually employ it in forming beliefs about the world is psycho-

93

logically implausible. Far from requiring representationalism, premise (4)

is well supported by a denial of representationalism.
In conclusion, "no good inference" arguments run on an innocent
appearancereality distinction together with some assumptions about the
nature of our evidence for beliefs about the world. Such arguments do
not depend on mindbody dualism, realism about objects in the world,
or a representationalist theory of ideas.

II. What Is Sufficient for Skepticism about the World

We have seen that the modern ontology is not necessary for generating
arguments for skepticism about the world. But many philosophers have
thought that it is sufficient. Their reasoning is that once we posit an
ontological gap between mind and world, there is no way that the gap
can be bridged. They accept the skeptical reasoning that appearances
cannot be good evidence for a mind-independent world, and attempt to
block the conclusion by denying that there is any mind-independent
reality that we could want to know. Unfortunately, these philosophers
fail to notice that the skeptical reasoning works as well for minddependent objects as it does for mind-independent ones. As we have
seen, "no good inference" arguments work as well on Berkeley's idealism as they do on Descartes' realism.
What this shows is that skeptical arguments make some other mistake.
Specifically, "no good inference" arguments make some mistake in their
assumptions about our evidence for beliefs about the world, and this
mistake is preserved throughout different interpretations of their ontology. This means that the modern ontology is not sufficient for generating skepticism about the world you need the bad epistemology to get
the skeptical result. Or to put things more carefully, this shows that the
modern ontology is not sufficient for generating skepticism within the
context of "no good inference" arguments.7
To establish these claims it will be necessary to identify some mistaken
7

In Chapter 8, I will argue that the modern ontology is not sufficient to generate skepticism
via any argument at all. In this regard I apply the relevant possibilities approach to argument
(D4) from Chapter 2, and argue that an adequate theory of evidence reveals that the skeptical
scenarios are not relevant possibilities. This, in turn, shows that (D4) makes a mistake about
the nature of evidence. But if the modern ontology is not sufficient here either, then all of
the best arguments for skepticism about the world will have been considered and I will be
warranted in drawing this conclusion without qualification.

94

epistemological assumption in the skeptical arguments we have been

considering. This is the purpose of Section II. 1. Second, we will need
to show that a rejection of the argument along such lines is consistent
with the modern ontology. This is done in Section II.2.
1. THE BIG MISTAKE

What part of the skeptical argument can be plausibly rejected? As I have

said, premises (1) and (2) of (NGI) are mere platitudes. Part of what I
mean by this is that anyone should accept (1) and (2), no matter what
her ontology, her philosophy of mind, or her theory of evidence.
Many attempts to answer the skeptic have focused on premise (4),
but I think that this approach is misguided. I have already said that I
find Hume's argument about circularity convincing when it is understood appropriately. But more importantly, it seems to me that it is
empirically false that human cognition employs anything like an inference from appearance to reality. The very notion of an inference involves a movement from belief to belief on the basis of deductive or
inductive relationships of their contents. But it is psychologically implausible that we typically have beliefs about sensory appearances, and, even
if we do, it is implausible that we infer anything from them on the basis
of anything approaching adequate reasoning.
The point I am making here is not just about conscious beliefs and
conscious inferences. I think that the attribution of even subconscious
beliefs or inferences is empirically implausible. First, if we typically have
beliefs about appearances, then we should be able to remember them.
But on the contrary, I cannot reconstruct the way things appeared to
me even moments ago. I take this as evidence that I never had beliefs
about appearances in the first place. Second, it is well documented that
people are often wrong about the way things appear to them. For
example, people will say that a facing wall appears to be uniform in
color, but a more careful introspection reveals that it presents a myriad
of different shades. In a similar fashion, people will report that shadows
on a field of snow appear gray, while in fact such shadows appear blue,
as is also revealed by a closer introspection. A good explanation for this
kind of mistake is that people are not used to thinking about appearances
at all, and so when our attention is called to the way things appear we
are not very good at forming accurate beliefs about this.
Someone might nevertheless think that such beliefs are subconscious,
even if they are difficult to recall or articulate. For even if I am not very

95

good at remembering or describing appearances, surely I can say something about them when asked. But here we have to make a distinction
between a subconscious belief and a disposition to believe.8 That is, the
fact that I am disposed to form some beliefs about my sensory experience
when asked does not show that I had those beliefs all along. It is more
likely, I think, that such beliefs are originally formed upon consideration
of the question. In this case beliefs about appearances would be like
beliefs about previously unconsidered sums; if I ask you the sum of 122
and 345, you will form an accurate belief about this, but that is no
indication that you believed the answer all along. When we have this
distinction in mind, it is hard to maintain that we typically have even
subconscious beliefs about how things appear. And if we do not have
the beliefs, then we do not make inferences from them. Premise (4) of
the argument looks good.
What options for rejecting something in (NGI) are left? I suggest that
we focus on the move from premises (1) and (2) to (3). That move
seems initially plausible, but a closer look will reveal that it is not as
obvious at it might first appear. More specifically, the move implicitly
assumes that all evidential relations are inferential relations or at least it
assumes that the evidential relation between sensory appearances and
reality is inferential. Although these assumptions seem plausible and are
even widely accepted, I believe that this is the mistake that "no good
inference" arguments make.9
In the remainder of this section I try to make an initial case that this
is indeed the mistake made by the skeptical arguments we have been
considering. In Chapter 7, I defend a theory of knowledge that is
consistent with this diagnosis, and that provides a theoretical basis for it.
In other words, agent reliabilism not only shows that not all evidential
relations are inferential but explains why this is so. Very roughly, the
idea is as follows: According to agent reliabilism, knowledge is true
belief that arises out of reliable cognitive faculties and habits, or what we
may call "cognitive virtues." Such faculties and habits reliably put us in
touch with the world, providing a good cognitive fit between ourselves
8

A similar distinction is treated in Robert Audi, "Dispositional Beliefs and Dispositions to

Believe," NOUS 28 (1994): 419-434.
9 Many discussions of skepticism seem to endorse the assumption, at least implicitly. I have
already mentioned Ayers' classic account of skeptical arguments in The Problem of Knowledge
(Harmondsworth: Pelican, 1956), and Williams' endorsement of this. See also Baysean and
traditional coherentist accounts of evidence, both of which attempt to define evidential
relations in terms of various kinds of inference relations.

96

and our environment. But not all cognitive faculties are inferential faculties. Our perceptual faculties, for example, put us in touch with the
world by reliably producing true beliefs directly on the basis of sensory
inputs. Here "directly" does not mean without prior training, concepts,
or presuppositions. Rather, it means non-inferentially that is, not on
the basis of inferences from prior premises serving as reasons. If this very
general picture is correct, then sensory appearances can be good evidence without the benefit of any inference from appearances to reality.
Inference from prior beliefs to further beliefs is one way to reliably form
true beliefs about one's environment, but there are non-inferential ways
as well.
I say this is only the "rough idea" because knowledge requires more
than reliability. For one thing, beliefs must be subjectively justified as
well as objectively reliable to count as knowledge. But we will see that
agent reliabilism can accommodate this condition on knowledge as well.
What we need is an account of subjective justification on which the
forming of beliefs about the world directly on the basis of sensory
appearances turns out to be so justified. Again, I begin the initial case in
the remainder of this section and continue it in Chapter 7.
a. The Elements of Perception

I am presently challenging the implicit assumption of argument (NGI)

that all evidential relations involve inferential relations. Put another way,
I am challenging the assumption that all evidential relations involve a
deductive or inductive inference from one set of beliefs to some further
set of beliefs. What one would like is a theory of evidence that allows
this diagnosis of the skeptical argument.
The most general characterization of evidence is as follows: A cognitive state is evidence for another cognitive state if and only if being in
the first state tends to confer some positive epistemic status on the second
state. For example, hearing the doorbell ring is evidence for my belief
that there is someone at the door, insofar as hearing the doorbell ring
tends to make the belief in question rational, or justified, or warranted.
What we are looking for is a theory of evidence that allows cognitive
states to be in such an evidential relationship to our beliefs about the
world, but that does not characterize that relationship in terms of deductive or inductive inferences from belief to belief. Later in this chapter I
argue that there are at least two broad approaches to evidence that would
do the job, and I throw my hat in with one of them. But at this point a

97

short detour is necessary. Before we can take a closer look at these

approaches it will be necessary to consider the elements of perception in
a very general way.
I am interested here in quickly making some distinctions that we can
all agree upon, even if we would not agree about how to cash them
out. On a very general level, then, we may distinguish three elements in
the perception of an object: the phenomenal content involved in the
perception, the conceptual or representational content involved in the
perception, and the perceptual judgment arising out of such contents.
These elements need not be thought of as being temporally or even
ontologically distinct, and the distinctions are not meant to be exhaustive. My point is only that one can make these distinctions in a roughand-ready manner, and I make them here because they will help us to
understand the theories of evidence to be discussed shortly. Accordingly,
we can talk about the following elements of perception, my stipulated
name for each appearing in italics.
i. The uninterpreted qualia of sensations. (Sensory appearances characterized as
lacking conceptual content.)
ii. Interpreted experience. (Sensory appearances characterized as involving conceptual content.)
iii. Beliefs about objects in the world.
The elements of perception distinguished here are commonly considered to be in relations which are in a broad sense causal, but they are
also considered to be in broadly evidential relations. As we have seen,
which elements are the essential ones for the evidential relation is a
matter of dispute, but in general philosophers agree that one or both of
the elements in i and ii act as evidence for the beliefs in iii, in the sense
(explained earlier) that being in such states tends to confer positive
epistemic status on beliefs about the world. Some philosophers argue
that we should collapse i and ii into a thick concept of experience,
thinking that further distinctions are irrelevant to the evidential story.
The important relation, according to this way of thinking, is between
thick experience and belief. Others collapse the distinction between ii
and iii, thinking that the important relation is between uninterpreted
qualia and interpreted experience/belief. A third possibility is that we
must keep the distinctions among qualia, experience, and belief, thinking
that there is not one important evidential relation but two: one between
qualia and interpreted experience, the other between interpreted experience and belief.

98

In Chapters 7 and 9, I defend a substantive account of perception

that endorses the first position. On this view of perception, sensory
appearances qua perceptual evidence have both phenomenal and conceptual content. My purpose in the present chapter, however, is to avoid
all such controversies. This is the appropriate way to proceed at this
point in the discussion, since I am arguing that the problem with (NGI)
is not that it assumes one account of sensory appearances or another. The
argument does not depend on a characterization of sensory appearances
as thin or thick, and it does not depend on any particular ontology of
sensory appearances.
There are now two points that need to be emphasized in this context.
First, it is relatively uncontroversial that the elements we have distinguished really are elements of perception. No matter how we understand
the perception of objects, it must be acknowledged to somehow involve
phenomenal qualia, conceptual content, and belief Second, no matter
how we understand these elements of perception, epistemological problems arise. For on any plausible theory of evidence, either qualia are
evidence for interpreted experience/belief, or interpreted experience is
evidence for belief, or both. But how are such evidential relations to be
understood? I have claimed that these relations are not to be understood
in terms of deductive or inductive inferences from belief to belief. Once
again, we are looking for a theory of evidence that allows either qualia
or thick experience to be evidence for beliefs about the world, but that
does not characterize this relationship in terms of inferences.
We are now in a position to look at two approaches to evidence that
can do this. As I have said, I will not argue here that my favorite
approach is correct. For the purposes of this section I will only attempt
to motivate the intuition that one of these approaches is correct, and to
show that either of them would provide the materials for rejecting the
skeptical reasoning in (NGI). It will then be relatively easy to show that
both of these approaches to evidence are consistent with a modern
ontology. That will be done in Section II.2.
b. Two Approaches to Evidence

In order to understand the two approaches to evidence I have in mind

it will be useful to make an analogy to ethics. In ethics, theories of right
action may be distinguished into two general camps. The first understands right action in terms of an objective relationship to morally
valuable consequences, so that an action is right if and only if, as a matter

99

of fact, it produces (or has a tendency to produce) such consequences.

The second general approach understands right action in terms of correct
norms or rules. Here what makes an action right is not whether some
relation to consequences in fact obtains, but whether the action is allowed by some relevant set of action-guiding norms. The first approach
is broadly consequentialist, whereas the second approach is broadly deontological.
We can characterize approaches to evidence in an analogous fashion.
Broadly speaking, a consequentialist approach in epistemology understands positive epistemic status in terms of epistemically valuable consequences. On this approach, a cognitive state is evidence for a belief p if
and only if believing p on the basis of the cognitive state tends to result
in the agent believing truly with respect to p. Sometimes the basing
relation here is understood to be inferential, so that a belief q is good
evidence for a belief p if inferring p from q (when q is true) tends to
lead to believing p truly. But that is only one way to satisfy the basing
relation. Alternatively, on this approach we may say that an experience
e is good evidence for a perceptual belief p if believing p on the causal
basis of e tends to result in believing p truly. In this latter case there is
no question of a deductive or inductive inference from beliefs about
experience to beliefs about objects in the world. Rather, the point is
that the experience itself is evidence for the belief, insofar as having that
experience tends to result in believing the truth.
I have talked here in terms of the relationship between experience
and belief. But a similar account could be given regarding the relationship between qualia and belief, or between qualia and interpreted experience.10
Not surprisingly, a deontological approach in epistemology understands epistemic status in terms of correct epistemic norms. On this
approach a cognitive state is evidence for a belief if and only if correct
epistemic norms permit forming that belief on the basis of that cognitive
state. Here again the basing relation is sometimes understood as inferential, the idea being that correct epistemic norms permit some inferences and not others. But again the basing relation can be satisfied in
other ways as well. Thus it is possible that correct epistemic norms
permit a certain belief whenever one has a certain kind of experience.
10 Alvin Goldman has developed a series of this broad kind of position. See, e.g., his "What
Is Justified Belief?", in George Pappas, ed., Justification and Knowledge (Dordrecht: Reidel,
1979).

100

Alternatively, there could be norms governing the interpretation of

qualia, thus taking a cognitive agent from uninterpreted qualia to interpreted experience or even to full-blown belief. The idea here is that
there are epistemic norms that govern the formation of beliefs directly
on the basis of either qualia or experience, as opposed to norms which
govern inferences from beliefs about qualia and experience.11
Consequentialist theories of positive epistemic status stress "objective" or de facto reliability, whereas deontological theories stress "subjective" factors such as conformance to countenanced norms. A third
approach to evidence would be a mixed theory, requiring both an
objective and a subjective element for positive epistemic status. This is a
plausible approach in epistemology, since knowledge would seem to
require both an objective relation to the truth and appropriate cognitive
conduct. In somewhat different terms, knowledge would seem to require both objective reliability and subjective justification. Accordingly,
a mixed approach might hold that a cognitive state is evidence for a
belief p if and only if both (a) believing p on the basis of the cognitive
state tends to result in the agent believing truly with respect to p, and
(b) believing p on that basis is permitted by some relevant set of belief
guiding norms. But there are other ways to capture the idea of appropriate cognitive conduct, or subjective justification, without reference to
norms or rules. Such an alternative is preferable, I will argue, insofar as
the thesis that our cognition is governed by norms or rules is a controversial one. This second kind of approach is defended in Chapter 7,
under the guise of agent reliabilism.
We are now in a position to draw the conclusion of this section:
namely, either of the two broad approaches to evidence just considered
would ground the objection to (NGI) that I have proposed. If you will
remember, that objection questioned the move from premises (1) and
(2) to (3) in the skeptical argument. My thinking was that this move in
the argument implicitly assumes that all evidential relations involve inferential relations, and I objected to that assumption. Clearly, either of
the two approaches to evidence just considered would ground the present objection. If either approach is correct about the nature of perceptual
evidence, then argument (NGI) is invalid.

11 For a developed version of this kind of position, see John Pollock, Contemporary Theories
of Knowledge (Totowa, NJ: Rowman and Littlefield, 1986), esp. Ch. 5.

101

2. THE MODERN ONTOLOGY AGAIN

In the preceding section, I argued that the skeptical argument depends

on a controversial assumption about the nature of evidence, and that
either of two broad approaches to evidence would vindicate an objection to that assumption. My final task is to show that those theories of
evidence are consistent with the modern ontology. Doing so establishes
that the modern ontology is not sufficient for generating the skeptical
argument in (NGI), and that rejecting the modern ontology is not
necessary to defeat that argument.
It is fairly easy to see that the two theories of evidence are consistent
with the modern ontology. For both theories were explicated in terms
of various relations among qualia, thick experience, and belief, with no
commitment being made regarding the ontological statuses of these
entities; rather, I purposely characterized such entities in the most general and neutral terms. That being the case, it is now easy to show how
the modernist can accommodate the various moments of perception,
taking such moments themselves to be uncontroversial data that any
ontology would have to allow. For good measure we will note how the
adverbialist and materialist can do this as well.
How would a proponent of the modern ontology understand our
categories of perceptual elements? Presumably she would understand
qualia as some sort of mental objects and would understand experience
in that way too. Beliefs about the world would then be intentional
attitudes which somehow go through these mental intermediaries. A
strong version of this story weds representationalism to modernism, so
that beliefs about the world are directly about the mental intermediaries,
and only indirectly about mind-independent objects insofar as the mental entities represent, or copy, or picture those objects.
On the other hand, we need not understand our categories in that
way. Thus the adverbial theorist considers both qualia and thick experience to be modes of the knower, thereby avoiding ontological commitment to separately existing mental objects of any kind. On this view
beliefs are intentional attitudes that can be directly about objects in the
world, there being no intermediary objects for intentionality to go
through. Alternatively, the materialist thinks of qualia, experiences, and
beliefs as various material or functional states of the knower, invoking
no kind of non-material object or state or attitude.
But then each of the two theories of evidence considered above is

102

consistent with a modern interpretation. On the consequentialist theory

this amounts to saying that a quale or experience e (now understood as
mental objects serving as intermediaries between subject and object) is
evidence for an external object belief p if and only if believing p on the
causal basis of e tends to result in believing p truly. The deontological
approach can make a similar application. In this case a quale or experience e (again, conceived as mental intermediaries) is evidence for an
external object belief p if and only if some relevant set of epistemic
norms permit believing p on the basis of e. A mixed approach could
accommodate the modern ontology by combining the consequentialist
and deontological conditions in the obvious manner.
What all of this shows is that the skeptical argument in (NGI) depends
on a controversial assumption about evidence that is independent of the
modern ontology. Therefore the modern ontology is not sufficient for
generating the skeptical argument, and rejecting that ontology is not
necessary for defeating the argument.
3. REPRESENTATIONALISM AGAIN
Section II.2 also shows that representationalism is not sufficient for
generating skepticism, or at least not within the context of "no good
inference" arguments. This is because both consequentialist and deontological theories of evidence are consistent with representationalism,
and so the denial that all evidential relations are inferential is also consistent with that position.
It is interesting to ask whether representationalism is sufficient to
generate skepticism by a different route. After all, one would think that
there must be something to the critique of skepticism that has been
taken for granted in so many circles, and this would seem to be the last
possibility that the modern ontology, or at least a representationalist
version of it, is closely related to skepticism.
It might seem obvious that representationalism does lead to skepticism. For if thought about objects in the world requires thought about
ideas, then thought about objects is indirect. And if thought about
objects is indirect, then it must proceed by inferences from thought
about ideas. But then the materials for a "no good inference" argument
are in place, and the skeptical conclusion would seem to follow in this
way from representationalism. Let us take a closer look at this reasoning,
which in effect suggests the following skeptical argument:

103

(R)

1. The immediate or direct object of thought is always some idea in

the mind.
2. All thought about other things must be indirect, mediated by
thought about ideas of those things.
3. Objects in the world are not ideas.
4. Therefore, all thought about objects in the world is indirect, mediated by thought about ideas of those objects. (1,2,3)
5. Therefore, thought about objects in the world requires an inference
from thought about our ideas of them. (4)
6. But there is no good inference from ideas to objects in the world.
7. Knowledge requires good inferences.
8. Therefore, there is no knowledge of objects in the world. (5,6,7)

Premises (1) and (2) of the argument amount to representationalism, and

(1) and (3) suggest a modern version. The remaining independent premises, (6) and (7), are common property of "no good inference" arguments. So it looks as though a combination of the modern ontology and
representationalism is sufficient to generate skepticism about the world.
That analysis is mistaken, however, because (5) does not follow from
(4). The subconclusion stated in (4) is that thought about objects in the
world must be mediated by thought about ideas. This indeed follows
from the representationalist assumptions stated in (1) through (3). Let us
call this kind of mediation "conceptual," since it involves conceiving
(or thinking about) one thing in order to conceive (or think about)
another thing. The subconclusion stated in (5), however, is that thought
about objects must be mediated by inferences. It only remains to note that
conceptual mediation does not imply inferential mediation, or mediation
by inferences from prior premises. One might hold that thinking about
Xs requires thinking about Ys, without holding that beliefs about Xs
require inferences from beliefs about Ys.
To see that this is so, consider the process of interpreting a piece of
art. To interpret a painting, for example, presumably one must think
about it: for example, one must form beliefs about certain physical
aspects of the painting. But the way in which interpretation takes place
is not that we infer an interpretation from premises about the physical
features of the painting; beliefs about the painting's physical features do
not figure in an argument from which we deductively or inductively
draw a conclusion about the painting's meaning. If you think that this is
how interpretation works, then try to reconstruct the argument. At best
the result will be awkward and unconvincing.
These considerations show that interpretation can take place in some

104

other manner. And this might be the case for the interpretation of
sensory appearances as well as for the interpretation of art. But then,
even if representationalism were correct even if we did form beliefs
about the world on the basis of thought about sensory appearances it
would not follow from this that we infer beliefs about the world from
beliefs about sensory appearances.
Is there another way in which representationalism entails skepticism?
Reid sometimes speaks as if he thinks that there is. Reid thinks that a
representationalist theory of ideas was shared by ancient and modern
philosophers alike, and that it was sufficient to generate skepticism:
Modern philosophers, as well as the Peripatetics and Epicureans of old, have
conceived that external objects cannot be the immediate objects of our thought;
that there must be some image of them in the mind itself, in which, as in a
mirror, they are seen. And the name idea, in the philosophical sense of it, is
given to those internal and immediate objects of our thoughts. The external
thing is the remote or mediate object; but the idea, or image of that object in
the mind, is the immediate object, without which we could have no perception,
no remembrance, no conception of the mediate object. (Essays, p. 226ab)
We ought, however, to do this justice to the Bishop of Cloyne and to the
author of the "Treatise of Human Nature," to acknowledge, that their conclusions are justly drawn from the doctrine of ideas, which has been so universally
received. . . . The theory of ideas, like the Trojan horse, had a specious appearance both of innocence and beauty . . . but carried in its belly death and destruction to all science and common sense. (Ibid., p. 132ab)
In these passages Reid calls our attention to a representationalist theory
of ideas that he thinks is common property throughout the history of
philosophy, and he credits Berkeley and Hume with recognizing that
theory's skeptical consequences. But is it the representationalism of the
theory of ideas that is the culprit? Reid attributes the following skeptical
argument to Berkeley. Notice that it is not a "no good inference argument," but instead trades on the inability of ideas to represent a material
world.
Bishop Berkeley gave new light to this subject, by shewing, that the qualities of
an inanimate thing, such as matter is conceived to be, cannot resemble any
sensation; that it is impossible to conceive anything like the sensations of our
minds, but the sensations of other minds. . . . But let us observe the use the
Bishop makes of this important discovery. Why, he concludes, that we can have
no conception of an inanimate substance, such as matter is conceived to be, or
of any of its qualities; and that there is the strongest ground to believe that there

105

is no existence in nature but minds, sensations, and ideas. . . . But how does this
follow? Why, thus; We can have no conception of anything but what resembles
some sensation or idea in our minds; but the sensations and ideas in our minds
can resemble nothing but the sensations and ideas in other minds; therefore, the
conclusion is evident. (Inquiry, pp. 131b132a)

The argument that Reid here attributes to Berkeley may be reconstructed as follows:
(B)

1. The only immediate objects of thought are ideas and sensations.

2. All thought of other things must be by means of ideas or sensations
that represent them.
3. In the case of material objects, the ideas or sensations that mediate
our thought must be images or resemblances of those objects.
4. No idea or sensation resembles any material object.
5. Therefore, there is no thought or perception of material objects.
(1,2,3,4)

Premises (1) and (2) of Berkeley's argument amount to representationalism, and so that theory is playing a role here. Moreover, Reid is
convinced by other arguments from Berkeley that (4) is true. But premise (3) is a thesis of Berkeley's empiricism rather than an essential aspect
of representationalism. That is, representationalism per se is not committed to the thesis that representation must be by resemblance. But then
representationalism is not sufficient to generate skepticism via Berkeley's
inconceivability argument. For that you need a bad theory of representation; that is, you need the assumption that representation must be by
resemblance.
Here we might have a case where the analysis of a skeptical argument
drives positive philosophy of mind rather than epistemology. For premise (3) is not an epistemological thesis, but a thesis in the philosophy of
mind, or perhaps in semiotics. On the other hand, the thesis that representation requires resemblance is so implausible that it is hard to say that
we have learned any lesson at all. This is perhaps one of those cases
where an assumption can be rejected as soon as it is made explicit.

4. CONCLUSIONS

I close by drawing some conclusions from the discussion so far. First,

neither the modern ontology nor representationalism is necessary for
constructing powerful arguments for skepticism about the world. Specifically, "no good inference" arguments require neither a mindbody

106

dualism, a realism about objects in the world, nor a representationalist

philosophy of mind. Such arguments are consistent with these but by
no means need them.
Second, neither the modern ontology nor representationalism is sufficient for generating skepticism about the world, since the best skeptical
arguments make an epistemological mistake. "No good inference" arguments assume that all evidential relations are inferential, or at least that
the evidence of sensory appearances must be inferential. But although
these assumptions are pretheoretically plausible and even widely accepted, this is not how perceptual knowledge works. An adequate theory of knowledge and evidence should show why this is so, explaining
how sensory appearances can act as evidential grounds for beliefs about
the world, even if there is no good inference from appearance to reality.
Third, all refutations of skepticism that are pragmatic or rhetorical are
also misguided. Since the problem of skepticism is not an existential
one, pragmatic and rhetorical considerations regarding the skeptic's way
of life, or her sanity, or her position in a debate are all out of place.
What makes skepticism a philosophical problem is not that there are
skeptical people, but that there are skeptical arguments. So long as those
arguments proceed on assumptions and reasoning that seem plausible to
us, they present a problem about how we are to rethink our previous
opinions about knowledge and evidence. As such, skeptical arguments
are not merely problems, but are also valuable methodological tools for
driving us toward better epistemologies.
Finally, all of this means that Heidegger, Dewey, Rorty, and others
are wrong about the legitimacy of epistemology's project. If skepticism
is conceived as a philosophical problem that is, as an argument that
begins from plausible assumptions and ends in an unacceptable conclusion then the only way to refute it is to challenge something essential
to the argument and to replace it with something better. This is a
legitimate philosophical enterprise, and it is exactly what epistemology
tries to do. If, on the other hand, skepticism is conceived as an existential
problem, then there is no problem. No one lives in the grip of skepticism - not even philosophers who proclaim themselves to be skeptics.
For this reason there is no need for philosophical "therapy" in addressing
skepticism. There is no existential sickness for therapy to cure.

107

The Argument from an Infinite

Regress of Reasons

In the first four chapters I have been claiming that skeptical arguments
play an important role in philosophical inquiry. Such arguments act as
heuristic devices for driving positive epistemology in particular, as opposed to ontology or philosophy of mind. I mean this thesis to be both
prescriptive and descriptive. On the one hand, I am claiming that skeptical arguments ought to play this methodological role. On the other, the
claim is that such arguments do in fact play it. Nothing supports the
descriptive thesis more than the literature on the skeptical argument
from an infinite regress of reasons. That argument is beautifully simple,
but it has inspired debate over the nature of knowledge and evidence
for over two millennia.

1. THE REGRESS ARGUMENT AND STRONG PARTICULARISM

The problem arises because it seems that one must have good reasons
for whatever one claims to know. But not any reason is a good reason;
one must have reasons for thinking that one's reasons are true. Accordingly, it seems that knowledge requires (per impossibile) an infinite regress
of reasons. An early version of the argument is attributed to the ancient
skeptic, Pyrrho. The passage quoted next is taken from Sextus Empiricus's discussion of Agrippa's five skeptical modes leading to the suspension of judgment. Agrippa, in turn, was systemizing the skeptical teachings of Pyrrho.1
1

For an excellent discussion of Agrippa's "Five Modes" and their relationship to contemporary theories of knowledge, see Fogelin, Pyrrhonian Reflections on Knowledge andJustification,

esp. Part II.

108

The Mode based on the extension to infinity is the one in which we say that
the proof offered for the verification of a proposed matter requires a further
verification, and this one another, and so on to infinity, so that since we lack a
point of departure for our reasoning, the consequence is suspension of judgement. (Sextus Empiricus, Outlines of Pyrrhonism, 1.15)

A similar problem is found in Plato's Theaetetus. Socrates is considering

the definition of knowledge as "true opinion, combined with definition
or rational explanation" and raises the problem of where an explanation
might end. Here Socrates recounts something he heard in a "dream":
But none of the primeval elements can be defined; whereas the things which
are compounded of them, as they themselves are complex, are defined by a
combination of names, for the combination of names is the essence of a definition. Thus, then, the elements or letters are only objects of sense perception,
and cannot be defined or known; but the syllables or combinations of them are
known and expressed, and are apprehended by true opinion. When, therefore,
any one forms the true opinion of anything without rational explanation, you
may say that his mind is truly exercised, but has no knowledge; for he who
cannot give and receive a reason for a thing, has no knowledge of that thing;
but when he adds a rational explanation, then, he is perfected in knowledge and
may be all that I have been denying of him. (202bc)
In Plato's terms, the problem is that knowledge seems to require a
rational explanation, but no explanation is possible with the "primeval
elements." Eventually Socrates seems to endorse a kind of foundationalism, insisting that the "simple elements" are k n o w n even though they
are not susceptible to further explanation:
Then, if we argue from the letters and syllables of which we have experience to
other simples and compounds, we shall say that the letters or simple elements as
a class are much more distinctly known than the syllables, and much more
indispensable to a perfect knowledge of any subject; and if someone says that
the syllable is known and the letter unknown, we shall consider that either
intentionally or unintentionally he is talking nonsense? (206b)
T h e problem is more clearly articulated by Aristotle in the Posterior
Analytics, and he more clearly opts for foundationalism:
Some people think that because you must understand the primitives there is no
understanding at all; others that there is, but that there are demonstrations of
everything. Neither of these views is either true or necessary.
The one party, supposing that you cannot understand in any other way, claim
that we are led back ad injinitum on the ground that we shall not understand the

109

posterior items because of the prior items if there are no primitives. And they
are right for it is impossible to survey infinitely many items. And if things
come to a stop and there are principles, then these, they say, are unknowable
since there is no demonstration of them and this is the only kind of understanding
there is. . . .
The other party agree about understanding, which, they say, arises only through
demonstration, But they argue that nothing prevents there being demonstrations
of everything; for it is possible for demonstrations to proceed in a circle or
reciprocally.
We assert that not all understanding is demonstrative: rather, in the case of
immediate items understanding is indemonstrable. And it is clear that this must
be so; for if you must understand the items which are prior and from which
the demonstration proceeds, and if things come to a stop at some point, then
these immediates must be indemonstrable. (Posterior Analytics, Book Alpha,
Chapter 3)
In these passages the problem is put in terms of giving "proofs,"
"rational explanations," and "demonstrations." Accordingly, these historical formulations might be criticized for setting the requirements for
knowledge too high. Alternatively, someone might think that the arguments do not threaten knowledge at all, since they are directed at
"understanding," or sdentia, or some other property requiring stronger
conditions than knowledge does.2 But either of these responses would
miss the real force of the skeptical reasoning being considered. The
central theme that drives the regress argument is that positive epistemic
status in general requires being based on good evidence. To put the idea
another way: anything that counts as knowledge (or understanding, or
sdentia) must be believed on the basis of good reasons for thinking that
the thing in question is true. Accordingly, the infinite regress argument
can be reconstructed as follows:
(IR)

1. To know that something is true one must believe it on the basis of

good reasons, or reasons that indicate that the thing is likely to be
true.
2. But not any reason is a good reason. Good reasons themselves must
be backed up by good reasons for thinking that they are true, which
reasons will in turn be in need of further good reasons.

For a position along these lines, see Eleonore Stump, "Aquinas on the Foundations of
Knowledge," Canadian Journal of Philosophy, supplementary volume 17 (1992): 125-158.

110

3. Therefore, to know that something is true one must believe it on

the basis of an infinite number of good reasons. (1,2)
4. No human is capable of basing her belief on an infinite number of
reasons.
5. Therefore, knowledge (for humans) is impossible. (3,4)

When we understand the argument this way it is a powerful one, for it

depends only on premises that are initially plausible and reasoning that
is seemingly valid. Moreover, the argument is directed at knowledge in
the ordinary sense. Premise (1) says that knowledge must be based on
good evidence - a platitude in most circles and true by definition in
some. Premise (2) makes the plausible claim that not anything counts as
a good reason or good evidence. Presumably one must have some good
reason for thinking that one's evidence is true. But since the question
regarding the value of one's evidence can be raised at each level of
supporting reasons, it would seem to follow that an infinite number of
reasons is required to ground a knowledge-claim. Premise (4) points out
that human cognition is incapable of such a feat, and the skeptical
conclusion follows from there straightaway. Something in argument
(IR) must be wrong, but it is not obvious what it is. As we will see,
refuting (IR) requires taking substantive positions regarding the nature
of knowledge and evidence.
It should be clear at this point that (IR) is immune to dismissive
responses to skepticism. Pragmatic and rhetorical responses are irrelevant
here for the same reasons they were irrelevant regarding skepticism
about the world. The skeptical problem represented in (IR) has nothing
to do with how we should speak or act, and therefore cannot be solved
by making pragmatic or rhetorical points. Rather, (IR) takes assumptions
that are initially plausible in the sense that they would be accepted by
nearly anyone pre-theoretically and shows that those assumptions lead
to a total skepticism about human knowledge. The only way to solve
that kind of problem is to find something in the argument to reject.
It is also clear that (IR) does not trade on unrealistically high standards
for knowledge. Nothing like absolute certainty is being required, for the
argument insists only that our evidence makes it likely that our knowledge-claims are true. That is hardly a high standard. For the same reason
it is also clear that deduction is not being required. Sextus and Aristotle
talk in terms of "proof and "demonstration," but nothing in the
skeptical reasoning requires that the argument be put that way. The
requirement that knowledge be based on inductive evidence issues in

111

the regress just as effectively, and so deduction has nothing to do with

it.
Finally, it is worth mentioning that (IR) does not depend on a
modern ontology. Nothing whatsoever in (IR) makes use of one ontology over another, and therefore the idea that skepticism about the world
depends on Cartesianism is refuted by that point alone. The scope of
(IR) is total skepticism rather than skepticism about the world, but the
narrower kind of skepticism follows from the broader. Accordingly, if a
completely general skepticism does not depend on a modern ontology,
then skepticism about the world does not either.
With dismissive responses aside, we are now in a position to consider
non-dismissive challenges to the regress argument. The three most important positions that have been taken in response to (IR) are foundationalism, coherentism, and contextualism. Foundationalism is the position that some things are known even though they are not believed on
the basis of justifying reasons. As such, foundationalism denies premise
(1) of (IR). Foundationalism also denies (2), since within a foundationalist framework "basic" or non-inferred knowledge can also serve as
good evidence for knowledge that is inferred from the foundations.
Coherentism accepts premises (1) and (2) but rejects the reasoning to
(3). The idea is that although all knowledge must be based on good
reasons, there is no implication of an infinite regress since justifying
reasons can be in a relation of mutual support. This relation is called
"coherence," with different versions of coherentism spelling out the
nature of the relation in different ways.
A third response to (IR) is contextualism. The central thesis of contextualism is that the requirement for further good reasons depends on
context. For example, in normal situations a person can know that there
are two sleeping cats in the room merely by seeing each cat separately
and then inferring that there are at least two. What is not required is
that the person have further good reasons for thinking that each individual cat is there it is enough in a normal context just to see them. On
the other hand, we can imagine other contexts where seeing each cat
would not be enough. If one were visiting a taxidermist, for example,
one would need further reasons for thinking she sees individual living
cats rather than individually stuffed cat skins. In different contexts, different beliefs will be contextually basic.
Contextualism therefore heads off an infinite regress of reasons by
allowing a stopping point at beliefs that are contextually basic. But a
question arises whether contextually basic beliefs are themselves cases of

112

knowledge. Contextualists have come down on both sides of the issue.

If the position is that such beliefs are knowledge, then contextualism
denies both premises (1) and (2) of the skeptical argument. If the position
is that these beliefs are not knowledge, then contextualism denies only
premise (2).
Below we will look more closely at each of these various positions.
But at this point I want to emphasize that all three are driven by
consideration of the regress argument. Foundationalists, coherentists, and
contextualists all think that their position is necessary to avoid something
like (IR), confirming that the analysis of skeptical arguments can and
does drive positive epistemological theory.
We have already seen that Aristotle's foundationalism is a direct
response to the problem of an infinite regress. The same is less clear in
Plato, although he seems to follow a similar dialectic, embracing foundationalism rather than admitting that "compounds" are known whereas
their "simple elements" are unknown. In this respect Plato and Aristotle
have been followed by many over the centuries. In the folio wing passage
Chisholm endorses foundationalism as the correct response to the regress
problem:
In many instances the answers to our questions will take the following form:
"What justifies me in thinking that I know a is F is the fact that it is evident to
me that b is G." For example, "What justifies me in thinking I know that he
has that disorder is the fact that it is evident to me that he has those symptoms." . . .
We might try to continue ad infinitum, justifying each new claim that we
elicit by still another claim. Or we might be tempted to complete a vicious
circle: in such a case, having justified "a is F" by appeal to "b is G," and "b is
G" by reference to "c is H," we would then justify "c is FT9 by reference to "a
is F." But if we are rational beings, we will do neither of these things. For we
will find that our Socratic question leads us to a proper stopping place.3
Not only foundationalists are driven by the regress argument. Consider
the following passage from the coherentist Laurence Bonjour.
The main watershed which divides the [coherentist theory of knowledge] from
opposing epistemological views is a familiar problem which I shall call "the
regress problem." This problem arises directly out of the justification condition
of the traditional explication of knowledge as adequately justified true belief.
The most obvious way in which beliefs are justified is inferential justification. In
3 Chisholm, Theory of Knowledge, 2nd ed., pp. 18-19.

113

its most explicit form, inferential justification consists in providing an argument

from one or more other beliefs as premises to the justificandum belief as a
conclusion. But it is obviously a necessary condition for inferential justification
that the beliefs appealed to as premises be themselves already justified in some
fashion; that a belief follows from unjustified premises lends it no justification.
Now the premise-beliefs might also be justified inferentially, but such justification would only introduce further premise-beliefs which would have to be
justified in some way, thus leading apparently to an infinite, vicious regress of
epistemic justification. . . . Any adequate epistemological position must provide
a solution to this problem, a way of avoiding the skeptical result and the
character of that solution will determine, more than anything else, the basic
structure of the position.4

Finally, David Annis contends that his contextualism is an alternative

to foundationalist and coherentist responses to the regress argument.
Philosophers who have accepted foundationalism have generally offered a version of the infinite regress argument in support of it. Two key premises in the
argument are the denial of a coherence theory of justification and the denial
that an infinite sequence of reasons is sufficient to justify a belief. But there is
another option to the conclusion of the argument besides foundationalism. A
contextualist theory of the sort offered above stops the regress and yet does not
require basic statements in the foundationalist's sense.5

These passages show that foundationalism, coherentism, and contextualism are offered quite explicitly as solutions to the regress argument.
As such the literature on this debate offers a clear confirmation that the
analysis of skeptical arguments drives positive epistemology. The dialectic at work here is that theorists assume that there must be some mistake
in the argument. A particular theory is then offered as explaining exactly
where the argument goes wrong. Moreover, it is not uncommon for
philosophers to argue against an alternative solution on the grounds that
it does not really avoid skepticism. In the following passage Goldman
criticizes Bonjour's coherentism for placing unrealistic demands on
knowledge and therefore having unacceptable skeptical results:
It should be clear on reflection that this is a severely unrealistic demand. It is
most implausible to suppose that garden-variety perceptual beliefs and memory
4 Bonjour, "The Coherence Theory of Empirical Knowledge," in Paul Moser, ed., Empirical
Knowledge (Totowa, NJ: Rowman and Littlefield, 1986), p. 117. Reprinted from Philosophical
Studies 30 (1976): 281-312.
5 David Annis, "A Contextualist Theory of Epistemic Justification," in Moser (1986), p. 208.
Reprinted from American Philosophical Quarterly 15 (1978): 213-219.

114

beliefs are based on inferences from premises of the indicated sort. Yet we
commonly impute knowledge and justification in these cases. Not only are very
few beliefs actually based on such inferences, it seems likely that the only people
who possess the relevant premise beliefs (or even possess the constituent concepts) are people with epistemological training and sophistication. It would
therefore follow on Bonjour's view that only these people are deserving subjects
of the terms "knower" and "justified believer." But is it plausible to suggest
that philosophical sophisticates are the only people with knowledge or justified
belief?6
Here the assumption is that skepticism is false, and that therefore any
theory of knowledge that has skeptical consequences, intended or otherwise, is also false.
In this regard Fogelin charges the literature on the regress argument
with blatant question begging. According to Fogelin, it is not appropriate to assume that skepticism is false and then defend one's own position
on that basis. But this, he says, is exactly what so many contemporary
analytic philosophers do:
The third success condition for a theory of epistemic justification is that it not
beg the question against Pyrrhonism by making the argument depend on assuming its falsehood. It is remarkable how often epistemologists do this, quite
explicitly, without a blush. The following specimen comes from Chisholm:
There is the Aristotelian argument to the effect that some of the things I
am justified in believing are self-justifying. The argument is easier to
ridicule than to refute. If my justification for accepting a certain proposition q requires me to go beyond and to appeal to a certain other proposition p, then I'm also justified in accepting q. Therefore these are the
three possibilities; either there is an infinite regress; or there is a circle; or
some propositions I'm justified in believing are self-justifying. But the
first two of these three possibilities are inconsistent with the fact that I do
know something. Therefore some propositions are self-justifying.
The underlying assumption of this passage is that beliefs that we take to be
justified already are justified. The task is to show how. 7
H e remarks,
6

Goldman, "Bonjour's The Structure of Empirical Knowledge," in Bender (1989), p. 108. Several
essays in Bender object to coherentism on the basis of psychological implausibility. See esp.
Hilary Kornblith, "The Unattainability of Coherence," and James Bogen, "Coherentist
Theories of Knowledge Don't Apply to Enough Outside of Science and Don't Give the
Right Results."
7 Fogelin, p. 141. He quotes Chisholm, "Comments and Replies," Philosophia 7 (1978): 597636, at p. 598.

115

This is a feature, expressed in various ways, of a great many contemporary

theories of epistemic justification. All such theories beg the question against
Pyrrhonism, and do so blatantly.8
But if so many epistemologists blatantly beg the question against skepticism, and without even a blush, then maybe Fogelin has missed what
they are up to. My claim has been that charges of question begging are
irrelevant in the context of analyzing skeptical arguments because there
is no skeptic with whom we ought to be engaged in a debate, or against
whom we have to be careful not to beg the question. Rather, the
philosophers that Fogelin is criticizing are trying to give an adequate
account of the nature of knowledge, and a condition of adequacy for
such an account is that it explain our pre-theoretical intuitions about
what does and does not count as knowledge. Since these intuitions are
overwhelmingly non-skeptical, this means that an adequate account
must avoid skepticism. That is why it makes perfect sense to look at the
regress argument and to argue that one's position offers a solution to it,
and why it makes perfect sense to point out that some alternative
account does not.
In other words, what Fogelin is missing is contemporary epistemology's particularism. Philosophers like Chisholm quite explicitly adopt
the methodological rule that epistemic theories ought to be consistent
with what we pre-theoretically take ourselves to know.9 Philosophers
like Goldman quite explicitly add the rule that epistemic theories ought
to be consistent with what we know empirically about our own cognitive capacities. As I have argued, these two methodological strategies are
related, since any theory of knowledge that is psychologically implausible
will generate skeptical results, and will therefore fail to capture our pretheoretical intuition about what we do and do not know.
As we saw in Chapter 1, Fogelin would not admit that a skeptical
position must fail to account for our pre-theoretical intuitions. According to him, our intuitions about what we know and do not know
become unstable under the weight of skeptical reasoning, and a sophisticated skepticism is able to explain this very feature of our pretheoretical judgments. In other words, a sophisticated skepticism carries

8 Fogelin, p. 143.
See Chisholm, "The Problem of the Criterion." See also The Theory of Knowledge, 2nd ed.,
p. 16: "We presuppose,first,that there is something that we know and we adopt the working
hypothesis that what we know is pretty much that which, on reflection, we think we know.
This may seem the wrong place to start. But where else could we start?"

116

with it an explanation of why we think we know in ordinary contexts,

and why we tend to change our minds about this in the extraordinary
contexts in which we take skeptical arguments seriously. But still, it
seems to me that a non-skeptical theory that explains our intuitions
remains overwhelmingly preferable, other things being equal. This is
because a non-skeptical theory accounts for our common sense intuitions about what we know by showing that they are for the most part
true. A skeptical theory accounts for those intuitions only by showing
that they are false, and by adding an explanation about why we do not
normally realize that they are false. Only a non-skeptical theory,
therefore, explains the majority of our intuitions in a sense that preserves
them. But if such a theory is preferable, then the methodological assumption that radical skepticism is false is warranted. For the purposes
of theory construction we should assume that radical skepticism is false,
and we should work out a position that entails this. This methodology
is in fact implicit in the passages that I have quoted. When foundationalists and coherentists argue against other positions, they are not merely
assuming that skepticism is false; rather, they are implicitly claiming that
their own position accounts for our intuitions about knowledge successfully.
So far we have seen how foundationalism, coherentism, and contextualism can be understood as responses to the skeptical regress argument.
Each of these positions assumes that a completely general skepticism is
false, and that therefore something in the argument is wrong. Accordingly, each tries to identify exactly where the argument is mistaken, and
to replace that mistake with a substantive thesis about the nature of
knowledge and evidence. In the remainder of the chapter I argue that
foundationalism provides the best account of knowledge in light of the
regress argument.
More exactly, I argue that foundationalism provides the best account
of human knowledge, or knowledge for beings with our kind of cognition. First, I argue that plausible versions of contextualism are best
understood as a kind of foundationalism and therefore do not represent
an alternative means of stopping the regress. Second, I argue that coherentism does represent an alternative to foundationalism, but that it is
psychologically implausible as an account of knowledge for beings like
us. The conclusion is that a contextualist version of foundationalism
gives the only account of human knowledge that both avoids the regress
argument and is psychologically plausible.
"Contextualist foundationalism" might sound like an oxymoron, but

117

that is because critical treatments of foundationalism have failed to pay

sufficient attention to what that theory is supposed to be about. In other
words, foundationalism is supposed to be a response to the skeptical
argument from an infinite regress ofjustifying reasons. As we have seen,
the way that foundationalism does this is to posit beliefs that are known
even though they are not backed up by further beliefs that act as their
evidence. But nothing about stopping the regress of reasons this way
requires that foundational beliefs be insensitive to context. Therefore,
there is no legitimate reason to burden foundationalism with an absolutist or non-contextualist theory of evidence.
In Section 2, I look at foundationalism more closely and show how
attention to (IR) reveals what is and is not required by a foundationalist
response to the regress argument. In Section 3,1 consider several versions
of contextualism and argue that the plausible ones reduce to foundationalism. In Sections 4 through 7, I look at coherentism. The best-known
objections raised against that position might be called a priori; they charge
that, in principle, coherence is insufficient for justification and knowledge. However, I argue that coherentism falls to a posteriori objections,
or objections to the effect that coherentism is psychologically implausible
as an account of knowledge for beings like us. I end by drawing some
conclusions.
2. FOUNDATIONALISM

The purpose of this section is to sketch the strategy that foundationalism

employs to respond to argument (IR) and to give a preliminary defense
of the position in the face of a few objections. However, my defense of
foundationalism is only preliminary; there are many interesting objections to the position that I do not even consider. This is because my
main concern is to show how attention to skeptical arguments can drive
progress in epistemology. Accordingly, I restrict my attention to a certain class of objections: those that are ineffective against foundationalism
precisely because they fail to attend to the skeptical reasoning to which
foundationalism is supposed to be a response.
The central thesis of foundationalism is that some things are known
even though they are not based on further justifying reasons. A second
thesis is that foundational knowledge can be used as supporting evidence
for other beliefs, thereby giving rise to non-foundational knowledge that
is based on justifying reasons. As we have seen, this means that foundationalism denies premises (1) and (2) of (IR). What I want to do next

118

is look more closely at the various ways that knowledge might be

foundational. Traditionally, foundationalists have embraced four kinds
of foundational knowledge.

a. Four Kinds of Foundation

i. First, some foundationalists have claimed that foundational knowledge may
be based on states of awareness such as sensory or introspective experience.
The idea is that foundational knowledge is based on evidence, but the
evidence is experience rather than belief. The regress of justifying reasons is
thereby ended, because experience is not a "reason" that in turn needs to
be justified or made known by further reasons. This way of understanding
foundational knowledge is sometimes called "givenism," presumably because experience is simply "given" to consciousness, without further need
for justifying grounds.
ii. Some philosophers have held that knowledge is foundational due to the way
in which it is believed. For example, Descartes thought that any proposition
that is clearly and distinctly conceived amounts to certain knowledge. Such
beliefs are sometimes called "self-justifying" or "self-evident," because they
are justified or evident, but not on the basis of any further belief or experience.
iii. Some foundationalists have claimed that beliefs may constitute foundational
knowledge in virtue of their being of a certain intrinsically specified kind.
Thus it has been claimed that beliefs about one's current mental states are
foundational.
iv. Finally, some beliefs have been considered foundational by virtue of their
external relationship to truth. For example, historically it has been claimed
that infallible beliefs qualify as foundational knowledge. More recently it has
been claimed that beliefs that are formed in highly reliable ways are thereby
justified and can even amount to knowledge. If the way in which such a
belief is formed does not involve further beliefs that serve as its evidence,
then the belief would qualify as foundational.
Notice that any of these four strategies for understanding foundational
knowledge would provide an adequate basis for rejecting (IR). Each is
such that, if it is correct, then premises (1) and (2) of (IR) are false. This
is because each describes a way in which a belief can amount to knowledge even though that belief is not based on further reasons acting
as its evidence. But then not all knowledge must be based on further
reasons, and not all reasons are in need of further reasons for their own
justification.

119

b. Objections

There have been many objections to foundationalism, but by far the

most common is that foundational beliefs are impossible. More specifically, a great amount of time and energy has been spent showing that
no beliefs are absolutely certain, or infallible, or incorrigible, or indubitable, or irrevisable. But this kind of objection is easily dismissed if we
understand foundationalism as a response to the regress argument. For
nothing in the foundationalist strategy for stopping the regress requires
that foundational beliefs be absolutely certain, or infallible, or incorrigible, or indubitable, or irrevisable. What is required is that some knowledge not be based on further beliefs that act as its evidence. It is also
required that this foundational knowledge can act as evidence for further
knowledge. But that's it. There is nothing associated with stopping the
regress that requires the high-powered epistemic properties just mentioned and commonly invoked in critiques of foundationalism.10
Why have so many philosophers missed this point? One reason, I
suggest, is that they have misconceived the project of epistemology and
therefore the motivation for foundationalism as well. If the project of
epistemology were to prove to the skeptic that knowledge is possible,
then the requirement that foundational knowledge be infallible, or incorrigible, or indubitable might make sense. The idea would be that we
need premises that not even a skeptic could intelligibly deny, and that
we need to show how the remaining knowledge we have can be based
on those unchallengeable premises. But we have seen that it is a misconceived enterprise to offer proofs to skeptics. Similar things can be said
about engaging skeptics in debate. If this were epistemology's project,
then, again, the quest for absolutely certain foundations might make
sense. But we should not be engaging skeptics in debate any more than
we should be offering them proofs. It is not a misconceived enterprise,
however, to investigate how knowledge must be structured to avoid the
problem of an infinite regress of reasons. This is one important aspect of
constructing an adequate theory of knowledge, which is the real project
of epistemology and the one in which foundationalists, coherentists, and
contextualists alike are engaged.

10 Similar points are made by William Alston in "Has Foundationalism Been Refuted?",
Philosophical Studies 29 (1976): 287-305, and by Laurence Bonjour in "Can Empirical
Knowledge Have a Foundation?", American Philosophical Quarterly 15 (1978): 1-13. See also
Robert Audi, The Structure of Justification (Cambridge: Cambridge University Press, 1993).

120

A second major objection to foundationalism is that there are not

enough foundational beliefs to support all of the knowledge that we
have. Even if some beliefs are infallible, or incorrigible, or indubitable,
and so forth, there are not enough of these to act as evidence for
knowledge that is not in the foundations. The answer to the first objection, however, answers this one as well. Since the foundations of knowledge can be understood much more broadly than both objections assume, there is no reason to think that there will not be enough
foundational beliefs to do the job. Beliefs based on sensory appearances,
for example, will be plentiful if givenism can be made to work.
Of course, such beliefs would not be plentiful if they had to be about
sensory appearances. We have already seen that it is psychologically
implausible that we typically have beliefs about how things appear to us
via the senses. But once we give up requirements like infallibility and
incorrigibility there is no reason to think that beliefs based on sensory
appearances must be about sensory appearances. It is more plausible that
appearances give rise directly to perceptual beliefs about objects in the
world, making fallible beliefs about material objects rather than infallible
beliefs about appearances foundational. This account of perception is
consistent with the theories of evidence that we considered in Chapter
4, and if it is correct, then there would be no shortage of foundational
perceptual knowledge. An account of perception and perceptual evidence along these lines is further developed in Chapters 7 and 9.
3. CONTEXTUALISM
We have noted that the central thesis of contextualism is that the need
for good reasons is context dependent, so that different beliefs are contextually basic in different situations. As such, contextualism denies at
least premise (2) of argument (IR). We also noted that there are two
major kinds of contextualism, depending on whether contextually basic
beliefs are themselves considered to be instances of knowledge. Let us
call the position that basic beliefs are not knowledge "type 1" contextualism, and the position that they are knowledge "type 2" contextualism. Type 2 contextualism denies premise (1) of (IR) as well.
The position that contextually basic beliefs are not knowledge is
suggested by Ludwig Wittgenstein in On Certainty. The idea is that
knowledge and other epistemic concepts are essentially tied to the practice of giving reasons for one's claims, or at least being able to do so.
But then concepts like knowledge and justification are simply out of

121

place where a belief is contextually basic. In such a case there will be

nothing to appeal to as a reason. This interpretation of Wittgenstein is
endorsed by Marie McGinn:
The second criticism Wittgenstein makes of the traditional epistemologist's use
of the words "I know" concerns the connection between knowing and giving
grounds. Our use of the expression "I know" is, he claims, connected 'in
grammar' with the possibility of saying how we know, and hence with the
possibility of stating the grounds on the basis of which our claim to know is
made.11
In support of the position, McGinn quotes Wittgenstein:
One says "I know" when one is ready to give compelling grounds. "I know"
relates to the possibility of demonstrating the truth. (On Certainty, # 243)
Wittgenstein holds that knowledge rests on a foundation, but not an
epistemic foundation. The reason the foundation is not epistemic is that
concepts such as knowledge and justification apply only to beliefs that
are based on reasons or grounds. Or, as Wittgenstein writes,
What we have here is a foundation for all my action. But it seems to me that it
is wrongly expressed by the words "I know." (Ibid., #414)
The WittgensteinMcGinn position has some plausibility as a thesis
about the way certain expressions are used. It is not implausible that
when one uses the expression "I know" one implies that one is ready
to give good grounds. Moreover, epistemic terms such as "reasonable,"
"justified," and "evident" are closely connected to having reasons, giving justifications, and offering evidence. But these verbal points should
not be allowed to mask a more fundamental issue, which is whether
contextually basic beliefs must have some sort of positive epistemic status
to play their grounding role. Whatever we call it, it seems that there
would have to be some feature of contextually basic beliefs that makes
them fit to serve as reasons or evidence or justifications for other beliefs.12
Consider that not just anything counts as a contextually basic belief,
even for Wittgenstein and McGinn. McGinn argues with Wittgenstein
that what makes a belief contextually basic is its special role in the
language game rather than its epistemic certainty.
11

Marie McGinn, Sense and Certainty (Oxford: Blackwell, 1989), p. 109.

12 Sosa makes this point forcefully in Knowledge in Perspective, p. 89.

122

These judgements are absolutely secure, but not in the sense that they are
'intrinsically certain,' as this is understood by the traditional epistemologist. Our
conviction in them is not properly conceived as epistemic certainty regarding
the truth of empirical propositions, for which the question of justification must
inevitably arise, but as the immediate exercise of our practical mastery of our
techniques for describing the world, for which the question of justification
makes no sense.

Whatever the merits of this account of basicality, the relevant point here
is that there must be something that distinguishes beliefs that are appropriately basic in a given context from beliefs that are not; there must be
something that gives some beliefs but not others their epistemic efficacy,
so to speak. But once this point is clearly understood, what is left to the
claim that contextually basic beliefs are not known or epistemically
justified? If the point is just a verbal one about the terms "known" and
"justified," then it does not address the substantive issue at hand. But if
the point is that such beliefs have no positive epistemic status whatsoever, then it is implausible. Perceptual beliefs that describe the world
around me and that are used as evidence for believing other things are
superior in epistemic status to flightful fancies and wild guesses. That is
why they can be used as evidence for believing other things, while
fancies and guesses cannot.
Versions of contextualism that deprive contextually basic beliefs of
positive epistemic status are therefore implausible. Once the issue is
clearly put, it appears that the contextualist should say that basic beliefs
do have positive epistemic status, although their having such depends on
the context in which a given belief arises. This is what I called type 2
contextualism.
This second position is explicitly endorsed by Annis. According to
his version of contextualism, a central issue of justification is whether a
believer is capable of answering relevant objections. Second, justification
is relativized to an "issue-context," which determines both the level of
understanding required of the believer and the persons whose objections
must be answerable. Without going into the details, a belief is contextually basic on this view so long as it evokes no relevant objections. The
idea is that after a point all legitimate objections will be met and the
regress of reasons appropriately ends at that point. But since this stopping
place is partly determined by context, Annis's view is rightly considered
a version of contextualism.13
13 Annis, "A Contextualist Theory."

123

A different version of type 2 contextualism is implicitly endorsed by

Rorty. In a passage where he is arguing against foundationalism Rorty
writes the following:
There is, to be sure, a place for the notion of "direct knowledge." This is
simply knowledge which is had without its possessor having gone through any
conscious inference. But there is no suggestion that some entities are especially
well suited to be known in this way. What we know noninferentiaUy is a matter
of what we are familiar with. Some people (those who sit in front of cloudchambers) are familiar with, and make noninferential reports of, elementary
particles. Others are familiar with diseases of trees, and can report "another case
of Dutch elm disease" without performing any inferences.14
Rorty's contextualism is different from Annis's. It does not centrally
involve the ability to answer relevant objections, but instead trades on a
notion of familiarity that Rorty does not spell out. But clearly Rorty
thinks that some beliefs are contextually basic in the sense in which we
have been understanding that notion: some beliefs are known even
though they are not inferred from other beliefs, and their status as
knowledge depends on contextual features.
Again, I am not interested here in either the details or the merits of
these accounts. Rather, the question I want to raise is why type 2
contextualism is not foundationalism. Both Annis and Rorty present
their views in opposition to foundationalism, but it seems to me that
their views are versions of foundationalism rather than alternatives to it.
To see the point, recall the way that foundationalism was defined.
Since foundationalism is a response to the regress argument, we defined
it in relation to that argument, saying that the central theses of the
position are the denial of premises (1) and (2) of (IR). More explicitly,
foundationalism is the position that some things are known not on the
basis of inferences from further reasons, and that such foundational
knowledge can act as reasons that allow other things to be known. I
propose that the present brand of contextualism answers the regress
argument in exactly the same way and therefore constitutes a version of
foundationalism. Contextualist versions of foundationalism simply add
the thesis that whether a belief is foundational depends on contextual
features.15
Critics of foundationalism might accuse me of abusing terminology;
they might insist that foundationalism has never been understood so that

15

14 Philosophy and the Mirror of Nature, p. 106.

For a similar assessment of Rorty, see Sosa, Knowledge in Perspective, p. 93.

124

foundational beliefs can be determined by context. Thus Michael Williams writes,

For foundationalism, a judgement derives its epistemological status from some
highly abstract feature of the content of the proposition it contains. . . . Given
foundationalism, we might as well talk about basic propositions as about basic
judgements (expressed by propositions advanced in particular circumstances).
But if Wittgenstein is right, the distinction is crucial: a terminating judgement
derives its status from its context, and not just from its content. . . . Whatever
the merits of this response to the regress, it is nothing like traditional foundationalism.
The contrast we have just noted provides the first hint of what I shall claim
to be foundationalism's decisive feature, its commitment to a strongly realistic
conception of epistemic relations. For the foundationalist, there are relations of
epistemic priority that hold between propositions independently of the circumstances in which those propositions are advanced, the interests that govern their
assessment, or any other such "contextual" factors.16
But as I have already argued, there is no legitimate motivation for
burdening foundationalism with theses that are irrelevant to stopping the
regress of reasons. In Williams' terminology, there is no good motivation
for saddling foundationalism with epistemic realism. There is simply
nothing in the logic of the foundationalist strategy for stopping the
regress that requires a "realist" position.17
Annis seems to recognize this point when he defines foundationalism
at the beginning of his essay:
Foundationalism is the theory that every empirical statement which is justified
ultimately must derive at least some of its justification from a special class of
basic statements which have at least some degree of justification independent of
the support such statements may derive from other statements. Such minimal
foundationalism does not require certainty or incorrigibility; it does not deny
the revisability of all statements, and it allows an important role for intrasystematic justification or coherence.18
But for some reason Annis fails to recognize that, by his own definition,
his contextualism is a version of foundationalism.
16 Williams, Unnatural Doubts, p. 67.
17 Williams makes a distinction between formal and substantive foundationalism, where only
the latter is committed to epistemic realism. However, he does not use the distinction to
make the present point; rather, his discussion of the relationship between foundationalism
and skepticism focuses on substantive foundationalism. See Unnatural Doubts, esp. pp. 114
121.
18 Annis, "A Contextualist Theory," p. 203.

125

I now want to take the point a step further. Specifically, I want to

challenge the notion that there has ever been a foundationalist who was
not also a contextualist. I take Descartes as my test case, since he is
usually the whipping boy for critics of foundationalism. In the Meditations Descartes claims that only what is certain amounts to knowledge.
Some things are known for certain immediately, insofar as they are
clearly and distinctly conceived by the light of reason, whereas others
must be demonstrated from what it known immediately. We therefore
have a classic foundationalist structure, with non-inferential knowledge
acting as a basis for inferential knowledge. But Descartes says quite
explicitly that whether something is conceived clearly and distinctly
depends on context, as does whether something is known immediately.
And although amongst the matters which I conceive of in this way, some indeed
are manifestly obvious to all, while others only manifest themselves to those
who consider them closely and examine them attentively; still, after they have
once been discovered, the latter are not esteemed as any less certain than the
former. . . . And as regards God, if my mind were not pre-occupied with prejudices, and if my thought did not find itself on all hands diverted by the
continual pressure of sensible things, there would be nothing which I could
know more immediately and more easily than Him. (Meditation V, Meditations,
p. 183)
Descartes thinks that whether something is known immediately depends
on contextual features such as prior training, degree of attention applied,
the influence of prejudices, and the presence of distractions. What this
shows is that the real issue is not whether there are contextual factors
involved in the determination of basic beliefs, but which contextual
factors are involved. Different versions of contextualism try to give an
adequate account of exactly this issue. Moreover, no version of foundationalism ignores context entirely, not even the extreme rationalism of
Descartes. We may conclude, therefore, that type 2 contextualism reduces to foundationalism.

4. COHERENTISM

The alternative to foundationalism is not contextualism but coherentism.

According to that position, the regress ofjustifying reasons is not infinite
because reasons can stand in a relationship of mutual support or "coherence." Accordingly, coherentism accepts premises (1) and (2) of argument (IR) but it rejects the move to (3); even though all knowl-

126

edge must be supported by justifying reasons, no infinite regress is

entailed.
Coherence theories may be divided into two kinds, according to how
a given theory fills in the details regarding the coherence relation. Traditional coherence theories understand that relation in terms of the usual
inference relations recognized by logicians and other philosophers, foundationalists included. Thus coherence is defined in terms of deductive
and inductive inference relations, including, for example, inferences to
the best explanation. A leading advocate of this kind of theory has been
Bonjour.19 Nontraditional coherence theories develop some alternative
account of coherence. In recent years examples of this kind of theory
have been provided by Keith Lehrer.20
The two best-known objections to coherentism are (a) that highly
coherent belief systems can nevertheless be isolated from reality and (b)
that there are a multiplicity of equally coherent systems with no nonarbitrary means of choosing among them. These may be called a priori
objections to coherentism, since they are objections to the effect that, in
principle, coherence relations cannot give rise to knowledge or justified
belief. Coherentists have been much occupied with objections of this
kind. However, I will not be concerned with them here. Rather, I want
to argue that coherentism falls to a posteriori objections, or objections
that coherentism cannot give an adequate account of knowledge and
justified belief for beings with our cognition.
I argue that the coherentist position in general is threatened by the
following line of objection: Traditional coherence theories hold that
inference always figures into the justification of belief, and so such
theories must claim (implausibly) that even perceptual beliefs are inferentially justified. Nontraditional theories hold that traditional inferences
never play a role in justification, and so must deny (implausibly) that
making a correct inference ever makes a difference. Finally, no kind of
coherence theory can give an adequate account of the role of sensory
appearances in perceptual knowledge. Because coherence theories allow
only other beliefs to serve as evidential grounds, coherence theories must
misdescribe the way that appearances ground perceptual knowledge.

19 See Bonjour, "The Coherence Theory of Empirical Knowledge," and The Structure of
Empirical Knowledge (Cambridge, MA: Harvard University Press, 1985).
20 See, e.g., Keith Lehrer, "The Coherence Theory of Knowledge," Philosophical Topics 14
(1986): 525, and Lehrer, "Coherence and the Truth Connection: A Reply to My Critics,"
in Bender (1989).

127

According to this line of objection, there may be nothing wrong with

coherence theories in principle. Rather, coherence theories fail to provide an empirically adequate account of human knowledge. Some human knowledge is inferential and some is not, and human perceptual
knowledge is grounded directly on sensory appearances rather than
belief.

5. THE OBJECTION TO TRADITIONAL COHERENCE

THEORIES

In the following sections we will be pursuing the argument that coherentism is psychologically implausible. But since different kinds of coherentism are implausible in different ways, it will be necessary to consider
traditional and nontraditional theories separately.
The following passage from Bonjour indicates that he conceives of
the coherence relation in terms of traditionally accepted inference relations.
What, then, is coherence? Intuitively, coherence is a matter of how well a body
of beliefs "hangs together": how well its component beliefs fit together, agree
or dovetail with each other, so as to produce an organized, tightly structured
system of beliefs, rather than either a helter-skelter collection or a set of conflicting sub-systems. It is reasonably clear that this "hanging together" depends on
the various sorts of inferential, evidential, and explanatory relations which obtain
among the various members of the system of beliefs, and especially on the more
holistic and systematic of these. Thus various detailed investigations by philosophers and logicians of such topics as explanation, confirmation, probability, and
so on, may reasonably be taken to provide some of the ingredients for a general
account of coherence.21

Bonjour's view is therefore an example of a traditional coherence theory. The objection I want to raise against this kind of theory is summed
up by John Pollock's observation that perception is not inference. The
problem for such theories is that at least some of our beliefs are both
non-inferential and justified, but according to traditional coherence theories all beliefs having positive epistemic status get it by being inferentially supported by other beliefs. In the following passage Pollock
considers the proposal that his perceptual belief that a book on his table
is red depends for its justification on an inference from reasons:
21

Bonjour, The Structure of Empirical Knowledge, p. 93.

128

The trouble with this proposal is that there are no plausible candidates for such
a reason. What could such a reason be? One suggestion I have heard is that our
reason is the second-order belief that we believe the book to be red, but the
claim that we ordinarily have such second-order beliefs is no more plausible
than the foundationalist claim that we ordinarily have appearance beliefs. Furthermore, what could our reason be for the second-order belief? . . . The general difficulty is that perception is not inference. When I believe on the basis of
perception that the book is red, I do not infer that belief from something else I
believe. Perception is a causal process that inputs beliefs into our doxastic system
without their being inferred from or justified on the basis of other beliefs we
already have.22
Bonjour's strategy for avoiding this kind of objection is to show just
h o w inferential justification for perceptual beliefs proceeds. Bonjour asks
what sort of inferential justification might be available for the perceptual
belief that there is a red book on the table:
First, the belief in question is a visual belief, i.e. it is produced by my sense of
sight; and I am, or at least can be, introspectively aware of this fact. Second, the
conditions of observation are of a specifiable sort: the lighting is good, my eyes
are functioning normally, and there are no interfering circumstances; and again,
I know or can know these facts about the conditions, via other observations and
introspections. Finally, it is a true law about me (and indeed about a large class
of relevantly similar observers) that my spontaneous visual beliefs in such conditions about that sort of subject matter (viz., medium-sized physical objects)
are highly reliable, i.e. very likely to be true; and, once more, I know this law. 23
These observations suggest to Bonjour the following general account of
inferential justification for perceptual beliefs:
(i) I have a spontaneous belief that P (about subject-matter S) which is an
instance of kind K.
(ii) Spontaneous beliefs about S which are instances of K are very likely to
be true, if conditions C are satisfied.
(iii) Conditions C are satisfied.
Therefore, my belief that P is (probably) true.
Therefore, (probably) P. 24
Several comments are in order. First, it would seem that to be justified
in believing P via premises (i) through (iii), one must be justified in
22 Pollock, Contemporary Theories of Knowledge, p. 75.
23 Bonjour The Structure of Empirical Knowledge, p. 291.

24

Ibid., p. 295.

129

believing each of the three premises. It is not enough, as Bonjour

suggests, that the person "can be" introspectively aware, or that she "can
know" that the premises obtain. One reason is that evidence can do me
good only if I use it, and I cannot use it unless I am already aware of it.
Another reason is that evidence-beliefs cannot give rise to justification
unless they enjoy positive epistemic status themselves. For example, if I
am just guessing that the postcard in my mailbox is from you, then that
hardly allows me to know that you are in Paris. This is so even if I "can
know" that it is from you just by looking more closely.
Second, premises cannot give rise to justification unless the person
actually makes the relevant inference; the mere availability of a justifying
inference is not enough to give a belief positive epistemic status. If I
know how many apples and how many oranges are in the refrigerator,
but I have not put two and two together, then I do not know how
many fruits are in there. Third, it is extremely implausible that we are
justified in believing the relevant premises or that we make the relevant
inference every time we are justified in some perceptual belief. As we
have already noted, typically we believe no such premises and make no
such inference.
These considerations are sufficient to illustrate the problem inherent
in any theory that conceives coherence in terms of traditional inference
relations. Any such theory must give an implausible account of perceptual knowledge, since it is implausible that perceptual beliefs are inferred
from premises about sensory appearances, or from premises about the
conditions in which perceptual beliefs are formed, or from anything else.
6. THE OBJECTION TO NONTRADITIONAL COHERENCE
THEORIES

An example of a nontraditional coherence theory is provided by Lehrer's

position in "The Coherence Theory of Knowledge." According to
Lehrer, knowledge depends on "complete justification," which in turn
involves both an objective and a subjective element. Each of these
elements is understood in terms of coherence with a certain system of
beliefs. More specifically, S is personally (subjectively) justified in accepting that p if and only if p coheres with the acceptance system of S.
S is verifically (objectively) justified in accepting that p if and only if p
coheres with the verific system of S.25 S's acceptance system corresponds
25 Lehrer, "The Coherence Theory of Knowledge," p. 6.

130

roughly to those propositions that S accepts for the purposes of obtaining

truth and eschewing error. S's verific system is obtained by deleting
falsehoods from S's acceptance system.
Next Lehrer defines the key idea of coherence in terms of the undefined locution "it is more reasonable for S to accept that p on the
assumption that c than to accept q on the assumption that d on the basis
of system A." The central intuitive idea is that "something coheres with
a system of a person if it is more reasonable to accept it than to accept
anything with which it conflicts on the basis of the system."26 This
intuitive idea is made explicit by means of the following definitions:
p competes with q for S on the basis of system A if and only if it is more
reasonable for S to accept p on the assumption that not q than on the assumption
that q on the basis of system A.
p beats q for S on the basis of system A if and only if p competes with q for S
on the basis of system A and it is more reasonable for S to accept p than to
accept q on the basis of system A.
n neutralizes q as a competitor of p for S on the basis of system A if and only if
q competes with p for S on the basis of system A, and n is such that the
conjunction of q and n does not compete with p for S on the basis of system A
when it is as reasonable for S to accept the conjunction of q and n as to accept
q alone on the basis of system A.
Lehrer then defines the coherence relation as follows:
p coheres with the system A of S if and only if, for every q that competes with
p for S on the basis of the system A, q is either beaten or neutralized for S on
the basis of the system A.27
It is clear that on Lehrer's account coherence is not understood in terms
of the inference relations commonly recognized by traditional theories
of knowledge. In place of the familiar inference relations of deduction
and induction Lehrer defines coherence in terms of his technical notions
of competition and neutralization. Accordingly, Lehrer's theory is what
I have called a nontraditional coherence theory.
The objection I want to raise against nontraditional theories like
Lehrer's is that they cannot account for inferential knowledge. More
exactly, we know some things because we infer them from other things
we know. But nontraditional coherence theories do not recognize a
justificatory role for traditional inferences.
26

Ibid., p. 9.

27

131

Ibid., pp. 10-11.

An example from Wayne Davis and John Bender will help to illustrate the problem:
Consider an acceptance system containing the following propositions.
(1)
(2)
(3)
(4)
(5)

All and only those with at least a 90 on a prelim get an A on it.

All and only those with an A on every prelim are exempt from the final.
I got at least a 93 on every prelim.
I got an A on every prelim.
I am exempt from the final.

On Lehrer's theory, any subject with such an acceptance system is justified in

accepting (5), as long as the propositions are all true for that subject. For the
probability of (5) relative to the rest of the system is 1. Any such subject has a
conclusive reason to accept (5). But compare Alan and Bob, for both of whom
all the propositions are true. Alan inferred (4) from (1) and (3), and he inferred
(5) from (2) and (4). Bob inferred (5) from (1) and (4), (1) from (2), (2) from
(3) and (4), and (4) from (5).28
Davis and Bender argue that on Lehrer's theory both Alan and Bob
come out justified, even though Bob's reasoning is viciously circular and
idiotic. But this is counterintuitive a person's belief cannot be justified
on the basis of circular and idiotic reasoning. H o w the person arrives at
his belief matters for justification and knowledge.
Davis and Bender argue that Lehrer is exposed to the counterexample because he does not distinguish between merely having reasons for a belief and believing for those reasons. But I think that the
problem is deeper than that. To see this, notice that we can revise
Lehrer's theory to accommodate the distinction between having reasons
and believing for reasons. For S to be personally justified in accepting
p, we might require not only that p coheres with S's acceptance system
in Lehrer's sense, but also that S is aware of this and accepts p for that
reason.
If we revise the theory in this way we will now get a correct ruling
on Bob. Since Bob does not believe (5) because (5) coheres with his
other beliefs, Bob's belief will not be ruled justified. However, the
revised version now rules incorrectly about Alan. For although Alan
clearly is justified in accepting (5), he does not accept (5) because he is
sensitive to the Lehrerian coherence relation between (5) and the remainder of his beliefs. This consideration presents a problem for any
28 Wayne Davis and John Bender, "Fundamental Troubles with the Coherence Theory," in
Bender (1989), p. 58.

132

nontraditional coherence theory. Namely, Alan has justified belief in the

example at least partly because he has made certain traditional inferences,
but a nontraditional coherence theory does not recognize a justificatory
role for these inferences, and that by its very nature.
Before accepting this last conclusion we should consider a final strategy by Lehrer to account for the role of traditional inferences in justification. In his latest defense of the coherence theory Lehrer argues that
in cases of traditional inferential justification, coherence is impossible
unless a person accepts that her belief follows (in a traditional way) from
her evidence for her belief. Thus Lehrer establishes a role for traditional
inferences by requiring that a person accepts that there is a traditional
inference relation between her belief and the evidence she has for her
belief.29
But in fact this latest move does not avoid the current objection
against coherentism. Waving aside the point that people typically accept no such thing about their beliefs, the current objection is that in
certain cases what matters for justification is that a traditional inference
has been made. What matters for justification is not that a person believes or accepts that her belief follows from her evidence, but that she
has actually made an inference from her evidence to the belief in question. Thus I take it that the above example from Davis and Bender
would not lose its force if we added that Bob accepts that (5) follows
from (1) through (4). So long as no good inference was actually made,
Bob is not justified in believing that he is exempt from the final, and
the fact that Bob thinks that a good inference could be made seems
irrelevant.
Consider a second example. Suppose that Mary is a geometry student
who accepts certain axioms and accepts a theorem that in fact follows
from those axioms. We may stipulate further that Mary accepts that the
theorem follows from the axioms and that she accepts that she can tell
whether this is so. Finally, let us suppose that all of these beliefs are true.
But now suppose that Mary has not in fact made a correct inference
from the axioms to the theorem. Atypically, Mary has reasoned fallaciously in this case. It is clear that Mary does not have either justification
or knowledge regarding the theorem. Furthermore, the reason she lacks
these is that she has failed to make a correct inference. The example
shows that in certain cases whether a person is justified in believing
something depends at least partly on whether she has made a traditional
29 Lehrer, "Coherence and the Truth Connection," p. 272.

133

inference. It is this fact about human knowledge that non-traditional

coherence theories cannot explain.
The considerations of Sections 5 and 6 show that traditional coherence theories cannot account for perceptual knowledge and nontraditional theories cannot account for inferential knowledge. This is because traditional theories must hold that traditional inferences are
always involved in knowledge, while nontraditional theories must hold
that traditional inferences are never involved. In the next section I want
to consider an objection that can be raised against all coherence theories. It is another a posteriori objection, claiming that coherence theories
cannot account for the role of sensory appearances in perceptual
knowledge.
7. THE ROLE OF SENSORY APPEARANCES
The essential point of this objection is that positive epistemic status
depends on what sensory experience a person actually has. Just as inferential knowledge depends on what inferences are actually made, perceptual knowledge depends on how things actually appear. But since the
way things appear is not itself a belief, coherence theories cannot account
for the importance of sensory appearances in perceptual knowledge.
To focus on the objection I have in mind we may consider a counterexample to coherentism formulated by Ernest Sosa. Sosa asks us to
consider person A, who has the belief that he has a headache, in a case
where A does indeed have a headache. Sosa correctly notes that the
belief in question will have few relevant relations with other beliefs that
A currently holds. These beliefs might include that he is not free of a
headache, that he is in pain, and perhaps some few others. But now
consider the case of person B, who also has a splitting headache, but
who has exactly the same beliefs as A save the following few changes:
change the belief that B has a headache to the belief that he does not
have a headache, change the belief that B is not free of a headache to
the belief that he is free of a headache, and make the other few changes
necessary to bring into coherence all of B's beliefs relevantly related to
the existence or nonexistence of a current headache. Sosa's point is that
although A is justified in believing that he has a headache, B is not
justified in believing that he does not have a headache. And this is so
despite the fact that B's beliefs enjoy the same coherence as A's. The
reason is that B's having a splitting headache affects the epistemic status
of B's beliefs about his headaches, even though his having a headache is

134

neither a belief nor a relation among beliefs and so does not affect the
coherence of B's beliefs.30
Sosa's example involves an introspective belief about one's own mental state, but a similar example can be constructed for sensory perception.
For example, visual beliefs are typically involved in few relevant relations
with other beliefs. In the case where I have a perceptual belief that there
is a tree before me, an equally coherent system can easily be constructed
for the belief that there is a stump before me. And now Sosa's reasoning
can be applied to these perceptual cases. In the first case, part of what
justifies me in believing that there is a tree is surely my experience that
there seems to be a tree. In the second case, I am not justified in
believing that there is a stump, despite the fact that, by hypothesis, this
new belief fits into an equally coherent system for my experience is
that there seems to be a tree and not a stump.
Notice that this kind of objection is effective against coherentism
whether we are thinking of appearances as thin or thick. On the one
hand, it seems that I cannot be justified in my belief that there is a stump
before me if things appear phenomenally as if there were a tree rather
than a stump. But even more clearly, I cannot be justified in believing
that there is a stump before me if the representational content of my
sensory experience is that the thing is a tree. Either way we think of
appearances, the justification of perceptual beliefs depends on the way
things appear, and not merely on our having beliefs about the way
things appear.
As I have already noted, the present objection mirrors the main
objection brought against nontraditional coherence theories; just as what
inferences one makes counts in determining whether an inferential belief
is justified, what experiences one has counts in determining whether a
perceptual belief is justified. But then, contra coherentism, the justification of perceptual beliefs is not reducible to beliefs or relations among
beliefs.

8. CONCLUSIONS

We may conclude that coherence theories cannot give an account of

human knowledge that is empirically adequate. Some of our knowledge
30

Ernest Sosa, "The Raft and the Pyramid: Coherence versus Foundations in the Theory of
Knowledge," in Knowledge in Perspective, pp. 184185. Reprinted from Midwest Studies in
Philosophy 5 (1980): 3 - 2 5 .

135

is non-inferential and some is inferred from what is already known.

Moreover, perceptual knowledge is based directly on sensory appearances as its evidential grounds. This suggests that foundationalism gives
the correct account of the structure of human knowledge, since it is the
only account that both answers the skeptical regress argument and is
psychologically plausible. We may also note that these conclusions confirm the lessons of Chapter 4: that not all evidential relations are inferential, and that perceptual knowledge is not inferred from beliefs about
the way things appear.
Finally, the kind of foundationalism here affirmed is contextualist.
This is because how things are perceived and what inferences are made
both depend on context. This is trivially so in that the causal antecedents
of perception and inference are dependent on contextual factors. But
the details of a correct view will not be trivial. Substantive issues arise
regarding which contextual factors influence human cognition, and how.
For example, substantive issues arise regarding how class, race, gender,
and other social factors affect human cognition. Different versions of
contextualism decide these questions differently. No epistemology that I
am aware of, however, has claimed that context is irrelevant to human
knowing.

136

Hume's Skepticism about

Unobserved Matters of Fact

In the preceding chapters we have considered arguments for skepticism

about the world from Descartes and Hume. We also looked at the
Pyrrhonian infinite regress argument, which is universal in its scope. A
third kind of skepticism regards our knowledge of unobserved matters
of fact. Arguments for this position have their most famous articulation
in Hume and charge that we can never make adequate inferences from
the observed to the unobserved. Skeptical arguments of this kind have
sometimes been called "the problem of induction."
Various objections have been raised against Hume's reasoning here.
The three that I will be concerned with are (a) that Hume's arguments
depend on his empiricist theory of ideas, (b) that the arguments assume
that absolute certainty is required for knowledge, and (c) that Hume's
arguments assume deductivism, or the position that inductive reasoning
cannot give rise to knowledge.
If any of these objections were effective, then Hume's skeptical arguments would teach no important lesson. No one today thinks that
Hume's theory of ideas is adequate, or that knowledge requires either
absolute certainty or deduction. But in fact none of these objections is
effective against Hume's skeptical reasoning. This is because Hume's
reasoning does not essentially depend on the various implausible assumptions that the objections attribute to it. As we have seen in other cases,
the skeptical reasoning in question can be reconstructed so as to employ
only assumptions that are pre-theoretically plausible, and that would be
accepted by nearly anyone outside the context of philosophical inquiry.
My strategy for defending Hume's reasoning against the present objections is to distinguish between two skeptical arguments in the Enquiry
one from Section IV and one from Section VII. With the arguments

137

clearly distinguished we can see that objection (a) is effective only against
the argument from Section VII. Objections (b) and (c) are effective
against neither argument.
In Part I of this of chapter, I reconstruct the two skeptical arguments
from Hume and evaluate the force of objections (a) and (b). I consider
these objections to be type-b dismissive responses, since they engage
Hume's actual reasoning only superficially. In Part II I consider objection (c), which we will see is not a dismissive response. However, I
think that this very common objection to Hume ultimately misunderstands his skeptical reasoning about unobserved matters of fact. Hume is
not a skeptic about the epistemic efficacy of inductive inferences.
Rather, he is skeptical that our beliefs about unobserved matters of fact
have even inductive support. Understanding Hume's argument this way
makes it far more powerful than it is usually understood to be, for almost
everyone thinks that inferences must be at least inductive to give rise to
knowledge. I argue that such is not the case, and that this is the real
lesson of Hume's argument. This in turn poses a problem for positive
epistemology: any adequate account of knowledge must explain how
evidence can produce knowledge even if that evidence does not inductively support its conclusions. I end by pointing to a theory of knowledge that would do the job. Chapter 7 develops this theory with an eye
on the lessons from the preceding chapters as well. In the next chapter,
therefore, we make the transition to positive epistemology.

I. Hume's Arguments and Dismissive Responses

We may distinguish two Humean arguments for skepticism about unobserved matters of fact. Hume runs the two arguments together in A
Treatise of Human Nature, and as a result commentators often fail to
distinguish them. Any confusion in the Treatise is nicely resolved in the
Enquiry, however, where Hume presents the two arguments in different
sections. By getting clear on the structures of these two arguments we
will be in a better position to evaluate the three objections that I have
just noted.
1. THE ARGUMENTS IN SECTIONS IV AND VII OF THE
ENQUIRY

The first argument I want to look at is from Section IV of the Enquiry.

It will be remembered that we have already seen the argument in

138

Chapter 2, where it was portrayed as an argument for skepticism about

the future. But the real point of the argument is that there is no knowledge of any matter of fact that is not observed. Future matters of fact are
just one example of unobserved matters of fact.
As we saw in Chapter 2, the argument centers on the claim that there
is a crucial assumption in all of our reasoning concerning what is not
observed. Namely, all reasoning about unobserved matters of fact involves the "regularity principle," or the principle that unobserved cases
will be similar to observed cases. Another way to state the principle is to
say that nature will continue to be regular. Hume argues that the regularity principle cannot be supported, and that therefore all of our beliefs
concerning unobserved matters of fact rest on an unsupported supposition. Here is how Hume states the argument in the Enquiry.
All reasonings may be divided into two kinds, namely, demonstrative reasoning, or that concerning relations of ideas, and moral reasoning, or that concerning matter of fact and existence. That there are no demonstrative arguments in
the case seems evident: since it implies no contradiction that the course of
nature may change, and that an object, seemingly like those which we have
experienced, may be attended with different or contrary effects. . . .
If we be, therefore, engaged by arguments to put trust in past experience,
and make it the standard of our future judgement, these arguments must be
probable only, or such as regard matter of fact and real existence, according to
the division mentioned. But that there is no argument of this kind, must appear,
if our explication of that species of reasoning be admitted as solid and satisfactory. We have said that all arguments concerning existence are founded on the
relation of cause and effect; that our knowledge of that relation is derived
entirely from experience; and that all our experimental conclusions proceed
upon the supposition that the future will be conformable to the past. To
endeavor, therefore, the proof of this last supposition by probable arguments, or
arguments regarding existence, must be evidently going in a circle, and taking
that for granted, which is the very point in question, (pp. 35-36)
In this passage Hume argues that neither demonstrative nor moral reasoning can justify the regularity principle. But his intended conclusion is
that the principle cannot be justified at all. Hume's suppressed premise
here is that the regularity principle must be justified by reasoning if it is
to be justified at all. I take it that Hume does not make the premise
explicit because he thinks that it is obvious. The only other sources of
justification besides reasoning for Hume are present sense, intuition, and
memory. Since it is clear that these are inadequate for the justification
of the principle, Hume explicitly considers only moral and demonstra-

139

tive reasoning. Here is a reconstruction of Hume's wider argument,

with the suppressed premise in brackets.
(UMF1)

1. All reasoning is either demonstrative reasoning or moral reasoning.

2. The principle that unobserved cases will be like observed cases
cannot be justified by demonstrative reasoning, since its denial
is not a contradiction. It does not merely state a relation of
ideas.
3. The principle cannot be justified by moral reasoning, since
moral reasoning presupposes the very thing in question, and so
such a justification would be circular.
[4. Neither can it be justified by intuition, present sense, or memory, since none of these is relevant to a factual supposition
about unobserved cases.]
5. Therefore, the supposition that unobserved cases will resemble
observed cases cannot be justified at all. (1,2,3,4)
6. All of our beliefs about unobserved matters of fact depend on
that supposition for their justification.
7. Therefore, all beliefs about unobserved matters of fact are
themselves unjustified, since they depend on an unjustified
supposition. (5,6)

Hume's second argument appears in Section VII. In this section

Hume considers the proper understanding of our ideas of necessary
connection and cause and effect. He concludes that the idea of cause
and effect contains the idea of necessary connection, so that when we
say that X is the cause of Y, part of what we mean is that there is a
necessary connection between X and Y. But the only idea of necessary
connection that we have is that of (a) two objects (or events) constantly
appearing together in experience, together with (b) an expectation in
the mind that the two objects (or events) will continue to go together
in the future:
The first time a man saw the communication of motion by impulse, as by the
shock of two billiard balls, he could not pronounce that the one event was
connected: but only that it was conjoined with the other. After he has observed
several instances of this nature, he then pronounces them to be connected. What
alteration has happened to give rise to this new idea of connexion} Nothing but
that he now feels these events to be connected in his imagination, and can readily
foretell the existence of one from the appearance of the other. When we say,
therefore, that one object is connected with another, we mean only that they
have acquired a connexion in our thought, and give rise to this inference, by

140

which they become proofs of each other's existence: A conclusion which is

somewhat extraordinary, but which seems founded on sufficient evidence.
(PP- 75-6)
Hume does not say so explicitly in Section VII, but it is easy to see how
this analysis of cause and effect has skeptical consequences. For if this is
all that is contained in the idea that X is the cause of Y, then we cannot
know that in the future X's occurring will result in Y's occurring. This
is because the fact that two things have gone together in the past,
together with the fact that we expect them to go together in the future,
does not guarantee that the things will go together in the future. Here is
a loose outline of the argument in Section VII, extended so as to
explicitly include the skeptical conclusion:
(UMF2)

1. All beliefs about unobserved matters of fact depend for their

justification on inferences involving the relation of cause and
effect.
2. The only idea we have of this relation is that of two objects
being constantly conjoined in the past, together with an expectation in the mind that the two objects will continue to go
together in the future.
3. There is no argument that objects conjoined in the past and
expected to go together in the future will go together in the
future.
4. Therefore there is no rational basis for such an inference. (1,2,3)
5. Therefore no knowledge can be acquired via such an inference.
(4)
6. Therefore there can be no knowledge of unobserved matters
of fact. (1,5)

2. OBJECTION (a): HUME RELIES ON AN INADEQUATE

THEORY OF IDEAS

Let us understand Hume's empiricist theory of ideas as involving two

main theses: (i) that ideas and beliefs can be analyzed in terms of sensations and (ii) that sensations are to be understood as essentially unrelated,
simple particulars. We may call the first thesis "empirical reductionism"
and the second thesis "empirical atomism." The present objection
charges that Hume's skeptical arguments regarding unobserved matters
of fact depend on this empiricist theory of ideas. But since this theory is
false or otherwise inadequate, Hume's reasoning is without force. This
kind of objection to Hume is associated with Dewey:

141

The philosophic empiricism initiated by Locke was thus disintegrative in intent.

It optimistically took it for granted that when the burden of blind custom,
imposed authority, and accidental associations was removed, progress in science
and social organization would spontaneously take place. . . . But after Hume
with debonair clarity pointed out that the analysis of beliefs into sensations and
associations left "natural" ideas and intuitions in the same position in which the
reformers had placed "artificial" ones, the situation changed. The rationalists
employed the logic of sensationalistic-empiricism to show that experience, giving only a heap of chaotic and isolated particulars, is as fatal to science and to
moral laws and obligations as to obnoxious institutions; and concluded that
"Reason" must be resorted to if experience was to be furnished with any
binding and connecting principles. The new rationalistic idealism of Kant and
his successors seemed to be necessitated by the totally destructive results of the
new empirical philosophy.1
In this passage Dewey is blaming the empiricist assumptions of Locke
and Hume for "destructive" skeptical results. But it is not clear which
aspect of modern empiricism is to blame. At the start of the passage
Dewey picks out "the analysis of beliefs into sensations and associations"
as the culprit. But by the end of the passage the problem is the conception of experience as a "heap of chaotic and isolated particulars." So it
is not clear whether Dewey thinks reductionism or atomism is the root
of skepticism. A few pages later Dewey emphasizes the latter diagnosis.
When experience is aligned with the life-process and sensations are seen to be
points of readjustment, the alleged atomism of sensations totally disappears. With
this disappearance is abolished the need for a synthetic faculty of super-empirical
reason to connect them. Philosophy is not any longer confronted with the
hopeless problem of finding a way in which separate grains of sand may be
woven into a strong and coherent rope or into the illusion and pretense of
one.2
I will pause to note that the two passages from Dewey are inconsistent.
The first implies that empirical reductionism is sufficient for skepticism,
and the second implies that empirical atomism is necessary for skepticism. But on the assumption that reductionism does not imply atomism
(which it does not), the two diagnoses of skepticism contradict each
other.
Waving this aside, the more important point is that both of Dewey's
objections are ineffective against Hume's skeptical reasoning. This is
1

John Dewey, Reconstruction in Philosophy (Boston: Beacon Press, 1957), pp. 8283.
2 Ibid., p. 90.

142

because the argument from Section IV presupposes neither an atomistic

conception of experience nor a reductionist account of belief. Both of
these components of the empiricist theory of ideas are inessential to
Hume's reasoning in argument (UMF1). That is not to say that Hume
did not accept both atomism and reductionism. Hume did accept these,
and his arguments are largely couched in the vocabulary of his empiricist
theory of ideas. The present point is that the argument from Section IV
need not be couched in such terms. Accordingly, Dewey's critiques
focus on inessential aspects of Hume's reasoning.
In fairness to Dewey we may observe that the argument from Section
VII does depend on Hume's reductionism. Specifically, premise (2) of
argument (UMF2) is driven by Hume's analysis of necessary connection,
and this analysis depends on Hume's position that every idea must be
traced back to some original impression or impressions. The reason
Hume thinks (2) is true is that he thinks there must be some impression
behind our idea of necessary connection, and the only one he can find
for the job is the expectation "we feel in the mind" when two things
are observed to be constantly conjoined.
Every idea is copied from some preceding impression or sentiment; and where
we cannot find any impression, we may be certain that there is no idea. In all
single instances of the operation of bodies or minds, there is nothing that
produces any impression, nor consequently can suggest any idea, of power or
necessary connexion. But when many uniform instances appear, and the same
object is always followed by the same event; we then begin to entertain the
notion of cause and connexion. We thence/ a new sentiment or impression, to
wit, a customary connexion in the thought or imagination between one object
and its usual attendant; and this sentiment is the original of that idea which we
seek for. (p. 78)
So Hume's argument in Section VII does presuppose his empiricist
theory of ideas. But how is the argument of Section IV supposed to
depend on that theory?
Here is one way that it might be thought to do so. It might be
charged that Hume's reductionism is implicit in premises (1) through (3)
of (UMF1). Those premises invoke the distinction between moral and
demonstrative reasoning, and these two concepts are defined in terms of
matters of fact and relations of ideas. Therefore, the distinction between
moral and demonstrative reasoning implies the distinction between matters of fact and relations of ideas. But this latter distinction, the present
objection goes, involves empirical reductionism, and therefore (UMF1)
really does depend on Hume's theory of ideas.

143

It should be conceded that Hume's distinction between matters of

fact and relations of ideas does involve empirical reductionism. Without
going into details of Hume scholarship, the distinction presupposes the
notion that complex ideas are built up out of simple ideas, and exist in
various kinds of relations. But what is important to recognize is that
(UMFl) can be reconstructed without the distinction between matters
of fact and relations of ideas, and without its reductionist presuppositions. For that reason (UMFl) does not depend on Hume's theory of
ideas in any essential respect.
What Hume needs to run his argument is not the distinction between
matters of fact and relations of ideas, but the distinction between contingent and necessary truths. More exactly, Hume requires a distinction
between contingent truths arrived at by empirical reasoning and necessary truths arrived at by a priori reasoning. It is true that Hume's theory
of ideas provides one way of cashing out the relevant distinction. But it
is not the only way or even the most plausible way, and therefore
(UMFl) does not essentially involve that theory.3
Another way in which (UMFl) might be supposed to involve
Hume's theory of ideas is in premise (4). That premise makes reference
to the notion of "present sense," and so it might be thought that Hume's
atomism about experience plays an essential role in the argument here.
But again, the notion of "present sense" need not be understood in
Hume's atomistic way. The relevant meaning of "present sense" in the
context of (UMFl) is empirical observation. In this regard premise (4)
claims only that a principle about unobserved occurrences cannot be
observed to be true, and therefore cannot be established by empirical
observation alone. If such a principle is to be justified at all, then its
evidence must involve some other source of knowledge, and presumably
some kind of inference from what is empirically observed. That claim
seems plausible enough, and certainly does not require the theory of
ideas in general or Hume's atomistic conception of experience in particular.
Taking all of this into account, we can reconstruct argument (UMFl)
as follows, purging it of any remnants of Hume's theory of ideas.

In fact, Hume does not even need this distinction. What he needs is the assumption that
the regularity principle must be justified by empirical reasoning, together with the assumption that empirical reasoning always involves the regularity principle. However, I will continue to employ the present distinction in reconstructing Hume's argument, since this more
closely parallels Hume's argument in the text.

144

(UMF1B)

1. All reasoning is either empirical reasoning about contingent

truths or a priori reasoning about necessary truths.
2. The regularity principle cannot be justified by a priori reasoning about necessary truths, since its denial is not a contradiction. It does not state a necessary truth.
3. The principle cannot be justified by empirical reasoning about
contingent truths, since all such reasoning presupposes the
very thing in question, and therefore such a justification
would be circular.
[4. Neither can the principle be justified by intuition, empirical
observation, or memory alone, since none of these can give
direct knowledge of a factual supposition about unobserved
cases.]
5. Therefore, the supposition that unobserved cases will resemble observed cases cannot be justified at all. (1,2,3,4)
6. All of our beliefs about matters of fact that go beyond current
empirical observation and memory depend on that supposition for their justification.
7. Therefore, all of our beliefs about unobserved matters of fact
are themselves unjustified, since they depend on an unjustified
supposition. (5,6)

We may conclude that Hume's skeptical argument in Section IV does

not essentially involve his empiricist theory of ideas, considered either as
a theory of experience or as a theory about the analysis of ideas and
beliefs. For this reason objection (a) is ineffective against Hume's reasoning. The objection focuses on superficial aspects of Hume's arguments
as they happen to appear in the text, and not on anything essential in
Hume's reasoning.

3. OBJECTION (b): HUME REQUIRES ABSOLUTE CERTAINTY

FOR KNOWLEDGE
The next objection charges that Hume's arguments require absolute
certainty for knowledge. If so, Hume's skepticism is without force, since
knowledge requires only some lesser degree of positive epistemic status.4
We may reply on behalf of Hume by again focusing on the argument
from Section IV of the Enquiry. According to (UMF1), the problem
4

For an example of this kind of objection see Rescher, Scepticism: A Critical Appraisal, esp.
Ch. II. We may also include here Dewey's railings against the quest for certainty and Peirce's
insistence on fallibilism.

145

with our beliefs about unobserved matters of fact is that they lack any
positive epistemic status whatsoever; it is not that they lack absolute
certainty.
We can see this by looking at the basic structure of (UMF1). In
premises (2) through (4) Hume identifies five possible sources of justification and declares four of the five to be irrelevant to the supposition
that unobserved cases will resemble observed cases. That supposition is
a contingent truth about unobserved cases. Memory deals with the past,
and empirical observation deals with present observed cases. Demonstrative reasoning and intuition deal with necessary truths. The only relevant
source ofjustification for the supposition would be moral reasoning, but
moral reasoning so employed would be circular.
The problem with circular reasoning, however, is not that it fails to
provide absolute certainty. Rather, the problem is that circular reasoning
provides no support at all for conclusions drawn from it. If Hume's
argument is sound, then our beliefs about unobserved matters of fact are
wholly lacking in positive epistemic status. Absolute certainty is not the
issue.

II. The Standard Objection against Hume

Objection (c) charges that Hume is a deductivist. That is, the objection
claims that Hume assumes that only deductive inferences can generate
knowledge and justified belief, and so Hume's skeptical arguments can
be avoided by simply recognizing the role of inductive inferences in
empirical matters. This is perhaps the most widely accepted critique of
Hume's skepticism about unobserved matters of fact. Barry Stroud calls
it "the standard interpretation" of Hume, and D. C. Stove says that
important aspects of it are "nearly common property among philosophers." 5 In Chapter 3, I rejected the charge of deductivism as a dismis5

See Barry Stroud, Hume (London: Routledge and Kegan Paul, 1977), p. 56, n. 11; D. C.
Stove, Probability and Hume's Deductive Scepticism (Oxford: Clarendon Press, 1973), p. 51.
Versions of the objection to Hume can be found in Reid and Peirce, but it has been more
recently articulated and further developed by Stove in several essays and in his book just
cited. Some form of the objection is defended in P. Edwards, "Bertrand Russell's Doubts
about Induction," in Anthony Flew, ed., Logic and Language, First Series (Oxford: Blackwell,
1951); P. F. Strawson, Introduction to Logical Theory (London: Methuen, 1952); Stroud, Hume;
and Anthony Flew, David Hume (Oxford: Blackwell, 1986), among others.

146

sive response to Hume's skepticism about the world. But as a response

to the argument in Section IV of the Enquiry the objection has some
plausibility. To see why, we will have to take a closer look at the
objection and at Hume's argument.

1. THE OBJECTION
At first glance the objection seems wrong. For Hume clearly distinguishes between demonstrative and moral reasoning and then argues that
neither of these can justify our beliefs about unobserved matters of fact.
If we make the plausible assumption that Hume's "moral reasoning"
roughly corresponds to what we mean by "inductive reasoning," then
Hume cannot be said to assume that inductive reasoning is not epistemically efficacious. On the contrary, Hume has given an argument for
that position: such reasoning always employs the regularity principle, but
that principle is unjustified and must remain unjustified. In particular,
inductive reasoning cannot justify the principle because such reasoning
presupposes it, and so that would be going in a circle.
How, then, does Hume assume that only deductive inferences give
rise to knowledge and justified belief? At this point the objection gets
interesting. Why, it is asked, does Hume think that all moral reasoning
presupposes the regularity principle? The answer, according to the present objection, is that Hume thinks that the principle is needed to make
moral reasoning deductively valid. So although Hume explicitly makes
a distinction between demonstrative and moral reasoning, he implicitly
assumes that only deductive reasoning is epistemically respectable.
Hume's assumption that the regularity principle must be involved in
moral reasoning is really an assumption that moral reasoning must be
deductive.
Stove articulates the objection nicely in the following passages.
Sometimes when we say of an argument from p to q, that it presupposes r, our
meaning is as follows: that, as it stands, the argument from p to q is not valid,
and that, in order to turn it into a valid argument, it would be necessary to add
to its premisses the proposition r. I believe that this is the sense in which
"presuppose" occurs in . . . Hume's argument.
Hume's argument in stage 2 may therefore be summed up in the following
way: from premisses which prove at most the invalidity of predictive-inductive
inferences, along with the unstated premiss that an inference is unreasonable if

147

it is invalid, Hume concluded that predictive-inductive inferences are unreasonable.6

This objection is sophisticated and interesting, and as was already noted,
it is popular as well. If the objection trades on an accurate understanding
of Hume's argument, then it is also devastating, since the rationalist
assumption that only deductive inferences are reasonable is no longer
taken seriously in philosophy, and rightly so. But in my opinion Hume's
argument can be reconstructed so as to avoid the present objection. For
that reason the objection misses the real force of Hume's skeptical
reasoning.
I will not argue that Hume's text forces the reconstruction I will
propose. Rather, I claim only that the text is consistent with it. But then
the principle of charity warrants my interpretation insofar as it makes
what Hume says not obviously wrong. More importantly, whether or
not the proposed interpretation is an accurate account of Hume's intentions, the substantive point holds: Hume's reasoning can be reconstructed
so that it does not presuppose deductivism. A close analysis of Hume's
argument teaches a more important lesson than is usually supposed.
How, then, should Hume's argument be reconstructed? My strategy
is to distinguish three kinds of inferential support. Thus the relationship
between the premises and the conclusion of an argument can be deductive-supportive, inductive-supportive, or nonsupportive. My thesis will
then be as follows: Hume does not assume that inductive inferences
cannot generate knowledge or reasonable belief. Rather, he makes the
weaker assumption that non-supportive inferences cannot. If this is right,
then Hume's argument is far more powerful than is generally acknowledged. For it is not at all obvious that non-supportive evidence can be
epistemically efficacious; many philosophers would agree with Hume on
that point. If Hume is granted this weaker assumption, however, then
the rest of his argument is flawless.
In Sections II.2 and II.3 of this chapter, I show how Hume's argument can be reconstructed along these lines. In the final section (Section
II.4), I focus on the real lesson of Hume's argument for positive epistemology. That lesson is not that our knowledge of unobserved matters of
fact rests on inductive inferences. Rather, it is the more surprising
conclusion that such knowledge rests on nonsupportive inferences, or
inferences that are not even inductive-supportive.

Stove, Probability, p. 43, p. 51.

148

2. DEDUCTIVE, INDUCTIVE, AND NONSUPPORTIVE

INFERENCES
In the Treatise and the Enquiry Hume employs the concepts of demonstrative and moral (or "probable") reasoning, whereas I have also been
using the concepts of deductive and inductive inference. It is now
necessary to say more about how Hume understands the former and I
am understanding the latter.
First, we should observe that Hume's concepts of demonstrative and
moral reasoning are multifaceted. Some of the descriptive and epistemic
dimensions of demonstrative and probable reasoning can be extracted
from the following passages:
All the objects of human reason or enquiry may naturally be divided into two
kinds, to wit, Relations of Ideas, and Matters of Fact. Of the first kind are the
sciences of Geometry, Algebra, and Arithmetic; and in short, every affirmation
which is either intuitively or demonstratively certain. . . . Propositions of this
kind are discoverable by a mere operation of thought, without dependence on
what is anywhere existent in the universe. . . . Matters of fact, which are the
second objects of human reason, are not ascertained in the same manner; nor is
our evidence for their truth, however great, of a like nature with the foregoing.
The contrary of every matter of fact is still possible; because it can never imply
a contradiction. . . . (Enquiry, p. 25)
All reasoning may be divided into two kinds, namely, demonstrative reasoning,
or that concerning relations of ideas, and moral reasoning, or that concerning
matters of fact and existence. (Ibid., p. 35)
Descriptively, we leam that moral reasoning deals with matters of fact,
whereas demonstrative reasoning deals with necessary truths, or truths
the contrary of which imply a contradiction. We also learn that demonstrative reasoning may proceed "by a mere operation of thought" or
"reasonings a priori," whereas later we learn that all probable reasoning
"arises entirely from experience" (ibid., p. 27).
The concepts of demonstrative and moral reasoning also involve
epistemic dimensions. Demonstrative reasoning gives rise to knowledge
that is "certain." The evidence gained by probable reasoning, however
great, is not "of a like nature." In the Treatise Hume speaks of "degrees
of evidence." Demonstrative reasoning gives knowledge in the strict
sense of the term. Moral reasoning gives only a lesser degree of evidence,
although "in common discourse we readily affirm, that many arguments

149

from causation exceed probability, and may be received as a superior

kind of evidence" (Treatise, p. 124).
The terms "deductive inference" and "inductive inference" are not
Hume's but ours. The terms have a variety of meanings, many of which
include descriptive and epistemic dimensions that are similar to those
implied by Hume's concepts of demonstrative and moral reasoning. But
there is also a common understanding of deductive and inductive inference that is neither descriptive nor epistemic, but semantic. On this
understanding an inference is deductive just in case, necessarily, if the
premises are true then the conclusion is true. This understanding of
deductive inference implies nothing descriptive of the premises or conclusion, nor does it imply that any of them are known or certain or
even reasonable. As I said, the concept so understood is semantic rather
than descriptive or epistemic. It is semantic in the sense that it concerns
a necessary relation between the truth of the premises and the truth of
the conclusion. This is the understanding of deductive inference that is
used in logic and mathematics.
The semantic notion of deductive inference may be opposed to a
semantic notion of inductive inference. On this understanding an inference is inductive just in case, necessarily, the truth of the conclusion is
likely in relation to the truth of the premises. Here "likely" is to be
understood in a non-epistemic sense, meaning something close to statistically probable, as opposed to epistemically probable or reasonable. The
concept is semantic in the relevant sense; it concerns a relation between
the truth of the premises and the truth of the conclusion, and that
relation is necessary. This relation is different from that which is involved
in deductive inferences, however. First, unlike entailment, inductive
support can change by adding premises. Second, inductive support is
weaker than entailment. It might be understood as a kind of partial
entailment, in that it does not preserve truth but it does preserve likelihood of truth.7
Examples of inductive inferences so defined are as follows:

Ian Hacking attributes a similar concept to Jeffreys and Keynes. "[I]n the early decades of
this century, there was much interest in the theory advanced by Harold Jeffreys and J. M.
Keynes, according to which the probability conferred on a hypothesis by some evidence is
a logical relation between two propositions. The probability of/*, in the light of e, is something like the degree to which h is logically implied by e." Ian Hacking, The Emergence of
Probability (Cambridge: Cambridge University Press, 1975), pp. 13-14.

150

i. Ninety-five percent of the dogs that live in New York City bite.
Fido is a dog that lives in New York City.
Therefore, Fido bites.
ii. Patent attorneys usually make good money.
Joe is a patent attorney.
Therefore, Joe makes good money.
iii. People who make promises they can't keep tend to regret it.
Therefore, if you make a promise you can't keep you will regret it.
iv. The grocery store used to be on Main Street.
Things haven't changed much around here.
Therefore, the grocery store is still on Main Street.

In each of these arguments there is a relationship of inductive support

between the premises and the conclusion in the sense defined; in each
argument, it is necessarily true that the conclusion is likely in relation to
the premises. This is not to say that in each argument the conclusion
would remain likely in relation to the premises if others were added to
the ones already stated. As was noted earlier, inductive support is unlike
deductive entailment in that adding premises can change the relationship
of support between the premises and the conclusion.
From what has been said it should be clear that arguments may be
divided into three categories: those that involve deductive inferences,
those that involve inductive inferences, and those that involve neither.
Alternatively, we may say that arguments can be deductive-supportive,
inductive-supportive, or nonsupportive. We may talk about evidence
for a conclusion in the same way. Thus evidence for a conclusion can
be deductive-supportive, inductive-supportive, or nonsupportive.
It is important to note that none of this concedes or denies anything
to the skeptic. This is because "supportive" is here understood as a
semantic notion, not an epistemic one. It remains an open question
whether epistemic grounding requires semantic support, and, if it does,
whether inductive support is sufficient.
Now that we have established a vocabulary, we may consider the
standard objection against Hume's skeptical argument. In our terminology the objection is that Hume assumes that only deductive-supportive
reasoning can give rise to knowledge and reasonable belief. Another way
to put the objection is that Hume thinks all evidence must be deductivesupportive, so that if one's evidence does not entail what is concluded
from it then that conclusion cannot amount to knowledge or even

151

reasonable belief. The objection, of course, insists that evidence which

is inductive-supportive can be sufficient for grounding knowledge.
Therefore, Hume's argument rests on a false assumption about what
kind of support is required for knowledge.
We get our first clue that the standard objection is misguided when
we look at various of Hume's characterizations of the regularity principle. In all of these Hume states the principle in completely general terms.
Hume always characterizes the principle as affirming that there are (in
general) regularities in nature, but never as affirming that some particular
regularity holds.
[A]ll our experimental conclusions proceed upon the supposition that the future
will be conformable to the past. (Enquiry, p. 35)
[A]ll inferences from experience suppose, as their foundation, that the future
will resemble the past, and that similar powers will be conjoined with similar
sensible qualities. (Ibid., p. 37)
If reason determin'd us, it wou'd proceed upon that principle, that instances, of
which we have had no experience, must resemble those, of which we have had experience,
and that the course of nature continues always uniformly the same. (Treatise, p. 89)

But if the regularity principle mentions no particular regularities, then

adding the principle to arguments from observed cases to unobserved
cases will not make those arguments deductive-supportive. And so we
cannot say that Hume is insisting on deductive-supportive reasoning
when he insists on the need for supposing the regularity principle.
Consider the argument that, in all past cases, bread has nourished me,
and that therefore this bread will nourish me as well. Using Hume's
second formulation we have,
v. In the past, in all observed cases bread has nourished me.
The future will resemble the past; similar powers will be conjoined with
similar sensible qualities.
Therefore, this bread will nourish me as well.
Because Hume's regularity principle does not say how the future will
resemble the past, or which powers will be conjoined with which
sensible qualities, the argument is not turned into a deductive one by
adding the principle. Of course, a principle affirming a particular regularity between bread and nourishment would yield a deductive argument
to the targeted conclusion. But as we can see from the various statements
of the principle just quoted, Hume never characterizes the regularity

152

principle as affirming particular regularities. For Hume the regularity

principle is general in this sense.8
If Hume does not insist that we need the regularity principle for the
purpose of making our reasoning deductive-supportive, then what is his
point? My suggestion is that Hume thinks the principle is needed to
make our reasoning inductive-supportive. That this is Hume's intention
is suggested by the following passage from the Enquiry:
If there be any suspicion that the course of nature may change, and that the past
may be no rule for the future, all experience becomes useless, and can give rise
to no inference or conclusion, (pp. 3738)
In this passage Hume is arguing that if the regularity principle is not
supposed, then our present and past observations become irrelevant to
future events. Without the assumption of regularity, premises about the
past and present would be irrelevant to that is, wholly nonsupportive
of conclusions about the future. And in the sense in which we have
been understanding "support," Hume is right. For it is not necessarily
true that, given observations of a past constant conjunction, that conjunction will likely continue in the future. If the universe is chaotic
rather than regular, then a past constant conjunction does not make a
future conjunction even likely.
The present point is not that past observations do not entail conclusions about future conjunctions. That is obvious. The point is that,
without the assumption that nature is regular, past observations do not
give even inductive support to conclusions about the future. However,
if we add the regularity principle to past observations, then our evidence
does become inductive-supportive. For this principle is equivalent to the
proposition that, in general, constant conjunctions in the past will continue. On this interpretation the role of the regularity principle is to
make the premises of probable reasoning inductively relevant to their
conclusions. The principle is needed to make probable reasoning inductive-supportive rather than nonsupportive.
It must be admitted that the passage just quoted from Hume does not
force the interpretation I have given it. But on the interpretation I am
suggesting, key passages from both the Treatise and the Enquiry become
8

Some commentators on Hume have noticed that adding the regularity principle (as Hume
formulates it) to one's evidence for unobserved matters of fact does not indeed generate
deductive inferences. But they have missed the significance of this, taking it to be a criticism
of Hume rather than an objection to their interpretation. See, e.g., Strawson, Introduction to
Logical Theory.

153

not only coherent but plausible. For example, the main argument from
Section IV of the Enquiry can now be read as follows:
(UMF3)

1. Reasoning can provide justification for its conclusion only if

it is supportive, either deductively or inductively, and its assumptions are themselves justified or reasonable to suppose.
2. Demonstrative reasoning is deductive-supportive, but it is not
relevant to matters of fact.
3. Therefore, beliefs about unobserved matters of fact must be
justified, if at all, by moral reasoning that is inductivesupportive. (1,2)
4. Moral reasoning is inductive-supportive only if it supposes
the regularity principle.
5. That principle cannot be justified by demonstrative reasoning, since the principle does not state a relation of ideas.
6. The principle cannot be justified by moral reasoning, since
moral reasoning already supposes it as a premise, and so
that would be circular.
[7. Neither can the principle be justified by intuition, present
sense, or memory, since none of these is relevant to a
factual supposition about unobserved cases.]
8. Therefore, the regularity principle cannot be justified at all.
(5,6,7)
9. Therefore, moral reasoning cannot provide justification for
beliefs about unobserved matters of fact. (1,4,8)
10. Therefore, beliefs about unobserved matters of fact cannot be
justified at all. (3,9)

The present interpretation of Hume runs the argument on premises (1)

and (4) rather than on the dubious assumption that only deductive
inferences are justification conferring. This is an advantage over the
standard interpretation of Hume insofar as it is more charitable. The
present interpretation, in fact, saves Hume from an embarrassing blunder.
The interpretation I am suggesting also does very well with the
following passage of the Treatise:
[T]he next question is Whether experience produces the idea by means of
understanding or of the imagination; whether we are determin'd by reason to
make the transition, or by a certain association and relation of perceptions. If
reason determin'd us, it wou'd proceed upon that principle, that instances, of
which we have had no experience, must resemble those, of which we have had experience,
and that the course of nature continues always uniformly the same. In order therefore

154

to clear up this matter, let us consider all the arguments upon which such a
proposition may be suppos'd to be founded, (pp. 8889)

Suppose that by "reason" Hume means any reasoning that is either

deductive-supportive or inductive-supportive, and by "imagination" he
means any transition that is wholly non-supportive. Then a natural
interpretation of the passage is as follows: "We have already established
that the transition from observed matters of fact to unobserved is not by
demonstrative reasoning. Therefore, if it is by reasoning that we make
this transition, it will have to be by moral reasoning, or reasoning that is
inductive-supportive. Now our reasoning here is inductive-supportive
only if it proceeds on the regularity principle. But that is the case only
if that principle is itself well founded, and so let us investigate that issue."
Any number of passages in the Treatise and the Enquiry can be given a
quite natural interpretation along similar lines.

3. AN OBJECTION CONSIDERED (AGAINST OUR

INTERPRETATION OF HUME)

I now want to consider the objection that the present interpretation of

Hume is only marginally more charitable than the standard one, and that
therefore (UMF3) is only marginally more plausible than the argument
commonly refuted. I have made (UMF3) run on the assumption that
only inductive and deductive inferences can confer justification on their
conclusions. One might argue that this is only slightly more plausible
than the assumption that only deductive inferences can confer justification. This is because I have defined inductive support in such a way that
almost none of our reasoning involves inductive inferences. That is,
almost none of our reasoning is such that, necessarily, the truth of the
conclusion is likely in relation to the truth of the premises.
To see the force of this objection we may note that even arguments
(i) through (iv) in Section 2 would not stand up to the standards I have
Hume endorsing, since the question of the reasonableness of their premises arises, and it is clear that there is no inductive-supportive argument
that will establish those premises and that does not itself depend on some
form of the regularity principle. The reasonableness of some of those
premises can be attributed to sense perception or memory, but other
premises would require an inference from sense perception and memory,
and there is no inductive-supportive argument available to do the job.
It might be thought that argument (i) is an exception, but even here the

155

point holds. If the first premise of that argument is supported by observing all of the dogs in New York City, then Fido was observed, and so
the conclusion does not go beyond observed cases. If the first premise is
not supported by observing all dogs, then it must be arrived at via an
argument involving the regularity principle or else via an argument that
is nonsupportive.
What this objection recognizes is that Hume's skeptical argument
runs as nicely on the assumptions I have attributed to him as it does on
the dubious assumption that only deductive inferences are justification
conferring. However, the objection goes, the assumptions I have put in
Hume's argument are just as implausible as the rationalist deductivism
others have attributed to him. That is the point I want to address now.
I do so by articulating a non-rationalist motivation for the assumption
that all evidence must be deductive-supportive or inductive-supportive.
I argue that the assumption arises quite naturally out of considerations
about what knowledge requires, and that rejecting the assumption
threatens results that many have found counterintuitive. My conclusion
is that rejecting the assumption amounts to a substantive and even
surprising position regarding the nature of evidence. 9
The assumption that I have been attributing to Hume amounts to
this: An inference can confer justification on its conclusion only if there
is a necessary relation of support between the premises and the conclusion. Alternatively, evidence can confer justification on a belief only if it
is deductive-supportive or inductive-supportive. Why might one think
that such an assumption is plausible? We can answer that question by
considering some cases of reasonable and unreasonable belief.
First, consider a mathematician who knows that certain axioms are
true and who proves a theorem on the basis of those axioms. By seeing
that the theorem is entailed by the axioms, the mathematician comes to
know the theorem. Contrast this situation with that of a novice who
also knows that the axioms are true but who merely guesses that the

The assumption that good inductive inferences should necessarily confer probability is manifested in Carnap's attempt to formalize inductive logic and in Bayesian attempts to reduce
inductive reasoning to the laws of probability. See, e.g., Rudolf Carnap, Logical Foundations
of Probability (Chicago: University of Chicago Press, 1950), and Paul Horwich, Probability
and Evidence (Cambridge: Cambridge University Press, 1982). It is perhaps also behind
Peirce's attempts to prove that the scientific method necessarily discovers the truth in the
long run. An informative historical overview of different concepts of probability and induction, but in somewhat different categories, is provided by Hacking, The Emergence
of Probability.

156

theorem is true. Obviously the novice does not come to know the
theorem, and a reasonable explanation is that he has no awareness of the
relationship between the axioms and the theorem. If the novice knew
that the axioms entailed the theorem, then he too would know the
theorem.
Next consider a case of reasoning about a matter of fact. A mechanic
sees green liquid dripping from underneath the front of a car and infers
that the car's radiator is leaking. The mechanic's belief is at least reasonable and may even amount to knowledge. Surely one relevant feature of
the case is that the mechanic knows that dripping green liquid is a
reliable indication of a leaking radiator. Again, the mechanic is aware of
the relationship between the truth of her premises and the truth of her
conclusion. Suppose that someone not very familiar with automobiles
observes green liquid leaking from his car. If the person does not understand that this is a reliable indication that his radiator is leaking, then
surely he cannot know on the basis of his observation that his radiator is
leaking. Or suppose that dripping green liquid were not a reliable indication of a leaking radiator. Then even if a person thought that it was,
the observation of dripping liquid could not allow him to know that his
radiator was leaking.
These examples support two conclusions about the nature of evidence. First, a body of evidence E can confer justification on a conclusion C only if the truth of E is a reliable indication of the truth of C.
Second, a body of evidence E can confer justification on a conclusion
C for a person S, only if S is sensitive to the fact that the truth of E is a
reliable indication of the truth of C. 10
The following question now arises: How is it that in cases of knowledge and reasonable belief, S becomes sensitive to the fact that the truth
of her evidence is a reliable indication of the truth of her conclusion? In
cases where the evidence is deductive-supportive or inductivesupportive a possible answer is that S can just "see" that the truth of the
conclusion is entailed or made likely by the truth of the evidence. The
"seeing" here is not literal seeing, but a kind of intuitive seeing associated with the knowing of self-evident necessary truths. The idea is that

10 Compare Richard Fumerton, Metaepistemology and Skepticism (Lanham, MD: Rowman and
Littlefield, 1995). According to Fumerton, the following "Principle of Inferential Justification" is a platitude: To be justified in believing one proposition P on the basis of another
proposition E, one must be (1) justified in believing E and (2) justified in believing that E
makes probable P.

157

deductive-supportive and inductive-supportive reasoning involves such

a logical intuition. When we employ either kind of argument, we can
"see" that, necessarily, the truth of the conclusion is indicated by the
truth of the premises.
Now suppose that there are inferences that are justification conferring
but that are neither deductive nor inductive in the senses we have
defined. That would amount to supposing that there are inferences
where the truth of the evidence is a reliable indication of the truth of
the conclusion, but where that fact is contingent rather than necessary.
It is plausible that the reasoning about the leaky radiator involves an
inference of just this kind. In fact, it is plausible that all of our reasoning
about unobserved matters of fact involves inferences of just this kind.
Indeed, this latter conclusion is what I take to be the import of Hume's
argument concerning unobserved matters of fact. More exactly, Hume's
argument establishes that if our reasoning about unobserved matters of
fact is reliable at all, then it is only contingently reliable. Such reasoning
is neither deductive-supportive nor inductive-supportive, and this is
equivalent to saying that it is not necessarily reliable.
So again, suppose that this is what our inferences are like. Why is
that a problem? The problem is that we are now at a loss to say how we
could be sensitive to the reliability of those inferences. The examples we
have considered naturally gave rise to two intuitions: that inferences
must be reliable to confer justification, and that we must be aware of
that reliability. If the reliability of our inferences is necessary, then we
can account for our sensitivity to their reliability by invoking our ability
to know self-evident necessary truths. But this account is blocked if our
inferences are only contingently reliable. If they are contingently reliable,
then their reliability cannot be self-evident. And if it is not self-evident
then how can it be evident at all? Of course one might begin to look
for an argument that our inferences are reliable, but that would only
raise all of the same questions regarding the reliability of that argument.
I will now summarize the main point of this section: The objection
was raised that (UMF3) is only marginally more plausible than the
argument from Hume that is commonly refuted. This is because
(UMF3) runs on the assumption that only supportive inferences can be
justification conferring, but given the notion of support employed, that
assumption is only slightly more plausible than the assumption that only
deductive inferences can be justification conferring. On the present
understanding, an inference is supportive only if it is necessarily reliable,
and so premise (1) of (UMF3) amounts to the assumption that only

158

necessarily reliable inferences are justification conferring. In response, I

have tried to motivate the assumption that only necessarily reliable
inferences are justification conferring. The argument is that it is a requirement of knowledge and reasonable belief that one be sensitive to
the reliability of any inference that is involved. But if an inference is not
necessarily reliable, then it is hard to see how one could be sensitive to
its reliability. Therefore, there is a strong motivation for thinking that
only necessarily reliable inferences can generate knowledge and reasonable belief.
4. THE LESSON OF HUME'S SKEPTICAL ARGUMENT

I have been arguing that skeptical arguments serve as heuristic devices

for driving positive epistemology. As such, skeptical arguments teach us
lessons about the nature of knowledge and evidence. What is the lesson
that Hume's argument teaches?
If Hume is granted the assumption that only logically supportive
inferences are justification conferring, then the rest of his argument goes
through and we must conclude that we have no knowledge or reasonable belief about unobserved matters of fact. To see that this is so we
need only review the argument. If the assumption is granted, then any
reasoning about unobserved matters of fact must be deductivesupportive or inductive-supportive. That there are no deductivesupportive arguments about unobserved facts from past and present
observations is now almost universally acknowledged, and I think Hume
must be conceded that point. Therefore our reasoning about unobserved
matters of fact will have to be inductive-supportive. But now such
reasoning will be inductive-supportive only if its premises include something like the regularity principle; without some such principle it will
not be necessarily true that the conclusions of such reasoning are likely
in relation to their premises. Furthermore, such a principle cannot be
supported by either deductive or inductive arguments. Once again, we
should concede the point that there are no deductive arguments. But
neither can the regularity principle be supported by inductive arguments,
and for just the reason Hume tells us. If all inductive reasoning already
assumes the regularity principle, then using inductive reasoning to support it would be "going in a circle" in the most obvious sense; it would
be assuming as a premise that which is supposed to be the conclusion.
So if we grant Hume the assumption that only supportive inferences
are justification conferring, then the rest of his skeptical argument goes

159

through. Therefore, one lesson of Hume's argument is straightforward:

we must conclude that nonsupportive inferences can be justification
conferring. Put another way, any adequate epistemology will have to
countenance evidence about unobserved matters of fact that is neither
deductive-supportive nor inductive-supportive. Providing the details
about how this is possible constitutes one condition of an adequate
theory of evidence.
But that raises another problem. We saw that the motivation for
accepting Hume's assumption was that knowledge seems to require that
one be sensitive to the reliability of one's inferences, and it is hard to see
how one could be sensitive to the reliability of inferences that are only
contingently reliable. An adequate epistemology will have to solve this
problem as well.
One way to solve the problem is to make the case that one need not
be sensitive to the reliability of one's inferences. This is the strategy that
simple reliabilism adopts. According to simple reliabilism, S is justified
in believing p just in case the cognitive process that generates S's belief
is in fact reliable. Thus if S's believing p is the result of some inferential
cognitive process, S need not be aware that the process or the inference
it involves is reliable; all that counts is that the process is in fact reliable.
Here we have an example of an epistemology that countenances nonsupportive inferences in the sense that any adequate epistemology must.
But it does so at the cost of denying one of our strongest intuitions
about the nature of evidence: namely, it denies that S need be sensitive
to the reliability of her evidence. Simple reliabilism has been widely
criticized on just this point.11
Moreover, simple reliabilism is not the only theory with this kind of
problem. Consider Robert Audi's view of indirect justification in his
book The Structure of Justification.12 According to Audi, a person S is
justified in believing p on the basis of r only if there is some support
relation C, and S believes C to hold between r and p. Audi calls this last
belief a "connecting belief." A trilemma arises, depending on how we
think of connecting beliefs and the support relation they involve. If
connecting beliefs express contingent reliability relations, then the problem of their justification arises as we have noted: their justification will
have to be inferential, and so will involve either a circle or a regress of
11

For example, see Bonjour, The Structure of Empirical Knowledge, and Fumerton, Metaepistemology and Skepticism.

12 See esp. Ch. 8.

160

further connecting beliefs. If connecting beliefs express necessary relations, then their justification can be non-inferential, but a different
problem arises for either of two further possibilities. If the beliefs express
a semantic relation implying reliability, then Humean problems arise as
before: r and p are semantically related only if r involves something like
the regularity principle, and so we encounter problems about the justification of r. If connecting beliefs express a non-semantic relation, perhaps about some kind of epistemic justification that does not imply
reliability, then Audi's position fails to accommodate the intuition
above: that knowledge requires that a person be sensitive to the reliability of her evidence.13
This suggests that an adequate epistemology ought to adopt a different
strategy for solving the problem at hand. Such an epistemology must
give some account of how we can be sensitive to the reliability of our
inferences, even when those inferences are not necessarily reliable, and
even if we do not have beliefs about those inferences. How such an
account would go is an open question, and the Humean considerations
we have just rehearsed should convince us that it is a difficult one. In
any case, Hume's skeptical argument suggests this further condition of
an adequate theory of evidence.
To sum up, in this chapter I have offered a reconstruction of Hume's
skeptical reasoning that is consistent with the text, and that avoids the
standard objection that Hume assumes only deductive inferences can be
justification conferring. Second, I have argued that the resulting skeptical
argument is a powerful one, and that rejecting it forces us to adopt a
substantive and even surprising position in our positive account of the
nature of knowledge and evidence. Finally, I have argued that Hume's
skeptical reasoning suggests two conditions for an adequate theory of
evidence. Specifically, any adequate theory will have to explain (a) how
contingently reliable inferences can be justification conferring and (b)
how knowers can be sensitive to the reliability of such inferences.
In Chapter 7, I defend a theory of knowledge and evidence that does
just this. The theory is a version of reliabilism, because it holds that
knowledge arises from the reliable cognitive powers of believers. As
such, the theory adopts the strategy of simple reliabilism in making the
reliability of empirical reasoning contingently reliable. But the theory
goes beyond simple reliabilism by affirming that knowledge requires a
13 A similar trilemma arises for Fumerton, since he accepts the Principle of Inferential Justification. See my note 11.

161

sensitivity to the reliability of one's inferences. The trick is to understand

such sensitivity in a way that both (a) avoids Humean circularity problems and (b) is psychologically plausible.
The main idea is as follows. First, inference patterns that give rise to
reliable reasoning are manifested in the cognitive behavior of intellectually virtuous agents, but not in a way that requires agents to have beliefs
about their inferences. Rather, intellectually virtuous agents are sensitive
to the reliability of their inferences in the sense that they are disposed to
engage in such inferences when they reason carefully. Put differently,
intellectually virtuous agents have a disposition to try to believe the
truth, and a disposition to reason in certain ways when so motivated. In
that sense, they recognize (are sensitive to) the reliability of those ways
in achieving the aim of that motivation.
Second, the cognitive dispositions that give rise to reliable empirical
reasoning allow inferences from observed cases to unobserved cases
without first justifying anything like the regularity principle. Such dispositions make us reliable because in fact nature is regular, and therefore
inferences based on observed cases will be in fact reliable. Understanding
the notion of an intellectual virtue in this way that is, as a disposition
to reliably form true beliefs and with that motivation we can say that
virtuous agents are sensitive to the reliability of their reasoning, but not
in a way that gives rise to circularity or that is psychologically implausible.
This broad theory of knowledge also accommodates the lessons of
Chapters 3, 4, and 5, explaining why not all evidential relations are
inferential, how perceptual knowledge can be based directly on sensory
appearances, and how knowledge not based on justifying reasons is
possible. The idea is that the cognitive dispositions manifested by virtuous agents include both inferential and non-inferential dispositions.
Cognitive powers such as mathematical and empirical reasoning involve
dispositions to make certain inferences, but non-inferential powers such
as memory, intuition, and perception involve dispositions of a different
kind. For example, perception disposes us to form beliefs about objects
in the world immediately (i.e. non-inferentially) on the basis of sensory
appearances. In all cases the cognitive dispositions that virtuous agents
manifest make them reliable in relevant environments that is, such
dispositions make virtuous agents contingently reliable. But they are
reliable nonetheless.
The ideas of the preceding two paragraphs are developed in Chapter

162

7. Accordingly, in that chapter we move from negative to positive

epistemology; I defend a positive account of knowledge and evidence
on the grounds that it explains why certain skeptical assumptions are
false. The theory also captures our pre-theoretical intuitions about which
cases count as knowledge, and is psychologically plausible as well.

163

7
Agent Reliabilism

We are looking for a theory of knowledge and evidence that confirms

and explains the conclusions of Chapters 2 through 6. More specifically,
we are looking for a theory that explains (a) why not all evidential
relations are inferential; (b) how sensory evidence in particular can be
non-inferential (or how beliefs about the world can be evidentially
grounded in sensory appearances yet not inferred from sensory appearances); (c) how some knowledge can be foundational (or how some
knowledge can be based on evidence which is not itself in need of
further justifying reasons); and (d) how inferences that are only contingently reliable can nevertheless give rise to knowledge. Our theory
should also explain (e) how knowers can be sensitive to the reliability of
their inferences, and even though such inferences are only contingently
reliable, so knowers cannot just "see" that they are reliable by a kind of
logical intuition into necessary relations.
I have already suggested that agent reliabilism does all of these things.
In this chapter I develop and defend that claim. The argument occurs in
two stages. In Part I, I argue that reliabilist theories in general confirm
all of the conclusions noted in (a) through (d). In other words, "simple"
or "generic" reliabilism explains why the skeptical assumptions rejected
in Chapters 2 through 6 are false. In Part II, I argue that agent reliabilism
is the best version of reliabilism. This is because, in part, agent reliabilism
explains (e), or how knowers can be sensitive to their contingent reliability without falling into Humean circularity problems.
In the next chapter, I continue to show how agent reliabilism addresses skepticism by revisiting argument (D3) from Chapter 2. This is
the version of Descartes' skeptical argument that I claimed was susceptible to a relevant possibilities approach. In Chapter 8, I develop an

164

account of what makes a possibility relevant, or what makes a possibility

such that it must be ruled out in order that one have knowledge, and I
show how agent reliabilism can provide a theoretical explanation of that
account. In other words, the theory explains why some possibilities need
to be ruled out to have knowledge, and why other possibilities are
irrelevant and can be ignored.
If a theory of knowledge can do all of these things, then that is
powerful evidence in its favor. Such a theory would capture a wide
range of our pre-theoretical intuitions about which particular cases count
as knowledge, would explain why skepticism is false, and would be
psychologically plausible. In the final chapter of the book, Chapter 9, I
argue that agent reliabilism also gives us insight into the possibility of
moral and religious knowledge. A consideration of various skeptical
arguments suggests that empirical knowledge is grounded in the faculties
and habits of cognitively virtuous agents, and teaches us something about
the nature of the cognitive virtues involved. This puts us in a position
to consider whether similar virtue is possible regarding religious and
moral beliefs. For example, we may consider the possibility of moral
perception in the light of our conclusions about the nature of empirical
perception. Traditional arguments against moral perception, it turns out,
are no longer persuasive in the context of our more adequate understanding of empirical perception. Progress in moral and religious epistemology would constitute further evidence in favor of the theory of
knowledge being defended.

/. Simple Reliabilism
I now turn to the argument that reliabilism provides a theoretical explanation for the conclusions of Chapters 2 through 6. I begin by sketching
the view and then turn to the concerns of (a) through (d) just reviewed.
1. SIMPLE RELIABILISM: THE BIG IDEA

Simple reliabilism is the view that knowledge arises from reliable cognitive processes. Here "reliable" cannot mean "reliable in producing
knowledge." That would be to give a circular account of knowledge.
Rather, reliable cognitive processes are ones that are reliable in arriving
at truth.

Note that we need not load much into the word "truth" here. For

165

example, we need not think of truth as correspondence with reality.

Although most reliabilists do endorse a correspondence theory of truth,
that theory is not part of reliabilism as such. Plug in your favorite notion
of what truth is, or better yet, just remain commonsensical about it. The
main idea is that knowledge is produced by cognitive processes that "get
things right" or are "accurate" a good deal of the time.
The following passage from Alvin Goldman expounds this simple but
powerful idea.
We can gain insight into this problem by reviewing some faulty processes of
belief-formation, i.e., processes whose belief-outputs would be classed as unjustified. Here are some examples: confused reasoning, wishful thinking, reliance
on emotional attachment, mere hunch or guesswork, and hasty generalization.
What do these faulty processes have in common? They share unreliability: they
tend to produce error a large proportion of the time. By contrast, which species
of belief-forming (or belief-sustaining) processes are intuitively justificationconferring? They include standard perceptual processes, remembering, good
reasoning, and introspection. What these processes seem to have in common is
reliability: the beliefs they produce are generally true. My positive proposal, then,
is this. The justificational status of a belief is a function of the reliability of the
process or processes that cause it, where (as first approximation) reliability
consists in the tendency of a process to produce beliefs that are true rather than
false.1
A number of points warrant consideration. First, by a "reliable" process
Goldman does not mean an infallible one he says that a reliable
cognitive process has a "tendency" to produce or sustain beliefs that are
true rather than false. The greater the reliability, the greater the degree
of justification or positive epistemic status, and Goldman implies that a
very high degree of reliability would be required for knowledge.
Suppose Jones believes he has just seen a mountain-goat. Our assessment of the
beliefs justifiedness is determined by whether he caught a brief glimpse of the
creature at a great distance, or whether he had a good look at the thing only 30
yards away. His belief in the latter sort of case is (ceteris paribus) more justified
than in the former sort of case. And, if his belief is true, we are more prepared
to say he knows in the latter case than in the former. The difference between
the two cases seems to be this. Visual beliefs formed from brief and hasty
scanning, or where the perceptual object is a long distance off, tend to be wrong
more often than visual beliefs formed from detailed and leisurely scanning, or
1

Alvin Goldman, "What Is Justified Belief?", in Moser, p. 179.

166

where the object is in reasonable proximity. In short, the visual processes in the
former category are less reliable than those in the latter category. 2

Second, whether an inference gives rise to true belief depends on the

premises one starts with, and so Goldman needs to introduce the idea of
conditional reliability. A reasoning (or inferential) process is conditionally reliable just in case, given true premise-beliefs, the process tends to
produce true conclusions. In this sense valid deduction is perfectly reliable, while good inductive reasoning is also reliable though not perfectly
so.3 The theory that results is what Goldman calls a "historical" or
"genetic" theory, because it makes positive epistemic status depend on
the history of the belief in question. "The theory says, in effect, that a
belief is justified if and only if it is 'well-formed/ i.e., it has an ancestry of
reliable and/or conditionally reliable cognitive operations."4

2. WHY SKEPTICAL ARGUMENTS GO WRONG

With this sketch in place we can see that simple reliabilism confirms the
conclusions of Chapters 2 through 6. In other words, simple reliabilism
gives us a theory of knowledge and evidence that explains why the
skeptical assumptions rejected in those chapters are false. Let us take a
closer look.
a. Why Not All Evidence Is Inferential

First, simple reliabilism explains why not all evidential relations are
inferential. I suggested in Chapter 4 that the most general characterization of the evidential relation is as follows: A cognitive state (such as a
belief or experience) is evidence for another cognitive state if and only
if being in the first state tends to confer positive epistemic status on the
second state. If reliabilism is true, then the way that this works is through
reliable cognitive processes. Such processes take an initial cognitive state

2 Ibid., p. 180.
If we define "deduction" as a cognitive process that employs valid (truth-preserving) reasoning, then deduction is perfectly (conditionally) reliable. However, we will need to define
another process, a pseudodeduction, to recognize cases where mistakes are made. Alternatively, we may use "deduction" to designate a process whereby one intends to reason validly,
in which case not even deduction will be perfectly reliable. Compare Sosa, Knowledge in
Perspective, pp. 227-233.

Goldman, "What Is Justified Belief?", p. 183.

167

such as experience or belief, and produce an additional state in a reliable

(i.e., truth-conducive) way.
Now let us say that an inference is a movement from a premise-belief
or premise-beliefs to a conclusion-belief on the basis of their contents
and according to a general rule. According to reliabilism, this is one way
in which a belief can be evidentially grounded, since using an inferencerule can be one way in which a belief may be reliably formed. But if
reliabilism is true, then that is not the only way that an evidential relation
can be manifested. For there might be other cognitive processes that
is, processes other than employing an inference rule from belief to belief
that ground one cognitive state in another in a reliable way. Put
differently, not every movement in thought constitutes an inference
from premise-belief to conclusion-belief according to a general rule.
Moreover, these other kinds of movement might nevertheless be reliable
or conditionally reliable.
It might be objected that the present point is merely a verbal one;
the way in which we make room for evidential relations that are noninferential is by employing a restricted sense of "inference." But the idea
is more important than that. The real point is that not all movements in
thought can be evaluated by the criteria governing inferences in the
narrower sense defined earlier. Some movements of thought may be
evaluated by considering the truth of initial premises together with the
conditional reliability of an inference-rule employed. If we have both
true premises and a conditionally reliable inference-rule then the reasoning in question is deductively or inductively sound. But what if there is
no adequate inference involved, deductive or inductive? Or what if
there are not adequate "premises," in the sense of true or justified beliefs
from which we are starting? Does that entail that the movement in
thought is inadequate for generating knowledge or justified belief? Only
if we insist on using the criteria for good inferences where such criteria
are out of place. This is exactly the mistake that many skeptical arguments make, and that many non-skeptical philosophers make as well. It
is, for example, the mistake of the hermeneutical school, which holds
that all good evidence must be from within the circle of belief5
A similar critique of the hermeneutical school is made by Ernest Sosa in "Philosophical
Scepticism and Epistemic Circularity," Proceedings of the Aristotelean Society (1994): 263-290,
esp. pp. 264-266. There he cites Heidegger, Gadamer, Habermas, Foucault, and Derrida as
holding the position that "a belief can be justified and amount to knowledge only through
the backing of reasons or arguments." He attributes the same position to Davidson, Rorty,
Bonjour, and Michael Williams. In earlier work Sosa calls this position the "Intellectualist

168

b. Why Knowledge of the World Need Not Be Inferred from Sensory

Appearances

If not all evidential relations are inferential, what might be an example

of a non-inferential evidence relation? This brings us to our second
conclusion from previous chapters: sensory appearances can be good
evidence for beliefs about the world, even though such beliefs are not
inferred from appearances. Reliabilism accounts for this, since it holds
that appearances are good evidence so long as they reliably cause true
beliefs, whether by employing good inferences or in some other way.
The present question is, What is that other way?
I have already suggested in Chapter 4 that the process is more direct
than inference. On this view, the way things appear is interpreted as
being of certain objects with certain properties. I am appeared to in a
characteristic way by a tree bearing apples by a brook, and I interpret
that, quite directly and without anything like an inference, as a tree
bearing apples by a brook. I say that there is nothing like an inference
involved because I do not first form other beliefs for example, beliefs
about how things appear and then construct a clever argument to the
conclusion that what I see is a tree. Certainly I do not do any such thing
explicitly, but there is no reason to think that I do so implicitly either.
And as we have seen, there is good reason to think that I do not.
At this point it might be useful to address a possible misconception.
To say that perception does not depend on inference is not to say that it
does not depend on prior assumptions, prejudices, conceptual schemes,
and the like. Quite the contrary: such things are what make at least
many perceptions possible. But the point here is that such things do not
function as premises in an inference: they do not function as grounds of
which we are aware, and from which we draw conclusions on the basis
of sensitivity to their content and according to a general inference-rule.
Rather, they shape perception in other ways. To put things too vaguely,
they function as the causal antecedents of belief by creating perceptual
dispositions: they cause us to see one way rather than another, to discriminate some of an infinite number of patterns, to recognize some
information while filtering out other information.6 But to evaluate these

Model of Justification" and subjects it to similar critique. See, e.g. his essay "The Raft and
the Pyramid," in Knowledge in Perspective, esp. pp. 169173. Compare also James VanCleve,
"Epistemic Supervenience and the Circle of Belief," Monist 68 (1985): 90-104.
6 In Chapter 9, I invoke schema theory in psychology to make these claims about non-

169

dispositions the way one evaluates discursive reasoning is to miss their

nature altogether. According to reliabilism, the relevant criterion for
evaluation is whether the dispositions governing our perceptions make
those perceptions reliable. Do our assumptions, prejudices, and schemes
serve to put us in better touch with the world or do they merely obscure
it? If the former, then perceptual beliefs have positive epistemic status
and can even amount to knowledge. 7
Accordingly, to ask whether there is a good inference from appearance to reality misunderstands the way that appearances function as
evidence for the world. This mistake is manifested both in the skeptical
demand for such an inference and in the non-skeptical philosopher's
attempt to reconstruct one.

c. Why Some Knowledge Is Not in Need of Justifying Reasons

From what has already been said we can see how simple reliabilism
explains the possibility of foundational knowledge, or knowledge not
based on further justifying reasons. If we conceive of justifying reasons
as evidential beliefs from which knowledge is to be inferred, then not all
knowledge needs further reasons, because some knowledge is based
directly on sensory appearances. In this way the regress of justifying
reasons comes to an end. In the preceding discussion we have already
seen the theoretical explanation for this. According to simple reliabilism,
positive epistemic status is a function of the reliability of cognitive
processes, and therefore so long as perceptual processes are reliable they
generate positive epistemic status. This is so even though such processes
involve no justifying reasons from which perceptual beliefs are inferred.
So if reliabilism is true, then not all knowledge is based on justifying
reasons. This is because we are conceiving reasons as further beliefs from
inferential perception less vague. The main idea is that perception involves "scripts" or stories
through which sensory appearances are processed. The perceptual dispositions that result are
theoretically loaded, to be sure. But they are best understood, I argue, as being noninferential they ground a process very different from inferring conclusions from premises.
Goldman discusses work in empirical psychology which shows that background assumptions
and other contextual features can increase reliability in perception. See Epistemology and
Cognition, esp. pp. 188191. There he cites G. Reicher, "Perceptual Recognition as a Function of Meaningfulness of Stimulus Material," Journal of Experimental Psychology 81 (1969):
275280; E. Tulving, G. Mandler, and R. Baumal, "Interaction of Two Sources of Information in Tachistoscopic Word Recognition," Canadian Journal of Psychology 18 (1964):
6271; and D. D. Wheeler, "Processes in Word Recognition," Cognitive Psychology 1
(1970): 59-85.

170

which knowledge is inferred, and the evidence of sensory appearances

does not work in that way. But we may ask a further question about
evidential grounds, where these are understood to include any kind of
justifying evidence whatsoever. Must all knowledge be based on evidential grounds, where we understand grounds to include sensory appearances, and perhaps other cognitive states, as well as beliefs?
If reliabilism is true, then not all knowledge need be based on evidential grounds. For what is essential on that theory is that beliefs be reliably
formed, and it remains an open question whether there are cognitive
processes that are both reliable and involve no evidential grounds whatsoever. It is possible that memory works in this way. Logical intuition
might be another candidate for such a process.
To elaborate, we may define three possible kinds of cognitive process.
First, a process is inferential if it takes initial beliefs as inputs and generates
other beliefs as outputs by employing some inference-rule. Second, a
process is experiential if it takes sensory appearances or other kinds of
experience as inputs and generates beliefs on that basis, but not by
employing some inference from input states. I have argued that sensory
perception is best understood as a process of this kind. Finally, a cognitive process is groundless if it does not operate on reliable grounds at all.8
Now it is an empirical question whether human cognition includes this
third kind of process. As we have just noted, it is possible that memory
or logical intuition work this way. But that is not obvious, since these
might involve their own phenomenology, and in a way that is consistent
with their involving reliable grounds.9
For present purposes we may keep the question open, since the point
is only that the third kind of process is possible. If such a process is also
reliable, then it would confer positive epistemic status according to
simple reliabilism. This would provide a different kind of foundational
belief from what we have seen so far. Such beliefs would have positive
epistemic status or even amount to knowledge, even though they are
not grounded in any evidence whatsoever.10
8

Here we must make a distinction between cognitive states that serve as truth-indicating
grounds for a belief, and those involved in a beliefs production in some other manner.
E.g., some beliefs might serve merely as background assumptions governing cognition;
others might act as defeaters of counter-evidence.
9 For useful discussions of this question with regard to memory and logical intuition, see
Plantinga, Warrant and Proper Function, esp. pp. 5764 and pp. 103110.
10 The phenomenon of Hindsight might offer an example of the third kind of cognitive
process. Individuals with certain kinds of brain injury lack the normal phenomenology of

171

d. How Contingently Reliable Inferences Can Give Rise to Knowledge

Simple reliabilism accounts for contingently reliable inferences in a

straightforward manner. According to simple reliabilism, positive epistemic status is generated by reliable cognitive processes. If the use of an
inference constitutes a reliable process, it does not matter that the inference is only contingently reliable. Put differently, simple reliabilism
makes de facto reliability the grounds of positive epistemic status; it makes
no difference whether the inference is contingently reliable or necessarily
reliable. Neither does it matter whether a believer knows that her inference is reliable, or is justified in believing that it is, or is even aware that
it is. Much less does it matter that she knows this to be a necessary
truth.11
If simple reliabilism is correct, then this solves the problem of induction. That problem can be characterized as follows. In order to have
inductive knowledge one must first know that inductive reasoning is
reliable. But because any knowledge that inductive reasoning is reliable
must rely on such reasoning itself, one cannot know that inductive
reasoning is reliable. In short, you need to know that induction is
reliable, but you can't. But as James Van Cleve has argued, if reliabilism
is true then the skeptical argument is wrong on both counts: to have
inductive knowledge you need not know that induction is reliable, but
you can.12
First, one need not know that induction is reliable, because de facto
reliability is what matters, not knowledge of reliability. But second, one
can know that induction is reliable, because now an inductive argument
to that conclusion is available. Specifically, it is now possible to do an
induction on inductive arguments without falling into circularity. To

vision. However, testing shows that they can retain the ability to detect spatial locations of
objects and can even improve it with practice. The explanation is that information about
light hitting the retina is successfully carried to relevant parts of the brain, even though the
information now bypasses parts of the brain associated with the production of visual imagery. However, the evidence for a third kind of cognitive process is ambiguous here. Even
in cases of Hindsight people report an odd kind of phenomenology, although it is not that
of visual appearances.
11 For excellent discussions on the relationship between reliabilism and induction, see Plantinga, Warrant and Proper Function, esp. Ch. 7; Hilary Kornblith, Inductive Inference and Its

Natural Ground (Cambridge, MA: MIT Press, 1993); and James Van Cleve, "Reliability,
Justification, and the Problem of Induction," Midwest Studies in Philosophy 9 (1984), 555567.
12 Van Cleve, "Reliability, Justification, and the Problem of Induction."

172

see why this is so, remember that reliabilism countenances contingently

reliable inferences. Accordingly, these inferences need not include some
premise to the effect that nature is regular or that induction is reliable;
one will be able to argue directly from observed cases of induction to
induction in general. But if such premises are not needed in inductive
arguments, then an inductive argument that concludes that induction is
reliable will not involve circular reasoning that is, will not involve its
conclusion as a premise.
The following example of enumerative induction illustrates the point:
(El)

1. In the past, in many observed cases, enumerative induction has been

reliable.
2. Therefore, in general, enumerative induction is reliable.

This argument amounts to an enumerative induction on enumerative

induction. But the argument is not circular, because the conclusion is
not used as a premise, even implicitly.
Now (El) would be circular if enumerative inductions had to include
a premise about their reliability. (El) would also run into circularity
problems if inductive arguments required the regularity principle as a
premise. For then the justification of that principle would be at issue,
and any inductive argument in its favor would, by hypothesis, have to
include the principle itself. But if simple reliabilism is true then enumerative inductions require neither the regularity principle nor a premise
about their own reliability. For again, what matters is de facto reliability,
and enumerative induction is in fact reliable even without such premises.
That is because the world is in fact regular, and that is all that is needed
for enumerative induction to be de facto reliable.
We have seen that according to simple reliabilism reasoning need
only be contingently reliable to confer positive epistemic status on its
conclusions. We have also seen that this characteristic of the theory
allows it to answer the problem of induction. Before leaving this section
I want to make one more general point. According to simple reliabilism,
all evidential relations require only contingent reliability, and therefore
the relationship between sensory appearances and the beliefs they cause
need only be contingently reliable for perceptual knowledge. This is a
fortunate result, since it is implausible that this relation could be any more
than contingent.
Consider that the perceptual faculties of other species are very different from our own. As such, very different sensory appearances may

173

indicate the same physical realities for these creatures; the way that a tree
appears to a bat using sound-based sonar is nothing like the way the
same tree appears to a human being using light-based vision. Consider
also that the way things appear to us now could have indicated different
physical realities. For that matter, we could have been built so that visual
appearances reliably indicated nothing at all about objects in the world.
The point is that there is no necessary relation between the way things
appear and the way things are. That certain appearances reliably indicate
certain real properties is merely a contingent fact.
This means that there is an important analogy between perceptual
evidence and our evidence for unobserved matters of fact: namely, in
each case our evidence for the relevant kind of belief is only contingently
reliable. We saw in Chapter 6 that this creates a problem for any
adequate understanding of inductive evidence. On the one hand,
knowledge seems to require that the knower be sensitive to the reliability of her evidence. On the other hand, it is hard to see how one could
be sensitive to this if one's evidence is only contingently reliable. I am
now suggesting that exactly the same problem arises for a theory of
perceptual evidence as well.
Finally, we should note that the problem cannot be avoided by
thinking of sensory appearances as having representational content. This
is because it remains a contingent matter whether sensory appearances
having a particular representational content are reliable indications of the
truth of beliefs having that same content. For example, in normal circumstances and in normal perceivers, there seeming visually to be a tree
bearing apples by the brook is a reliable indication that there is a tree
bearing apples by the brook. But this is a contingent fact about human
visual perception rather than a necessary truth. Just as it could have been
the case that visual phenomenal qualia do not reliably indicate anything
about objects in the world, it could have been the case that thick visual
seemings do not either.
3. FROM PROCESSES TO VIRTUES

We have seen that simple reliabilism has much in its favor. The theory
explains why each of the conclusions in (a) through (d) is correct and
therefore explains why the skeptical arguments in Chapters 2 through 6
are mistaken. It is, therefore, the kind of theory that we are looking for.
But simple reliabilism is not quite right. It is on the right track, to be
sure, but it is in need of some revision to bring it into better shape.

174

One problem with simple reliabilism has already been mentioned in

Chapter 6, and again just now. Namely, the theory fails to explain how
knowers can be sensitive to the reliability of their evidence. In this
regard, simple reliabilism says that evidence must in fact be reliable for it
to give rise to knowledge and justified belief. There is no further requirement that the knower realize that her evidence is reliable, or that
she know this, or that she be sensitive to this in some other way. A
second, related problem with simple reliabilism may be called "The
Problem of Strange and Fleeting Processes." Put simply, simple reliabilism is too weak, because some reliable processes (strange and fleeting
ones) do not give rise to knowledge and justified belief. In the remainder
of this section I explore this second problem, and I argue that agent
reliabilism can solve it. In Part II of the chapter, I argue that agent
reliabilism solves the first problem as well.
Here are three examples of reliable yet epistemically inefficacious
processes. First, consider "The Case of the Epistemically Serendipitous
Lesion."13 Imagine that there is a rare sort of brain lesion, one effect of
which is to cause the victim to believe he has a brain lesion.
Suppose, then, that S suffers from this sort of disorder and accordingly believes
that he suffers from a brain lesion. Add that he has no evidence at all for this
belief: no symptoms of which he is aware, no testimony on the part of physicians
or other expert witnesses, nothing. (Add, if you like, that he has much evidence
against it; but then add also that the malfunction induced by the lesion makes it
impossible for him to take appropriate account of this evidence.) Then the
relevant [cognitive process] will certainly be highly reliable; but the resulting
belief that he has a brain lesion will have little by way of warrant for S.14
As a second example, consider "The Case of the Absurd Reasoner."
Having little understanding of biology, but fascinated by deterministic
explanations of human behavior, Charles reasons as follows. If he witnesses two people ordering the same fruit drink on the same day, he
concludes on that basis that they are genetically related. As it turns out,
his whimsical reasoning process is perfectly reliable, since everyone is
genetically related.
Finally, consider "The Case of the Helpful Demon." Rene thinks he
can beat the roulette tables with a system he has devised. Reasoning
according to the Gambler's Fallacy, he believes that numbers which
13 This case is from Alvin Plantinga, "Positive Epistemic Status and Proper Function," p. 28,
and from Warrant: The Current Debate (Oxford: Oxford University Press, 1993), p. 199.
14

Plantinga, Warrant: The Current Debate, p. 199.

175

have not come up for long strings are more likely to come up next.
However, unlike Descartes' demon victim, our Rene has a demon
helper: every time Rene forms a belief that a number will come up
next, the demon arranges reality so as to make the belief come out true.
Given the ever present interventions of the helpful demon, Rene's
belief-forming process is highly reliable. But this is because the world is
made to conform to Rene's beliefs, rather than because Rene's beliefs
conform to the world.
These examples of strange but reliable processes show that simple
reliabilism is too weak. More exactly, it would seem that not just any
reliable cognitive process can give rise to positive epistemic status. That,
in turn, raises the question of what the appropriate restriction should be.
How can simple reliabilism be revised so as to exclude these strange
cases as counting for knowledge and justified belief? The answer is
suggested in the following passage from Sosa, where he is considering
how a certain move in ethics might fruitfully be applied to epistemology:
In what sense is the doctor attending Frau Hitler justified in performing an
action that brings with it far less value than one of its accessible alternatives?
According to one promising idea, the key is to be found in the rules that he
embodies through stable dispositions. His action is the result of certain stable
virtues, and there are no equally virtuous alternative dispositions that, given his
cognitive limitations, he might have embodied with equal or better total consequences, and that would have led him to infanticide in the circumstances. The
important move for our purpose is the stratification of justification. Primary
justification attaches to virtues and other dispositions, to stable dispositions to
act, through their greater contribution of value when compared with alternatives. Secondary justification attaches to particular acts in virtue of their source
in virtues or other such justified dispositions.
The same strategy may also prove fruitful in epistemology. Here primary
justification would apply to intellectual virtues, to stable dispositions for belief
acquisition, through their greater contribution toward getting us to the truth.
Secondary justification would then attach to particular beliefs in virtue of their
source in intellectual virtues or other such justified dispositions.15
Relevant to present purposes is Sosa's suggestion for a restriction on
reliable cognitive processes: it is those processes that have their bases in
the stable and successful dispositions of the believer that are relevant for
knowledge and justification. Just as the moral rightness of an action can
be understood in terms of the stable dispositions or character of the moral
15 Sosa, "Raft and Pyramid," pp. 167-168.

176

agent, the epistemic rightness of a belief can be understood in terms of

the intellectual character of the cognizer.
Sosa names this approach "virtue epistemology," since the stable and
successful dispositions of a person are appropriately understood as virtues.
For example, it may be one's faculty of sight operating in good light that
generates one's belief in the whiteness and roundness of a facing snowball. Is
possession of such a faculty a "virtue"? Not in the narrow Aristotelian sense, of
course, since it is no disposition to make deliberate choices. But there is a
broader sense of "virtue," still Greek, in which anything with a function
natural or artificial does have virtues. The eye does, after all, have its virtues,
and so does a knife. And if we include grasping the truth about one's environment among the proper ends of a human being, then the faculty of sight would
seem in a broad sense a virtue in human beings; and if grasping the truth is an
intellectual matter then that virtue is also in a straightforward sense an intellectual virtue.16
In this regard Sosa cites Book 1 of Plato's Republic, where Plato says that
vision is the virtue of the eyes and hearing the virtue of the ears.17 But
whatever the terminology we adopt, the important point is that the
move solves the problem of strange and fleeting processes. For the
cognitive faculties and habits of a believer are neither strange nor fleeting. They are not strange, because they make up the person's intellectual
character they are part of what make her the person that she is. They
are not fleeting, because faculties and habits by definition are stable
dispositions they are not the kind of thing a person can adopt on a
whim or engage in an irregular fashion.
On the present view, knowledge and justified belief are grounded in
stable and reliable cognitive character. Such character may include both
a person's natural cognitive faculties and her acquired habits of thought.
Accordingly, innate vision gives rise to knowledge if it is reliably accurate. But so can acquired skills of perception and acquired methods of
inquiry, including those involving highly specialized training or even
advanced technology. So long as such habits are both stable and successful, they make up the kind of character that gives rise to knowledge.
We may now explicitly revise simple reliabilism as follows: A belief p
has positive epistemic status for a person S just in case S's believing p
results from stable and reliable dispositions that make up S's cognitive
character. I will call this position "agent reliabilism," since the disposi16

Sosa, Knowledge in Perspective, p. 271.

17 For example, at 342 and 352.

177

tions referred to in the definition are dispositions of cognitive agents.

Accordingly, the definition makes the reliability of agents central to the
analysis of knowledge and justified belief.
I have been arguing that agent reliabilism solves the problem of
strange and fleeting reliable processes, and as such constitutes an improvement over simple reliabilism. We may briefly note that the position
has the same advantage over several other versions of reliabilism, such as
method reliabilism, social practice reliabilism, and evidence reliabilism.
For with each of these other views, it is possible to imagine cases where
the specified seat of reliability is fleeting, and therefore cases where the
view rules incorrectly that there is knowledge. For example, we can
imagine cases where a person bases her belief on evidence that is perfectly reliable, but where there is no corresponding disposition to form
beliefs on evidence of a relevant kind. In the isolated case where the
person happens to use evidence that is in fact reliable, it seems incorrect
to call the result knowledge. Similarly, we can imagine that a person
adopts a perfectly reliable method, but on a whim. Where there is no
corresponding disposition to employ the method, it seems wrong to say
that the person obtains knowledge by employing it in an isolated incident. This is because in each case the method, or the practice, or the
adoption of particular evidence amounts to a fleeting process. The belief
is formed by a reliable process, but there is a sense in which the reliability
of the process is accidental from the believer's point of view.
In each case the problem can be solved if the view is revised so as to
require a disposition to use the process in question. But then the view
would become a version of agent reliabilism. Since the relevant dispositions would be properties of agents, such a revision would have the
effect of requiring agent reliability. For example, consider William Alston's social practice reliabilism.18 If social practices are defined independently of people's dispositions to engage in them, then the view is
subject to counterexamples as just described: it will be possible that
agents adopt practices in isolated incidents, on a whim, and so forth. But
if social practices are by definition dispositions to act in certain ways, then
it will be impossible for an agent to engage in a reliable social practice
without having a reliable cognitive character.19 On this second version

18 William Alston, "A 'Doxastic Practice' Approach to Epistemology," in Marjorie Clay and
Keith Lehrer, eds., Knowledge and Skepticism (Boulder: Westview, 1989); and Alston, Perceiving God (Ithaca: Cornell University Press, 1991), esp. Ch. 4.
19 It is clear that Alston means to define a practice in the second way. He writes, "A doxastic

178

of the view, social practice reliabilism becomes a version of agent reliabilism, but with further details about the nature of cognitive dispositions.
In effect, it says that knowledge and justified belief are grounded in the
reliable dispositions of agents, and that these are to be understood in
terms of reliable social practices.
This means that agent reliabilism is sufficiently general to admit of
many versions, depending primarily on how one fills in the details
regarding the nature of reliable character. Understood thus generally, the
position is widely popular in contemporary analytic epistemology.20 Historically, the theory has its roots in the AristotelianThomistic conception of the person as the seat of intellectual powers, and in Reid's antiskeptical faculty epistemology.21
But despite current popularity and excellent historical credentials,
agent reliabilism is by no means uncontroversial. In Part II of the chapter, I want to look at one major issue facing the position and to address
some objections that arise within that context. The issue is how to
address the concern of (e) at the beginning of the chapter that is,
practice can be thought of as a system or constellation of dispositions or habits, or to use a
currently fashionable term, 'mechanisms,' each of which yields a belief as output that is
related in a certain way to an 'input'." Perceiving God, p. 153.
20 In addition to Sosa and Alston, Plantinga has developed a detailed position around the thesis
that knowledge is grounded in the stable and reliable faculties of believers, where "faculties"
are understood to include both natural and acquired dispositions. For the most recent
statement of Plantinga's position, see Warrant and Proper Function. Goldman, in his most
recent work, moves away from his earlier process reliabilism to a version of agent reliabilism,
calling his own view a "virtues theory" and comparing it sympathetically with Sosa's. See
"Epistemic Folkways and Scientific Epistemology" in Liaisons, esp. p. 163. Each of these
authors defends a different position. My claim here is that all are versions of the core position
that I call "agent reliabilism." Linda Zagzebski also understands knowledge and justified
belief as deriving from intellectual virtue, making the analogy to Aristotelian virtue ethics
even stronger than Sosa would have it. Early versions of Zagzebski's position were instances
of agent reliabilism. However, in the latest statement of her views Zagzebski explicitly
rejects the requirement that knowers have a reliable cognitive character. Therefore, although her latest position is a version of virtue epistemology, it is not a version of agent
reliabilism. For the earlier view, see Linda Zagzebski, "Religious Knowledge and the Virtues of the Mind," in Linda Zagzebski, ed., Rational Faith: Catholic Responses to Reformed

Epistemology (Notre Dame: University of Notre Dame Press, 1993), and "Intellectual Virtue
in Religious Epistemology," in Elizabeth Radcliffe and Carol White, eds. Faith in Theory
and Practice: Essays on Justifying Religious Belief (La Salle: Open Court, 1993). For the later

view, see Virtues of the Mind (Cambridge: Cambridge University Press, 1996). For a criticism
of Zagzebski's later position, see my "Two Kinds of Intellectual Virtue," Philosophy and
21

Phenomenological Research, forthcoming.

See Plantinga, Warrant: The Current Debate, p. v and pp. 183184, and Jonathan Kvanvig's
useful discussion in The Intellectual Virtues and the Life of the Mind (Savage, MD: Rowman

and Littlefield, 1992), esp. pp. 11-27.

179

how to explain the kind of sensitivity to one's own reliability that is

required for knowledge. Another way to put the question is in terms of
subjective justification: In what sense must knowledge be well formed
from the knower's own point of view, as opposed to objectively well
formed, or de facto reliable? My claim will be that agent reliabilism has
the resources to successfully address these issues.

II. Agent Reliabilism and the Question of Subjective

Justification
Agent reliabilism makes de facto reliability an important condition on
knowledge. In this sense a belief counts as knowledge only if it is
objectively well formed. But it would seem that knowledge has to be
subjectively appropriate as well as objectively reliable. As we have noted,
one way in which the subjective side of knowledge manifests itself is in
regard to the concern raised in (e): inductive evidence, and evidential
grounds in general, seem capable of generating knowledge only if the
believer has some insight into the reliability of those grounds. The
novice mathematician who does not "see" that her axioms entail her
theorem, we said, does not know the theorem on that account. So far
we have not seen how agent reliabilism can account for this strong
intuition regarding the requirements for knowledge.
The issue of subjective appropriateness is often discussed in terms of
epistemic justification. The question then becomes: How can agent
reliabilism understand epistemic justification? The question threatens to
turn into an objection. For our concept of knowledge does seem to
involve a subjective requirement, and if agent reliabilism cannot account
for this, then the view fails on that ground. But at this point the
objection that threatens has not been adequately stated. For we have not
articulated well just what kind of subjective justification is required for
knowledge. In what sense, exactly, must knowledge be well formed
from the knower's point of view? Or, in terms of (e), in what sense
must knowers be sensitive to their own reliability?
In Sections II. 1 through II.3 I review several senses of subjective
justification which should not be required for knowledge. In each case,
requiring this kind of sensitivity on the part of knowers would make
knowledge impossible. I then argue in Section II.4 that agent reliabilism
has the resources for understanding a different sense of subjective justification, and that this sort of justification is required for knowledge. In

180

this way, agent reliabilism explains (e) as well as (a) through (d), and in
general accounts for the intuition that knowledge must be subjectively
appropriate as well as objectively reliable.
1. KNOWING THAT ONE KNOWS
One way to articulate the problem of subjective justification is in terms
of the internalismexternalism debate in analytic epistemology. Internalism is the position that the conditions for justification must be appropriately internal to the knower's perspective. Roughly, something is internal to S's perspective so long as S is aware of it or could be aware of it
merely by reflecting. Externalism is simply the denial of internalism,
holding that the conditions for justification need not be within the
knower's perspective. In this terminology, agent reliabilism is a form of
externalism. For example, whether a person's perception is functioning
reliably might not be something of which he is aware or could be aware
just on reflection. The present issue is whether externalist theories like
agent reliabilism can account for the kind of subjective justification that
knowledge requires.
In the following passages Bonjour reflects on the significance of cases
like that of Norman the clairvoyant. Due to his highly reliable power of
clairvoyance, Norman believes that the president is in New York City
despite the fact that Norman has considerable evidence otherwise, and
has no suspicion that he has the power of clairvoyance. By hypothesis,
Norman's belief about the president is caused by a reliable cognitive
process: his little-used and completely unrecognized clairvoyance. But
according to Bonjour, it is counterintuitive to suppose that his belief is
justified, and neither do we want to say that it is a case of knowledge:22
We are now face-to-face with the fundamental and obvious intuitive
problem with externalism: why should the mere fact that such an external
relation obtains mean that Norman's belief is epistemically justified when the
relation in question is entirely outside his ken? As I noted earlier, it is clear that
one who knew that [the externalist's] criterion was satisfied would be in a
position to construct a simple and quite cogent justifying argument for the belief
that the President is in New York City: if Norman has property H (being a
completely reliable clairvoyant under the existing conditions and arriving at the
belief on that basis), then he holds the belief in question only if it is true;
Norman does have property H and does hold the belief in question; therefore,
22 The example is from Bonjour, The Structure of Empirical Knowledge, p. 41.

181

the belief is true. Such an external observer, having constructed this justifying
argument, would be thereby in a position to justify his own acceptance of
a belief with the same content. . . . But none of this seems in fact to
justify Norman's own acceptance of the belief, for Norman, unlike the hypothetical external observer is ex hypothesi not in a position to employ this argument. . . . 23
One reason why externalism may seem initially plausible is that if the external
relation in question genuinely obtains, then Norman will in fact not go wrong
in accepting the belief, and it is, in a sense, not an accident that this is so. . . .
But how is this supposed to justify Norman's belief? From his subjective perspective, it is an accident that the belief is true. And the suggestion here is that
the rationality or justifiability of Norman's belief should be judged from Norman's own perspective rather than from one which is unavailable to him.24

Bonjour's complaint is that Norman's reliability is "outside his ken"

or external to his "subjective perspective": thus the name "externalism"
for theories of justification and knowledge of this general kind. But in
what sense is Norman's reliability external to his perspective, and why is
that a problem? This is what we have to get straight if we are to judge
the force of Bonjour's objection against externalism in general and agent
reliabilism in particular.
We might interpret Bonjour's argument as follows: First, knowledge
is transparent: no one knows unless he knows that he knows, or at least
can know that he knows. But if externalism is true, then no one ever
does know that he knows. And therefore, if externalism is true, the
skeptic is right when he says that no one knows.
Whether or not this is what Bonjour intends, there is something
natural about the argument. Many philosophers have explicitly affirmed
the principle that knowledge entails knowing that one knows, and such
a position is at least implied in many reactions to externalism.25 But for
all its naturalness the argument is confused. For one thing, there is no
reason why one cannot know that she knows on agent reliabilism. On
that theory, one knows that one knows in the same way one knows
anything by believing so as the result of reliable cognitive character.
Another problem with the current argument is that its first premise is
false. That premise expresses the principle that one knows only if one
25

23 Ibid., p. 43.
24 Ibid., pp. 43-44.
For endorsement of this principle and ones like it see Roderick Chisholm, "Knowing That
O n e Knows," in The Foundations of Knowing; Carl Ginet, Knowledge, Perception and Memory

(Dordrecht: Reidel, 1975); and H. A. Prichard, Knowledge and Perception (Oxford: Clarendon Press, 1950).

182

knows that one knows (Kp only if KKp). But this principle cannot be
right, since it leads directly to total skepticism; if the principle were
correct, then I could not know even that I exist, and no matter what
theory of knowledge is correct. This is because, according to the principle, I know that I exist only if I know that I know that I exist. But by
a second application of that principle, I know that I know that I exist
only if I know that I know that I know that I exist. Kp implies KKp,
which in turn implies KKKp, which implies KKKKp, which implies
KKKKKp, and so forth. But sooner or later I will reach propositions
which I cannot even grasp, much less know. So if the principle is right,
I do not know even that I exist. But of course I do know that I exist,
and so the principle does not state an actual condition on knowledge. 26
Let us consider a second interpretation of Bonjour's objection. The
passages from Bonjour strongly suggest the following argument: In order
to know, it is not sufficient that one's belief be reliably formed; one
must also know that one's belief is reliably formed. But if this is Bonjour's
meaning, then his objection is again misguided. For like the principle
just discussed, such a requirement generates an infinite regress and leads
directly to skepticism. To see why, consider the conditions for knowledge that the objection suggests:
S knows that p only if
i. S believes that p is true;
ii. p is true;
iii. S's belief is reliably formed; and
iv. S knows that her belief is reliably formed.
The problem here is with condition (iv). Not only does it make the
conditions for knowledge circular it makes them impossible to fulfill.
For on this suggestion S must know p1: that p is reliably formed. But
knowing this will generate (through (iv)) the condition that S know p":
that p' is reliably formed. And the regress goes on. But since no one is
capable of knowing an infinite number of increasingly complex propositions, a total skepticism is entailed. What this shows is not that total
skepticism is true, but that the present objection to externalist theories is
misguided.
26 The same argument applies if we take a weaker version of the principle: that one knows
only if one can know that one knows. For more extended arguments that this principle
and ones like it should be rejected, see William Alston, "Level Confusions in Epistemology," Midwest Studies in Philosophy 5 (1980): 135150; and my "Internalism and Epistemically Responsible Belief," Synthese 85 (1990): 245-277.

183

Bonjour does not say so, but someone might think that the need to
know about one's own reliability is grounded in the following principle:
To have knowledge, one must know that the conditions for having
knowledge are fulfilled. This is another version of the idea that knowledge is transparent, so I will call it the "transparency principle." Since
the transparency of knowledge has been such an attractive position, it is
worth noting that the principle is not merely false but borders on being
incoherent. For to say that conditions X, Y, and Z are sufficient conditions for knowing p is just to say that nothing else is required for knowing
p. To go on to require that one must also know that conditions X, Y,
and Z are satisfied is to give up the original contention that X, Y, and
Z are sufficient conditions.27 Of course, one might think that knowing
p requires knowing that the conditions for knowing p are satisfied,
because one thinks that knowing implies knowing that one knows. But
we have already seen that this last assumption is false.
Finally, it is important to note that the transparency principle would
have disastrous consequences for any theory of knowledge whatsoever,
and not just for agent reliabilism. Suppose, for example, that we adopt a
crude relativism, endorsing the theory that knowledge is what society
agrees upon. According to the transparency principle, one can know
that p is true only if one also knows that society agrees that p is true.
And by a second application of the principle, one must also know that
society agrees that society agrees that p, and so on ad infinitum. Again,
since it is impossible to believe an infinite number of increasingly complex propositions, the transparency principle directly entails total skepticism, even for the theory that knowledge is what society agrees upon.
Clearly the principle is either incoherent or false.

2. UNDERSTANDING THAT ONE KNOWS

So far we have seen three ways not to understand the subjective requirement on knowledge. Before moving on to something better, I want to
look at a related objection against agent reliabilism. The objection is
raised by Stroud and may be stated in the form of the following argument: If agent reliabilism is true, then no one can understand that the
conditions for having knowledge are satisfied, since no one can reason
without circularity that one's cognitive faculties and habits are reliable.
27

Here I am indebted to James Van Cleve. See his "Foundationalism, Epistemic Principles,
and the Cartesian Circle," Philosophical Review 88 (1979): 77.

184

But if one cannot understand that the conditions for having knowledge
are satisfied, then one cannot understand that one has knowledge.
Therefore, agent reliabilism does not give us an adequate understanding
of our knowledge. Moreover, since the task of epistemology is to understand how we know what we know, agent reliabilism cannot be a
satisfactory epistemology.28 More formally, we have the following argument.
(STR)

1. A person can understand that she has knowledge only if she

understands that she fulfills the conditions for having knowledge.
2. If agent reliabilism is true, then a condition for knowing anything
is that one's cognitive powers (i.e., faculties and habits) are reliable.
3. Therefore, if agent reliabilism is true, then one can understand
that she has knowledge only if she understands that her cognitive
powers are reliable. (1,2)
4. But if agent reliabilism is true, then one cannot understand that
one's cognitive powers are reliable.
5. Therefore, if agent reliabilism is true, then no one can understand
that she has knowledge. (3,4)

Remember that the support for premise (4) is as follows:

(STR4)

1. If agent reliabilism is true, then one can understand that one's

cognitive powers are reliable only by reasoning in a circle.
2. Understanding cannot be based on circular reasoning.
3. Therefore, if agent reliabilism is true, then one cannot understand that one's cognitive powers are reliable. (1,2)

Like Bonjour's, Stroud's argument has a certain intuitive pull. However, a closer look reveals that it has several problems. First, premise (4)
of (STR) is false. According to agent reliabilism, one can know that
one's cognitive powers are reliable so long as one's belief to this effect is
itself grounded in reliable cognitive powers. But then the agent reliabilist
can say the same thing about understanding: so long as one's belief is in
fact well formed by virtuous faculties and habits, one understands what
one believes. Stroud insists that understanding involves having reasons,
but this added requirement on understanding is not a problem for agent
reliabilism. On any plausible account, one will have reasoned to the
28

See Barry Stroud, "Understanding Human Knowledge in General," in Clay and Lehrer;
and again in "Scepticism, 'Externalism,' and the Goal of Epistemology," Proceedings of the
Aristotelean Society (1994): 291-307.

185

belief that one's powers are reliable. So long as the reasoning is in fact
virtuous, understanding results.
(STR4) charges that any such reasoning would be circular, but that
too seems to involve a confusion. Specifically, to understand that one's
cognitive powers are reliable one must use those very powers, and in
particular one must use one's reasoning faculties. But one need not
reason from the reliability of one's powers as a premise. That would be
circular, reasoning to a conclusion that one is reliable from a premise
that one is reliable. But agent reliabilism requires no such circles.
For example, to know that your vision is in good working order you
might visit an eye doctor. And certainly your vision will be involved in
learning from the doctor that your vision is in fact in good working
order you will need to read a report, or at least identify the doctor as
the person you are talking to. So will reasoning be involved, since you
will have to infer the reliability of your vision from the evidence gathered. But at no time during this process would you reason from the
assumption that your vision and reasoning are reliable. Again, you learn
that your vision is reliable by using it and by using your reason, not by
reasoning from assumptions about the reliability of your vision or your
reason. Similar things can be said about other cognitive powers. For
example, cognitive scientists learn about the reliability of human inference by inferring their conclusions from relevant observations. But that
is the point: they infer the reliability of inference from observations, not
from some premise to the effect that inference is reliable.
Premise circularity is vicious, but it is not involved in determining
the reliability of one's cognitive faculties and habits. What about "faculty
circularity," or the process of using one's faculties to determine the
reliability of those same faculties? First, we should note that in general
one cannot escape using one's cognitive powers to determine the reliability of those very powers for, by definition, one's cognitive powers
are that by which one cognizes. But to see that this fact is benign,
consider that it is the condition of any cognizer whatsoever. Although
skepticism is presented by Stroud and others as a problem for the human
condition, it is true of any cognizer, God included, that one must think
by the ways that one thinks.29 Finally, if Stroud's objection were effective against agent reliabilism, a similar line of reasoning would preclude
understanding that we know on any theory of knowledge whatsoever.
29

I am indebted to Keith DeRose for this last point, which he made in a paper presented at
Fordham University.

186

For no matter what conditions on knowledge are laid down by a theory,

we will have to use our thinking to determine that our thinking satisfies
those conditions.

3. SOSA'S PERSPECTIVISM

We have seen three ways not to understand subjective justification and

a related misplaced objection from Stroud. Nevertheless there seems to
be something right about Bonjour's general point, acknowledged in (e)
at the start of the chapter: Even if knowledge does not require knowing
that one knows, or knowing that one's cognitive powers are reliable, it
does seem to require some kind of sensitivity to one's own reliability.
As Bonjour put the point, one's belief ought to be well formed from
one's "subjective perspective." The trick is in how to understand that
notion, if not in any of the ways we have considered so far.
Sosa is one agent reliabilist who has tried to tackle exactly this problem. According to him subjective justification is to be understood in
terms of a reliable perspective on one's own faculties. The central idea
is that for reflective knowledge (as opposed to mere "animal knowledge"), one must have a true grasp of the fact that one's belief is
grounded in a reliable cognitive faculty. This grasp must in turn result
from a faculty of faculties, which gives rise to the required epistemic
perspective:
For one is able to boost one's justification in favor of P if one can see one's
belief of P as in a field F and in circumstances C, such that one has a faculty (a
competence or aptitude) to believe correctly in field F when in conditions C.
. . . One thereby attributes to oneself some intrinsic state such that when there
arises a question in field F and one is in conditions C, that intrinsic state adjusts
one's belief to the facts in that field so that one always or very generally believes
correctly.30
According to Sosa, to "see" one's belief of P as in a field and circumstances is to have true beliefs to that effect, where those true beliefs are
themselves products of a cognitive virtue.
It is clear how this position can be applied to the case of Norman
mentioned in Section II. 1. Sosa can say that Norman has animal knowledge but not reflective knowledge, since Norman does not see himself
30 Sosa, Knowledge in Perspective, p. 282. Sosa develops this strategy in "Intellectual Virtue in
Perspective" and in "Reliabilism and Intellectual Virtue," both in Knowledge in Perspective.

187

as believing from a reliable faculty of clairvoyance. Further, Sosa can say

that Norman's belief is reflectively unjustified, since his belief actually
conflicts with his epistemic perspective on his faculties. Remember we
said that Norman has good evidence against his belief that the president
is in New York City, and so from his own perspective he ought not to
believe what he does.
Therefore Sosa's view correctly handles the cases of Norman the
clairvoyant: our intuition is that Norman does not have knowledge, and
Sosa's position has that result. However, a different problem arises for
the view, concerning its psychological plausibility. Specifically, it would
seem that we seldom have the perspective on our faculties that Sosa
requires for justification. In the typical case we have no beliefs at all
about the source of our beliefs. Much less do we have beliefs about our
reliability in particular fields and circumstances, or about whether a belief
is in such a field, or we in such circumstances.
Or at least we do not have occurrent beliefs like these. Is it plausible
that we typically have such beliefs dispositionally? Where we do have
such a dispositional perspective, the field and circumstances that our
perspective specifies are probably the wrong ones. To the extent that I
dispositionally attribute to myself certain cognitive faculties, those faculties are specified much too broadly to be of any use. For example,
consider my belief that there is a glass of water on the table. In the
typical case I have no occurrent beliefs about the source of this belief in
a given cognitive virtue. Let me now consider any dispositional beliefs I
might have. After considering the issue for a moment, it occurs to me
that my belief about the glass is the result of sight. But if you ask me to
get very specific about a field of propositions F, or a set of circumstances
C, such that I am highly reliable in that field when in those circumstances, I am at a loss. I simply do not have very specific beliefs in this
area, nor is it plausible that such beliefs are available dispositionally if
only I think about it a little more. But if I do not have such a perspective
in the typical case, then on Sosa's criteria I will not have justification
and knowledge either. As we have seen, psychologically implausible
conditions on knowledge lead directly to skepticism.
In reply to a similar objection Sosa writes the following:
[A person judging shapes on a screen] is justified well enough in taking it that,
in his circumstances, what looks to have a certain shape does have that shape.
He implicitly trusts that connection, as is revealed by his inferential 'habit' of
moving from experiencing the look to believing the seen object to have the
corresponding shape. So the 'belief involved is a highly implicit belief, mani-

188

fested chiefly in such a 'habit.' Habits, too, can be assessed as intellectually or

cognitively proper or improper, 'justified' or not. And they can even be assessed
for 'correctness.' Thus the habit of moving from 'looks round' to 'is round' is
strictly correct if, in the relevant circumstances, anything that looked round
would in fact be round (and we can view the 'habit,' alternatively, as a belief
that, in the circumstances, anything that looked round would in fact be
round). . . .
Since 'inferential habits' can be assessed as 'correct' or 'incorrect' in the senses
specified, and as 'justified' or 'unjustified,' (not just any habit, no matter how
acquired, being proper), therefore there is some motivation to view such habits
as implicit beliefs (that can be thus correct and justified) in the corresponding
conditionals, as suggested above.31

Sosa's strategy for meeting the psychological implausibility objection is

to stress the implicit nature of the perspective on our faculties. The
knower's perspective is "highly implicit" and is "manifested chiefly" in
his habits of inference rather than in any explicit acceptance or judgment
on the knower's part. But notice an ambiguity in Sosa's notion of a
"perspective" and then a shift in his position. The ambiguity is between
(a) dispositional beliefs and (b) dispositions to believe. In the passages
quoted Sosa begins by understanding our epistemic perspective in the
first way, but ends by understanding it in the second way. Let us take a
closer look.
In the beginning of the first passage the required perspective on our
faculties is constituted by highly implicit dispositional beliefs. These
beliefs are "revealed by" and "manifested in" our habits of inference.
But at the end of the first passage and in the second, our perspective on
our faculties is identified with the habits. Sosa writes, "[W]e can view the
'habit,' alternatively, as a belief," and "[T]here is some motivation to
view such habits as implicit beliefs." Sosa talks as if these alternatives are
equivalent, or at least equivalent for relevant purposes. But they are not.
To have a belief, even an implicit one, is not the same as having a
disposition to form beliefs.32
The distinction is important because it is not at all implausible that
we have habits of inference, or dispositions to form certain beliefs on
31 Ernest Sosa, "Virtue Perspectivism: A Response to Foley and Fumerton," in Enrique
Villanueva, ed., Philosophical Issues 5: Truth and Rationality (Atascadero CA: Ridgeview,

1994), pp. 4243. Sosa is responding to an objection from Richard Foley, "The Epistemology of Sosa," same volume.
32 For an extended discussion of the present distinction, see Audi, "Dispositional Beliefs and
Dispositions to Believe."

189

certain bases and in certain circumstances. But it is highly implausible

that we have beliefs about the details of belief production. For even
implicit beliefs are intentional they are beliefs about something. Accordingly, the view that we have an implicit perspective on our faculties
implies that in some sense we think about our first-level beliefs, the
details of how they are produced, and whether beliefs so produced are
reliable in the circumstances in which we find ourselves. As we have
seen, there is simply no evidence that such thinking takes place in the
typical case, and plenty of evidence against it.
Therefore we should conclude that Sosa's perspectivism does not
capture what it is to be subjectively justified not if a perspective is
intentional, so that a perspective on one's faculties is a perspective about
those faculties. But we said that the passages from Sosa contained a shift
in position a shift from thinking of our perspective as dispositional
beliefs to thinking of it as dispositions to believe. It seems to me that the
second position is the more fruitful one. Let us pursue it further.
4. FROM PERSPECTIVES TO DISPOSITIONS
My proposal is that we can understand subjective justification in terms
of the knower's dispositions to believe. More exactly, subjective justification can be understood in terms of the dispositions a person manifests
when she is thinking conscientiously when she is trying to believe
what is true as opposed to what is convenient, or comforting, or fashionable. Something like this is perhaps Sosa's considered position, but
then we should drop the terminology of "epistemic perspective," "reflective knowledge," and "seeing one's belief," since all of this implies
intentional attitudes that we do not typically have that is, thoughts
about one's beliefs and the sources of those beliefs. None of that is
implied here.
The proposal then is this:
(VJ) A belief p is subjectively justified for a person S if and only if S's
believing p is grounded in the cognitive dispositions that S manifests when S is thinking conscientiously.
A few comments are in order. First, by "thinking conscientiously" I do
not mean thinking with an explicitly voiced purpose of finding out the
truth. Neither do I mean thinking with this as one's sole purpose.
Rather, I intend the usual state that most people are in as a kind of

190

default mode the state of trying to form one's beliefs accurately. One
might say "thinking honestly" instead, and this is intended to oppose
such modes as trying to comfort oneself, trying to get attention, and
being pigheaded. The latter, we might say, reflect epistemic vices rather
than virtues.
Second, (VJ) does not equate justified belief with conscientious belief.
This is because a person might be conscientious in believing that something is true without manifesting the dispositions she usually does in
conscientious thinking. For example, a father might sincerely try to
discover the truth about a son accused of bad behavior and yet nevertheless violate norms of good reasoning that he would usually manifest
when thinking conscientiously: in this case his good judgment is undermined by affection for his child, and despite himself. In a similar fashion,
someone might try too hard to get at the truth, thereby failing to
manifest the good habits that he typically does. In such cases we say that
the person outthinks himself, much as players can press too hard in
sports.
Third, the dispositions that a person manifests when she is thinking
conscientiously are stable properties of her character and are therefore in
an important sense hers. Accordingly, in an important sense a belief
produced from such dispositions will be well formed from the person's
own point of view. The current proposal is that this fact solves the
problem posed in (e) at the beginning of the chapter: namely, it explains
in what sense knowers must be sensitive to the reliability of their own
inferences. The relevant sense is that inferences that generate knowledge
are appropriately grounded in the knower's cognitive character specifically, the character she manifests when thinking conscientiously. This
same line of reasoning addresses the requirement that knowers be sensitive to the reliability of their evidence in general, whether or not that
evidence is conceived as inferential. Just as knowers are disposed to form
beliefs on the basis of inferences from other beliefs, they are disposed to
form perceptual beliefs directly on the basis of sensory appearances. The
fact that they do this in some ways and not others constitutes a kind of
sensitivity to the reliability of their evidence, whether that evidence be
inferential or experiential.
This way of understanding subjective justification is quite natural on
a virtue theory, since intellectual virtues are stable dispositions constituting a person's cognitive character. A long tradition suggests that virtuous
belief is also properly motivated. This is captured in the present proposal

191

by the reference to conscientious thinking. A person's intellectual virtues

are identified with the dispositions she manifests when she is sincerely
trying to believe what is true.
Finally, the view proposed in (VJ) does not have the problem we saw
for the first interpretation of Sosa's perspectivism. This is because we
require for justification that a belief be grounded in one's character, but
not that one have beliefs about one's character. In this respect good
thinking is like good hitting; when a baseball player swings the bat, he
manifests dispositions that are a product of both innate capacities and
acquired learning. If he is a good hitter then these dispositions will
generate success in relevant conditions. But even so, the most successful
player need not be a good coach; he may not have any beliefs at all, or
may even have incorrect beliefs, about the nature and character of the
dispositions that he himself manifests when batting conscientiously.
What makes for a good hitter is that he hits well, and what makes for a
good thinker is that he thinks well. Accordingly, (VJ) makes no requirement concerning a perspective on one's faculties or character, implicit
or otherwise, and for this reason escapes the charge of psychological
implausibility.

5. THE PLACE OF EPISTEMIC NORMS

We may further explore the present proposal by considering the question of epistemic norms. A natural way to understand subjective justification is in terms of conformance to the norms of good thinking, and
the question arises whether (VJ) is consistent with this. The idea that
subjective justification can be understood in terms of conformance to
norms is suggested by John Pollock:
Norms are general descriptions of the circumstances under which various kinds
of normative judgments are correct. Epistemic norms are norms describing
when it is epistemically permissible to hold various beliefs. A belief is justified if
and only if it is licensed by correct epistemic norms. . . . The concept of epistemic justification can be explained by explaining the nature and origin of the
epistemic norms that govern our reasoning.33
Pollock stresses that epistemic norms govern our behavior without our
having beliefs about them. Like most other action-guiding norms, epi33 Pollock, Contemporary Theories of Knowledge, pp. 124-125.

192

stemic norms influence our behavior without our thinking about them
at all:
Now let us apply this to epistemic norms. We know how to reason. That means
that under various circumstances we know what to do in reasoning. This can
be described equivalently by saying that we know what we should do. Our
epistemic norms are just the norms that describe this procedural knowledge. . . .
They describe an internalized pattern of behavior that we automatically follow
in reasoning, in the same way we automatically follow a pattern in bicycle
riding.34
We are presently considering whether this view of justification is
consistent with (VJ). In another place I have argued that it is, suggesting
that the dispositions a person manifests when she is thinking conscientiously have their basis in her conformance to countenanced norms.35
On this view justified belief is grounded in cognitive dispositions, which
in turn are grounded in conformance to correct epistemic norms. But
although this position is possibly correct, it now seems to me that
whether it is correct is an empirical question. Specifically, it is an empirical question whether cognition is governed by norms, and therefore an
empirical question whether cognitive virtues have their bases in conformance to norms.
This issue arises due to the nature of norms. In particular, I am
thinking of norms as rules that govern behavior, as opposed to rules that
merely describe behavior. As such, norms in this sense play a causal role
in the production of behavior. We may think of such norms as conditionals with antecedents stating certain conditions and consequents directing behavior in those conditions. For example, a rule of the relevant
kind might have the structure, "In conditions C, believe B." Furthermore, C must state conditions to which we have cognitive access, since
a rule can be followed only if we can be aware that its antecedent
conditions are fulfilled.36 An example of such a norm in perception
would be, "Given such and such sensory experience, believe that there
is a cat on the couch." But now it is an empirical question whether
human cognition can be understood in this way - whether all human
thought can be understood as governed by such rules. Evidence is
mounting that it cannot be.
34 Ibid., p. 131.
35 John Greco, "Internalism and Epistemically Responsible Belief," and "Virtues and Vices
of Virtue Epistemology," Canadian Journal of Philosophy 23 (1993): 413-432.
36 On this point I follow Pollock, Contemporary Theories of Knowledge, p. 133.

193

The problem is not that we do not explicitly follow such rules in our
thinking, or even that implicit norms are exceedingly hard to articulate.
Rather, the problem is that at least some parts of human cognition might
not be rule governed at all. This would be the case, for example, if
connectionist theories of perception and memory are correct. Let us take
a moment to consider this last possibility, and to consider the consequences for understanding cognition as rule governed.
We can think of connectionism as a theory for modeling cognitive
processing.37 Unlike models of cognition suggested by traditional rulebased theories, connectionist systems are made up of numbers of simple
but connected units that can be "activated" or excited to some degree.
The units are set up so that the activation of each affects the activation
of others to which it is immediately connected. In the most interesting
models the interaction among units affects connection "strengths" over
time, so that the effect of one unit on another is increased or inhibited
on the basis of prior interaction between the two units. Processing takes
place in a connectionist system when an initial pattern of activation is
supplied to input units. This original activation sets off activity among
the various units of the system until stability in the system is achieved.
The pattern of activation over the output units then represents the
system's "answer" to a proposed problem.
What is important for our purposes is that processing in a connectionist system does not take place according to programmed rules. Rather,
the interaction among the units of the system is governed only by the
laws of thermodynamics initial activation together with initial connection strengths cause a pattern of activity that eventually settles the system
into the state of highest entropy.38
An example will illustrate. James McClelland, David Rumelhart and
Geoffrey Hinton have proposed a model for the perception of obscured
letters in visually presented words.39 Words are presented in partially
degraded condition, and the system is supposed to identify the obscured
letter. For example, the word W O R K is presented so that the fourth
37 William Bechtel, "Connectionism and the Philosophy of Mind: An Overview," in William
Lycan, ed., Mind and Cognition (Oxford: Blackwell, 1990), p. 253.
38 Ibid., p. 254.
39 The model is described in James McClelland, David Rumelhart, and Geoffrey Hinton,
"The Appeal of Parallel Distributed Processing," in Brian Beakley and Peter Ludlow, eds.,
The Philosophy of Mind (Cambridge, MA: MIT Press, 1992). Reprinted from James Rumelhart, David McClelland, and the PDP Research Group, Parallel Distributed Processing:
Explorations in the Microstructure of Cognition, vol. 1 (Cambridge, MA: MIT Press, 1986).

194

letter is partially obscured, and the problem for the system is to correctly
identify the letter as a K. Here is how it works.
The model assumes that there are units that act as detectors for the visual
features which distinguish letters, with one set of units assigned to detect the
features in each of the different letter-positions in the word. . . . There are also
four sets of detectors for the letters themselves and a set of detectors for the
words.
In the model, each unit has an activation value, corresponding roughly to
the strength of the hypothesis that what that unit stands for is present in the
perceptual input. The model honors the following important relations which
hold between these 'hypotheses' or activations: First, to the extent that two
hypotheses are mutually consistent, they should support each other. Thus, units
that are mutually consistent in the way the letter T in the first position is
consistent with the word TAKE, tend to excite each other. Second, to the extent
that two hypotheses are mutually inconsistent, they should weaken each other.
. . . [Inconsistencies operate in the word perception model to reduce the activations of units. Thus, the letter units in each position compete with all other
letters in the same position, and the word units compete with each other. This
type of inhibitory interaction is often called competitive inhibition. In addition,
there are inhibitory interactions between incompatible units on different levels.
This type of inhibitory interaction is simply called between-level inhibition.40
The present model does not operate like classical artificial intelligence
models, where a system calculates outputs from inputs according to
programmed rules. Nevertheless, the system works so as to give the right
answer. In the following illustration the letters W, O, and R are presented as completely visible and enough of a fourth letter is shown to
rule out all letters other than R and K.
Before onset of the display, the activations of the units are set at or below 0.
When the display is presented, detectors for the features present in each position
become active (i.e., their activations grow above 0). At this point, they begin
to excite and inhibit the corresponding detectors for the letters. In the first three
positions, W, O, and R are unambiguously activated, so we will focus our
attention on the fourth position where R and K are both equally consistent
with the active features. Here, the activations of the detectors for R and K start
out growing together, as the feature detectors below them become activated.
As these detectors become active, they and the active letter detectors for W, O,
and R in the other positions start to activate detectors for words which have
these letters in them and to inhibit detectors for words which do not have these
40 McClelland et al., pp. 273-275.

195

letters. A number of words are partially consistent with the active letters, and
receive some net excitation from the letter level, but only the word WORK
matches one of the active letters in all four positions. As a result, WORK becomes
more active than any other word and inhibits the other words, thereby successfully dominating the pattern of activation among the word units. As it grows in
strength, it sends feedback to the letter level, reinforcing the activations of the
W, O, R, and K in the corresponding positions. In the fourth position, this
feedback gives K the upper hand over R, and eventually the stronger activation
of the K detector allows it to dominate the pattern of activation, suppressing the
R detector completely.41

Here we have an illustration of perception without rules or norms. More

exactly, we have a model of how perception might work without
norms, since a machine does not literally perceive anything. The important point is that the model opens up this possibility for human perception. Whether human perception really does work this way is an empirical question.
An interesting feature of connectionist systems is that their behavior
manifests regularities as if governed by inferential and other rules. But it
is only "as if," since no such rules play any causal role in the mechanisms
that generate those regularities.42 Moreover, connectionist systems do
not mimic rule following perfectly: exceptions occur, and so even the
appearance of rule following is only approximate. This is in fact an
attractive feature of connectionist models, because humans also behave
as if their thinking is rule governed, but only approximately and with
exception. If connectionist models produce the appearance of rule following, but also the same kind of divergence from rules that we see in
human cognition, then that is good evidence for the thesis that human
cognition is connectionist, and so also merely "as i f by rules or norms.
A further example illustrates how this treatment of rules is suggested
by connectionist models. McClelland, Rumelhart, and Hinton describe
a model for learning the past tenses of English verbs through repeated
examples of root forms paired with past-tense forms. The model consists
of two pools of units, one for representing root forms and the other for
representing their past tenses. Training takes place when units for the
root forms are activated and a learning rule adjusts connection strengths
with the units for past-tense forms.

41 Ibid., p. 275.
42 Jerry Fodor and Zenon Pylyshyn, "Connectionism and Cognitive Architecture: A Critical
Analysis," in Beakley and Ludlow (1992), p. 298. Reprinted from Cognition 20 (1988).

196

The model is trained initially with a small number of verbs that children learn
early in the acquisition process. At this point in learning, it can only produce
appropriate outputs for inputs that it has explicitly been shown. But as it learns
more and more verbs, it exhibits two interesting behaviors. First, it produces
the standard ed past tense when tested with pseudo-verbs or verbs it has never
seen. Second, it "overregularizes" the past tense of irregular words it previously
completed correctly. Often, the model will blend the irregular past tense of the
word with with the regular ed ending, and produce errors like CAMED as the
past of COME. These phenomena mirror those observed in the early phases of
acquisition of control over past tenses in young children.43
T h e eerie similarity to the behavior of young children cannot help but
suggest that they learn the past tenses of verbs in the same way.
The model learns to behave in accordance with the rule, not by explicitly
noting that most words take ed in the past tense in English and storing this rule
away explicitly, but simply by building up a set of connections in a pattern
associator through a long series of simple learning experiences. The same mechanisms of parallel distributed processing and connection modification which are
used in a number of domains serve, in this case, to produce implicit knowledge
tantamount to a linguistic rule. The model also provides a fairly detailed account
of a number of specific aspects of the error patterns children make in learning a
rule. In this sense, it provides a richer and more detailed description of the
acquisition process than any that falls out naturally from the assumption that the
child is building up a repertoire of explicit but inaccessible rules.44
But as Fodor and Pylyshyn emphasize, it is misleading to put the point
in terms of implicit versus explicit rule acquisition. For on a connectionist model rules on the cognitive level do not govern at all. The point is
not that rules governing learning (or interpretation, or inference) are not
"explicitly noted" or "explicitly stored away"; it is that no such rules
play any causal role in cognition. Again, on a connectionist model it is
only "as i f such rules are being followed; in actuality they do not do
any work.45
The preceding discussion has consequences for how we should understand epistemic justification. Specifically, the discussion shows that
(VJ) is more general than a theory of justification in terms of epistemic
norms. The basis of our best cognitive dispositions might be conformance to epistemic rules or norms, but it might not be. Accordingly,
whether the details of (VJ) are to be cashed out in terms of epistemic
43

McClelland et al., p. 287.

44 Ibid.
45 Fodor and Pylyshyn, p. 293.

197

norms is therefore at least partly an empirical question, to be decided by

what empirical investigation into human cognition reveals.
This way of proceeding is in keeping with the methodology of strong
particularism. If connectionism turns out to be true and human perception is not governed by norms, then it would be misguided to require
conformance to norms for perceptual knowledge. It is obvious that I do
know by perception that there is a computer in front of me, and that
there is a cat on the couch. If a theory of knowledge and justification
has built into it presuppositions about human perception that are inconsistent with this, then so much for that theory and its presuppositions. It
will have turned out that a longstanding way of understanding subjective
justification that is, as conformance to correct intellectual norms was
just wrong.
Let us suppose that connectionism is true and that important areas of
human cognition are not governed by norms. Would this mean that
there is no place for epistemic norms whatsoever? Not at all, since there
are several ways in which norms might still be relevant to justification.
First, even if human thinking is not governed by norms, it might still be
the case that good thinking is describable by norms. In other words, even
if norms play no causal role in the production and regulation of belief,
good human thinking might nevertheless be characterized by a set of
rules describing how good thinking proceeds. Second, from the fact that
not all human cognition is governed by norms it does not follow that
none is. Perhaps connectionism is true for some areas of cognition, such
as perception and memory, but not others. Third, even where cognition
is not produced by norm-governed processes in its details, the beliefs
that result may nevertheless be covered by more general epistemic
norms. For example, even if connectionism is correct regarding the
mechanisms of perception, perceptual beliefs might come under some
such norm as "Trust your perception unless you have specific reason
not to." Such a norm might be involved in perception and relevant to
the justification of perceptual beliefs, even if perception is not norm
governed in its details.
Finally, cognitive norms might emerge from processes that are not
originally norm governed. For example, over a period of time perception might give rise to the norm, "Do not trust visual perception
regarding small objects at large distances." Or memory might produce
the norm, "Do not trust ostensible memories of emotionally charged
events." Such norms might serve to refine perception or memory, mak-

198

ing them reliable in conditions or fields where they otherwise would

not be. If so, then conformance to emergent norms can constitute a
significant if partial basis of reliable dispositions. In a similar way, norms
governing good reasoning might be successfully adopted through informal or formal training. For example, empirical research shows that
training in statistical rules can improve inductive reasoning. 46
But even if there is ample place for epistemic norms in human
cognition, we still should not define subjective justification in terms of
conformance to norms. First, it remains an empirical question whether
cognition is governed even by very general covering norms, or refined
by emergent or learned norms. Second, not all beliefs in conformance
with these would be justified. For example, suppose that unknown to
me my perception is malfunctioning, causing me to believe that there is
a cat on the couch on the basis of sensory appearances that would not
normally cause this belief. In such a case my belief would be in conformance with the norm, "Trust your perception unless you have specific
reason not to," but it would not be produced by any stable disposition
usually in place when I am thinking conscientiously. Would such a
belief nevertheless be subjectively justified? In a weak sense yes, since
the belief would violate no norm that I countenance. But it would not
be subjectively justified in a further sense that all knowledge requires:
the belief would not be appropriately grounded in my cognitive character.
Perhaps we should endorse the following relations between subjective
justification and countenanced norms. Let it be stipulated that a belief is
in conformance with a norm just in case the norm plays a relevant causal
role in the production of the belief. A belief is in accordance with a norm
just in case the belief is consistent with the conditional rule that the
norm specifies, even if the rule plays no causal role in the actual production of the belief.
(NJ1) A belief p is subjectively justified for a person S only if S's
believing p is in conformance with the norms that S countenances
when S is thinking conscientiously.
(NJ2) A belief p is subjectively justified for a person S only if S's
believing p is in accordance with norms that correctly describe
S's dispositions when S is thinking conscientiously.
46 John Holland, Keith Holyoak, et al. "Learning Inferential Rules," in Hilary Kornblith, ed.,
Naturalizing Epistemology, 2nd ed. (Cambridge, MA: MIT Press, 1994).

199

Note that (NJ1) and (NJ2) state only necessary conditions for justification, not sufficient ones. Second, (NJ1) applies only when there are
norms which govern S's thinking. If no such norms exist, (NJ1) will be
an empty condition on subjective justification.

6. THE PLACE OF EPISTEMIC RESPONSIBILITY

It has also been common to understand subjective justification in terms

of epistemic responsibility, where responsibility is conceived in terms of
either praiseworthiness or freedom from blame.47 Should we agree that
knowledge requires epistemic responsibility? Not if epistemic praise and
blame are to be understood along Kantian lines. For Kant, the moral
versions of these notions are intimately tied to duty, and duty to conformance to the moral law. But as we have seen, it is an empirical question
whether cognition is governed by laws. Therefore, we should not make
Kantian responsibility a necessary condition of knowledge.
What about Aristotelian responsibility? Aristotle's discussion of moral
praise and blame lays down two major conditions for responsible action:
knowledge of particular circumstances and freedom from compulsion.
Discussion of the latter "control" condition contains two themes, which
Aristotle does not seem to distinguish. First, responsible action must be
such that "the moving principle" is at least partly within the agent. This
is opposed to cases where the agent is coerced by external forces; "[T]hat
is compulsory of which the moving principle is outside, being a principle
in which nothing is contributed by the person who acts."48 Second,
responsible action must be such that the agent could have acted otherwise in a strong sense. Aristotle writes, "Now the man acts voluntarily;
for the principle that moves the instrumental parts of the body in such
actions is in him, and the things of which the moving principle is in a
man himself are in his power to do or not to do." 49 Let us call this
power to do or not to do an action "strong control."
Now the first aspect of control, that the principle of action is within
the agent, fits well with the idea that justified belief is grounded in
cognitive character. For if a belief arises from character, clearly the cause
47

See, e.g., Bonjour, Chisholm, and Ginet in the works previously cited. I have defended
responsibilist conceptions of epistemic justification in "Internalism and Epistemically Responsible Belief and "Virtues and Vices of Virtue Epistemology."
48 Aristotle, The Nicomachean Ethics, trans. David Ross (Oxford: Oxford University Press,
1980), 3.1.
49 Ibid.

200

of the belief is partly the knower herself. Problems arise, however,

regarding the second aspect of control. Specifically, it is not clear
whether knowledge is always such that the knower could have believed
otherwise.
To be sure, Aristotle does not require direct or immediate strong
control for moral responsibility. Thus he says that persons can be responsible for their states of character because they can be "part-causes" of
such.50 For example, a person can be blameworthy for ill health even if
it is not in his power now to be healthy. This will be so if earlier
voluntary actions were part-causes of his present illness. Similarly, one
might hold that we have at least indirect control over our doxastic lives.
For example, we can decide to gather more evidence before drawing
conclusions about some matter. Moreover, we can strive to develop
better cognitive habits or work to alleviate cognitive vices.
But still, the issue of strong control can be pressed. Is it realistic to
think that all human cognition is at least within our indirect strong
control? For example, is it in my power (in any relevant sense) to refrain
from believing that there is a truck bearing down on me when I see that
there is? The issue is complicated, since the ways in which our choices
and actions can affect even perception are not fully understood. But
perhaps this point alone is sufficient reason to reject any such condition
on subjective justification or knowledge: namely, it is largely an empirical question whether we have strong control over our cognition, and
to what extent. Therefore, it is an empirical question whether all human
knowledge can satisfy this strong Aristotelian condition on praiseworthiness and blameworthiness. This being the case, we should not say that
strong Aristotelian responsibility is a condition on knowledge. As in the
case of cognitive norms, we do not want to build into our theory of
knowledge conditions that we may in fact not satisfy.
Another consideration confirms the conclusion that knowledge does
not require this kind of epistemic responsibility. Suppose it turns out
that human perception is not within our strong control, even indirectly.
Suppose also that, nevertheless, our perception is as reliable as we ordinarily assume that it is. Would we hesitate to say that we can know
what we clearly see to be the case? For example, would we say that I
cannot know that there is a truck bearing down on me because, as it
turns out, I cannot help but believe that there is, given the truck that I
see? Or suppose that we come across alien beings who have far more
50

Ibid., 3.5.

201

accurate and reliable perception than we do, but who do not have the
power to believe otherwise than how they perceive. It seems wrong to
say that, for lack of control in this strong sense, such beings do not have
perceptual knowledge. These considerations suggest that epistemic responsibility, in any sense that requires the ability to believe otherwise, is
not a part of our concept of knowledge. Even if, as a matter of fact, all
of our cognition is within our strong control and so all knowledge is
praiseworthy, such praiseworthiness is not a necessary condition on
knowledge.
However, there remains a weaker sense of praiseworthiness in which
it is correct to say that knowledge requires epistemic responsibility:
namely, there is a sense in which something is praiseworthy just in case
(a) it is properly motivated and (b) the principle of action is within the
agent. The way that agent reliabilism understands subjective justification
guarantees that these weaker conditions are fulfilled in cases of knowledge.51

7. CONCLUSIONS

We have seen that reliabilism in general can provide a theoretical explanation of the conclusions expressed in (a) through (d), and that agent
reliabilism can account for the concern in (e) as well. In other words,
agent reliabilism provides a principled refutation of all of the skeptical
arguments that we have considered and also explains in what sense
knowers must be sensitive to the reliability of their inferences. The
theory also has resources for defining an important sense of subjective
justification. In this way, it accounts for our persisting intuition that
knowledge ought to be well formed from the knower's point of view,
or that knowledge must be subjectively appropriate as well as objectively
reliable. Finally, the theory does all of this in a way that is psychologically
plausible, and that captures our pre-theoretical intuitions about what
cases do and do not count as knowledge.
51 Matthias Steup argues that cognitive control can be fruitfully understood in terms of sensitivity to one's evidence. Just as some philosophers have attempted to define "could have
done otherwise" as "would have done otherwise if choice had been different," Steup
suggests that we might define "could have believed otherwise" as "would have believed
otherwise if evidence had been different." Perhaps this describes another sense in which
knowledge can be understood to involve epistemic responsibility. "Epistemic Obligation
and the Freedom to Believe Otherwise," typescript.

202

In the next chapter I continue the defense of agent reliabilism by

showing how it gives a theoretical explanation of what it is for an
alternative possibility to be epistemically relevant. That is, the position
explains which alternative possibilities to one's belief need to be ruled
out to have knowledge and which do not, and why this is so.

203

8
Agent Reliabilism and the Relevant
Sense of "Relevant Possibility"

Recall the skeptical argument (D3) from Chapter 2. The major premise
of that argument was that good evidence must discriminate among
alternative possibilities. In other words, my evidence for a belief p can
give me knowledge only if it would cause me to believe that p is true
in cases where p is true, and also would not cause me to believe that p
is true in cases where p is false and some alternative is true. But, the
argument continues, since sensory appearances do not discriminate in
this way between what we believe and various skeptical scenarios, sensory evidence cannot give rise to knowledge of the material world.
I suggested that a solution to (D3) is to deny the assumption that
evidence must rule out all alternative possibilities to generate knowledge.
The idea was that evidence must discriminate among relevant possibilities,
but not all possibilities are relevant. In particular the possibilities that I
am a brain in a vat or the victim of an evil demon do not seem to be
relevant ones. Of course if I were a brain in a vat or the victim of a
deceiving demon, then I would not know what I think I do, but so
long as these remain mere possibilities they fail to undermine my knowledge.
This strategy against (D3) gave rise to (D4). That argument ran as
follows:
(D4)

1. S can know that p is true on the basis of evidence E only if E

discriminates p's being true from all relevant possibilities that are
inconsistent with p's being true.
2. My evidence for my belief that I am sitting by the fire is my
sensory experience.
3. It is a relevant possibility that I am not sitting by the fire but only
dreaming that I am.

204

4. Therefore, I can know that I am sitting by the fire only if my

sensory experience discriminates my sitting by the fire from my
only dreaming that I am. (1,2,3)
5. But my sensory experience does not discriminate these possibilities
for me.
6. Therefore, I do not know that I am sitting by the fire. (4,5)

Since (D4) concedes that evidence need only rule out relevant possibilities, its first premise becomes unassailable. But now premises (3) and (5)
of (D4) become problematic. Specifically, the idea of dreaming in (3) is
ambiguous between normal dreams that occur in sleep and "philosophical" dreams involving evil demons and brains in vats. The problem is
that on either interpretation the argument contains a claim that is less
than wholly plausible.
If we interpret (D4) as making claims about normal dreams, then
premise (5) seems false. Perhaps it is at times a relevant possibility that I
am dreaming in the everyday sense of "dreaming," but in such cases my
evidence does discriminate that possibility from waking life; if I pay close
attention, or perhaps pinch myself, I can tell that I am not dreaming.
On the other hand, if (D4) is about philosophical dreams, then premise
(3) seems false, or at least less than obvious. Intuitively it is not a relevant
possibility that I am the victim of a deceiving demon, or that I am a
brain in a vat. These are possibilities in some broad sense, but not in any
sense that seems to present a problem for my knowing. Therefore, on
either interpretation of dreaming, the skeptical argument contains a
premise that we need not accept.
1. TWO TASKS FOR WORKING OUT THIS APPROACH
The relevant possibilities approach has been a popular one, but it needs
to be worked out further. First, we want an account of what makes a
possibility relevant and why the skeptical hypotheses do not constitute
relevant possibilities. In other words, we need an account of "relevant
possibility" that (a) accords with our pre-theoretical intuitions about
which possibilities need to be ruled out and which do not in ordinary
cases of knowledge and (b) rules that the skeptical possibilities do not
need to be ruled out. Second, we do not want our account to be ad hoc.
What we would like is an account of knowledge which explains why
our account of relevant possibility has the content that it does.

205

a. Providing an Account of "Relevant Possibility"

It has been popular to attempt an account of relevant possibility in terms

of objective probability.1 The problem with this solution is that we lack
an adequate account of "objectively probable." The most natural way
to understand that notion is in terms of percentage of a reference class.
However, it is hard to see how a reference class could be specified in
the present case, or even what a reference class would begin to look like
here. For this reason attempts to define relevant possibility along these
lines have failed.2
I propose that we understand relevant possibility in a different way,
making use of the "possible worlds" semantics of modal logic. The idea
is that it is a logical possibility that I am a brain in a vat hooked up to a
supercomputer programmed to deceive me. In other words, there are
possible worlds in which this is the case. But such possible worlds are
"far away" from the actual world. There are no close possible worlds in
which I am a brain in a vat, where "closeness" is to be understood,
roughly, in terms of overall world similarity. The current proposal, then
is, this: a possibility is relevant if it is true in some close possible world,
irrelevant if not. This proposal explains many of our pre-theoretical
intuitions about which possibilities need be ruled out and which do not
in order to know, and it explains why the skeptical possibilities do not
need to be ruled out.
More work needs to be done, of course. Most importantly, the proposed account of relevant possibility is too weak as it stands; more possibilities are epistemically relevant than those that are true in close possible
worlds. Also, we will have to say more about what constitutes world
closeness in order to make the account more informative. But this is the
main idea for completing the first task. The details presented in Sections
2 and 3 will fix up the account by making it stronger in the required
way and by making the notion of world closeness more informative.

See, e.g., Fred Dretske, "Epistemic Operators, "Journal of Philosophy 67 (1970): 1007-1023;
Alvin Goldman, "Discrimination and Perceptual Knowledge"; and Marshall Swain, "Revisions of'Knowledge, Causality, and Justification,' " in Pappas and Swain (1978). For objections to these views see, Sosa, "Knowledge in Context, Skepticism in Doubt," in James
Tomberlin, ed., Philosophical Perspectives 2: Epistemology (Atascadero, CA: Ridgeview, 1988).
The account I offer here avoids Sosa's objections by making the notion of objective likelihood more informative, and by adding subjective conditions to the account of "relevant
possibility."
Sosa makes this point in "Knowledge in Context, Skepticism in Doubt."

206

b. Providing an Account of Knowledge to Support Our Account of

Relevant Possibility

I have already said that agent reliabilism can provide a theoretical understanding of what makes a possibility relevant. In other words, the theory
can explain why some possibilities need to be ruled out to have knowledge and why others do not. In the present context, this is equivalent to
explaining why evidence must discriminate among possibilities that are
true in close possible worlds, but not among possibilities that are true
only in far-off worlds.
The main idea is as follows. Cognitive virtues are a kind of power
or ability. Abilities in general are stable and successful dispositions to
achieve certain results under certain conditions. But abilities cannot be
defined in terms of actual conditions only. Rather, when we say that
someone has an ability we mean that she would be likely to achieve
the relevant results in a variety of conditions similar to those that actually obtain. In the language of possible worlds, someone has an ability
to achieve some result under relevant conditions only if the person is
very likely to achieve that result across close possible worlds. But if
knowledge essentially involves having cognitive abilities, and if abilities
are dispositions to achieve results across close possible worlds, then this
explains why possibilities are relevant only when they are true in some
close possible world. Specifically, only such possibilities as these can
undermine one's cognitive abilities. In an environment where deception by demons is actual or probable, I lack the ability to reliably form
true beliefs and avoid false beliefs. But if no such demons exist in this
world or similar ones, they do not affect my cognitive faculties and
habits.
The remainder of the chapter develops this main idea. In Section 2 I
present and defend a more detailed account of relevant possibility. In
Section 3 I further explore the idea of a cognitive virtue by offering a
possible worlds analysis for abilities in general and cognitive abilities in
particular. In Section 4, I show how agent reliabilism explains the
account of relevant possibility presented in Section 2.
2. THE RELEVANT SENSE OF "RELEVANT POSSIBILITY"
Our first task is to provide an account of relevant possibility such that
(a) knowledge requires only that our evidence rule out (discriminate)
that kind of possibility and (b) the skeptical possibilities involving brains

207

in vats and evil demons turn out not to be relevant possibilities. As a

first try consider the following account of relevant possibility, which I
believe comes close to meeting these two criteria. Let us say that a
possible world W is close to a possible world Wf if and only if, roughly,
W is very similar to W\ Then,
q is a relevant possibility with respect to S's knowing that p is true if and
only if
i. If q is true, then S does not know that p is true, and
ii. In some close possible world, q is true.
What is important to notice is that this account effectively disallows that
evil demons and brains in vats are relevant possibilities. So long as the
actual world is anything like we think that it is, there are no close worlds
in which we are disembodied spirits or brains in vats. Accordingly, the
account has significant consequences for the force of (D4). Premise (3)
of that argument implies that the skeptical scenarios are relevant possibilities. On the current account this amounts to saying that the skeptical
possibilities are true in some close possible world. But of course there is
no reason whatsoever to accept that claim. Presumably such scenarios
are logically possible, so that each is true in some far-off possible world.
But we have no reason to believe that either is true in any close possible
world.
Therefore, if the current account of relevant possibility is correct,
then it seriously undermines the force of the skeptical argument. The
claim that the account is correct is supported by reflection on two
further examples. First, imagine that you see Tom Grabit take a book
from a library shelf and put it in his bag.3 On the basis of this evidence
you form the belief that Tom Grabit took a library book. Unknown to
you, however, Tom has a twin brother Jack, who was also in the library
at the time you saw Tom take the book. Intuitively, you do not know
that it was Tom who took the book, since if Jack had taken the book
you would have been fooled into thinking that Jack was Tom. In order
for you to know that Tom took the book you must be able to rule out
the possibility that it was Jack who took the book. But consider the case
where Tom does not have a twin brother. Must you still be able to rule
out the possibility that it was his twin brother who took the book, in
3

Tom Grabit cases were first introduced by Lehrer and Paxson. See Keith Lehrer and Thomas
Paxson, "Knowledge: Undefeated Justified True Belief," Journal of Philosophy 66 (1969):
225-237, reprinted in Pappas and Swain (1978).

208

order to know that Tom took the book? Intuitively the answer is no. In
the case where there is no twin brother it is enough that you saw Tom
take the book.
As a second example, consider the case of the barn facades. You are
driving through a part of the country where, unknown to you, the local
residents have constructed sides of barns in order to fool passers-by into
thinking that the community is wealthier than it actually is. You drive
by a real barn and form the belief that there is a barn ahead. In this case
the proposition that you see a barn facade rather than a barn is a relevant
possibility. You do not know the latter if you cannot rule out the
former. But must you always be able to rule out the possibility of a barn
facade before you can know that you see a barn from the highway?
Intuitively the answer is no, and this is in fact how the account rules.4
The proposed account of relevant possibility handles these two cases
well. However, it is not quite right and some revisions are necessary.
This is because there are possibilities that are not true in any close
possible world, but that nevertheless need to be ruled out in order for
someone to have knowledge. Consider a revised version of the Tom
Grabit example. Everything is as before, except that Tom Grabit has no
twin brother. However, S believes that Tom does have a twin brother. It
seems to me that S's evidence must rule out the possibility that Tom's
twin brother took the book, even though the possibility is not true in
any close possible world (since Tom in fact has no twin brother).
Another version of the Tom Grabit example raises a different problem.
Suppose that Tom does not have a twin brother and S does not believe
that Tom has a twin brother, but S ought to believe that Tom has a
twin brother S has good reason for believing that Tom has a twin
brother. Again, it seems to me that in such a case the possibility that
Tom's twin took the book must be ruled out by S's evidence. In these
cases the problem is not that the possibility is true in some close possible
world, but that S believes, or ought to believe, that the possibility is
likely to be true. Fortunately the account can easily be revised to accommodate both of the these cases. Thus we have,
(RP)

q is a relevant possibility with respect to S's knowing that p is true

if and only if
i. If q is true, then S does not know that p is true; and
ii. Either (a) in some close possible world q is true, or

The example is from Goldman, "Discrimination and Perceptual Knowledge."

209

(b) S believes that q is likely to be true, or

(c) S ought to believe that q is likely to be true.
I believe that (RP) covers all and only those possibilities that must be
ruled out in cases of knowledge. Secondly, (RP) continues to exclude
the skeptical scenarios as relevant possibilities, assuming that there are no
brains in vats or victims of evil demons in our environment, and assuming that we neither believe nor ought to believe that there are.
Before moving on we should address a possible misconception.
Namely, one might think that the present strategy against (D4) merely
ends in a stalemate with the skeptic. This is because the skeptic will
insist that, for all we know, some skeptical hypothesis is true in some
close possible world, perhaps because it is true in the actual world. But
this kind of objection is misguided in two respects. First, there is no
skeptic with whom we are debating and with whom we could fall into
such a stalemate. Rather, what we have is a skeptical argument and we
are evaluating its force so as to root out any mistaken assumptions
that it might contain. We are trying to learn a lesson about what
knowledge does and does not require, and how we might fulfill those
requirements.
Second, the objection misunderstands the dialectic of the investigation so far. We deemed Descartes' original argument to have force
precisely because it claimed only that the skeptical scenarios were in
some sense possibilities, and it does seem that we have to allow that
assumption. But we have now pushed the argument so that premise (3)
of (D4) claims that the scenarios are relevant possibilities, and on the
present interpretation this amounts to the claim that in some close
possible world I am a brain in a vat or the victim of an evil deceiver.
But now the argument is flawed if it depends on that premise. We will
have found an assumption that we need not accept, and that in fact there
is no reason whatsoever to accept.
In effect this shows that there is only one way to challenge the present
strategy against (D4): namely, one must challenge the account of relevant
possibility in (RP). For if that account is right, clearly the claim that the
skeptical scenarios are relevant is implausible. I will now go in the other
direction, however; that is, I will now defend the account in (RP) by
showing how it is explained by agent reliabilism.

210

3. A THEORY OF VIRTUES AND A VIRTUE THEORY OF

KNOWLEDGE

The central idea of agent reliabilism is that knowledge arises from cognitive abilities or powers. An ability, in turn, is a stable and successful
disposition for achieving some result under appropriate conditions. What
we need now is a more detailed account of what this amounts to. I will
proceed by offering an account of abilities in general and then an
account of cognitive abilities in particular. After that it will be relatively
easy to see how the resulting theory explains the proposed account of
relevant possibility in (RP).
a. What Is an Ability?
Roughly, an ability in general is a stable disposition to achieve some
result under appropriate conditions. For example, we say that Don
Mattingly has the ability to hit baseballs. By this we mean that Mattingly
has a disposition to hit baseballs under normal conditions for playing
baseball. Notice that we do not require that Mattingly have perfect
success, nor do we require that he have a disposition for hitting baseballs
under just any conditions. In general, how high a success rate is required
will depend on the kind of ability in question. Likewise, what conditions
are appropriate will depend on the kind of ability in question, as well as
on other contextual matters.
These remarks might suggest the following account of having an
ability to achieve a result:
(Al)

S has an ability to achieve result R in conditions C if and only if

S has a high rate of success achieving R when S is in C in actual
cases.

The current proposal, however, is both too weak and too strong. First,
(Al) does not distinguish between having an ability to achieve R and
having success achieving R due to good luck. Thus it is possible that in
all actual cases of swinging at baseballs, I hit the ball due to amazingly
good luck. My bat just happens to be where it ought to be on every
pitch. In that case I would have great success hitting baseballs in actual
cases, but it would be false that I have an ability to hit baseballs. Or
consider the case of Mr. Magoo. Magoo in fact is highly successful in
avoiding harm, meaning that in the actual world Magoo rarely comes to
harm. But we would not say that Magoo has an ability to avoid harm,

211

since he avoids harm only by amazingly good luck. (An inch to the left
and that anvil lands on his head!)
For a similar reason the current proposal is also too strong. For it
might be that S does have an ability to achieve R but fails in nearly all
actual cases due to amazingly bad luck. Just as it is possible to have an
ability to achieve R and yet fail in some actual case, it is possible (even
if improbable) to have the ability and yet fail in nearly all actual cases.
Finally, it makes sense to say that S has an ability to achieve R even if S
never achieves R because S never comes under the appropriate conditions. Thus it makes perfect sense to say that you have the ability to
count by fives to one thousand, even if you never do so because no
actual situation ever calls for it. And so we have several reasons why
(Al) fails to capture what it is to have an ability in general.
These last considerations suggest that having an ability is not a function of what S does in actual cases, but of what S would do in possible
cases. And this suggests the following subjunctive account of having an
ability:
(A2) S has an ability to achieve result R in conditions C if and only if,
if S were in C, then S would achieve R with a high rate of success.
But this account is also both too weak and too strong, and for somewhat
similar reasons as with (Al). First, consider the case where S is often in
C and has success achieving R in actual cases due to good luck. Then
the subjunctive conditional in (A2) is true, but S does not have an ability
to achieve R. Second, consider the case where S is often in C but fails
to achieve R in actual cases due to bad luck. Then the subjunctive
conditional in (A2) is false, but it is possible that S does have an ability
to achieve R.
I suggest that the following account avoids all of the problems that
we have raised for (Al) and (A2):
(A3) S has an ability to achieve result R in conditions C if and only if,
across the range of close possible worlds where S is in C, S achieves
R in C with a high rate of success.
Two examples will help to explain the motivation for (A3). First, consider that in some close possible world a fair coin will come up heads
one hundred times in one hundred tosses. However, across a range of
close possible worlds, the coin will come up heads 50 percent of the
time, since improbable strings of heads will be offset by improbable

212

strings of tails. Now let us consider an example of an ability. There is a

close possible world where Mattingly hits .000 for the month of April,
due to an improbable string of bad luck. But across the range of close
possible worlds, Mattingly presumably hits near .300, since strings of bad
luck are offset by strings of good luck. The current proposal is that this
is why Mattingly has the ability to hit baseballs; he in fact hits near .300
in the actual world, but, more importantly, he hits near .300 across the
range of close possible worlds.
The current proposal, then, does not equate abilities with actual
successes, since it requires that S have a high rate of success across close
possible worlds. For this reason (A3) handles the case of Mr. Magoo.
Although Magoo avoids harm in the actual world, there are close possible worlds (large numbers of them in fact) where Magoo is harmed at
every step. Further, the proposal does not rule out abilities where there
is no actual high rate of success. Thus where S has little actual success
achieving R, but only due to strings of bad luck, it will still be the case
that S has a high rate of success through close possible worlds. The case
in which S never finds herself in the relevant conditions is also handled
nicely by the current account. For even if S is never in C in the actual
world, it will still be either true or false that, across the range of close
possible worlds, S has a high rate of success with respect to R when
under the relevant conditions. (In the close worlds where you must
count by fives to one thousand to save your life, you do it.)
Finally, consider how the current proposal distinguishes Mattingly
from the lucky batter. We say that Mattingly has an ability to hit
baseballs. Not only does he have good success hitting baseballs in the
actual world, but presumably he has a high rate of success across close
possible worlds. Thus in worlds where the pitch comes in a little lower,
Mattingly adjusts his swing accordingly. Not so with the lucky batter.
The lucky batter swings without an ability and just happens to connect
with the pitch. If the pitch had come in a little lower, the lucky batter
would have swung over it.
We may therefore accept (A3) as a correct account of what it is to
have an ability in general. Of course (A3) is vague in many respects. It
does not say how high the rate of success across close possible worlds
must be, or which conditions are the relevant ones. But this is as it
should be. First, which conditions are relevant will depend on the ability
in question, as well as on context. For example, when we say that
Mattingly can hit baseballs we obviously do not mean at night and
without lights, but this is well understood in the context of a conversa-

213

tion about baseball. Similarly, contextual features set different boundaries

when we say that Mattingly has the ability to hit baseballs and when we
say that a Little Leaguer has the ability to hit baseballs.
Context does not eliminate vagueness completely, but this too is as it
should be. This is because the concept of an ability is vague even within
contexts. For example, when I claim that a rookie has the ability to hit
baseballs, do I mean that he can hit in pressure situations? Do I mean
that he can hit against the very best pitchers in the league? Normally my
claim will be at least somewhat vague in these respects. Sometimes closer
attention to context or further stipulation will eliminate such vagueness,
and sometimes it will not.
Other kinds of vagueness are associated with the concept of world
closeness, and these might seem more problematic. We said that closeness is a function of overall world similarity, but what kinds of similarity
count, and what degree of similarity is important? Again, this will depend on the kind of ability in question, and it will depend on contextual
features as well. But also again, fixing kind of ability and contextual
features will not eliminate all of the vagueness involved.
Having said this, we can nevertheless give our concept of world
closeness a substantial degree of content. First, what counts for world
closeness is partly determined by our account as it stands. Specifically,
since (A3) relativizes abilities to a range of relevant conditions, only
worlds in which S is in such conditions will be relevant for consideration. Second, it seems clear that when we say that S has an ability, we
mean that S has the ability as S is actually constituted. For example,
when we say that Mattingly can hit baseballs, this does not become false
because in some possible worlds Mattingly is injured or drugged or out
of shape and therefore can no longer hit baseballs. For similar reasons it
is clear that only worlds with the same laws of nature as hold in the
actual world will count as close.
The issue can be pressed, however. Which aspects of S's constitution
are relevant for determining world closeness, where the issue is whether
S has the ability to achieve result R in conditions C? The answer seems
to be that it is those aspects of S's constitution which, given the actual
laws of nature and given conditions C, are relevant to whether S's
actions in C result in producing R. In other words, we will want to
keep constant any characteristics of S which, given the laws of nature
and the relevant conditions, figure into whether the relevant results
occur.
Two examples here should lend some clarity. First, we want to say

214

that Mattingly has the ability to hit baseballs because, given Mattingly's
actual constitution, the normal conditions for hitting baseballs, and the
laws of nature which actually hold, Mattingly's actions result in a high
ratio of hit baseballs. This occurs in the actual world, but it would also
occur if the world were a little different: if the pitch came in a bit lower,
Mattingly would still hit the baseball. On the other hand, it makes no
sense to say that Mattingly lacks an ability to hit baseballs because in
some possible worlds he is nearsighted. This is because the constitution
of Mattingly's eyes figures importantly into the laws of nature that
govern the occurrence of the result in question.
Let us consider a second example. Imagine that due to amazingly
good luck my bat hits the baseball whenever I swing at a pitch in the
actual world. It does not follow that I have the ability to hit baseballs.
This is because, given the constitution of my eyes, and given the laws of
nature that actually hold, if the world were a little different I would miss
the baseball entirely. When the pitch comes in a little lower, Mattingly,
constituted as he is, adjusts his swing. But when the pitch comes in a
little lower to me, I see no difference and I swing over the ball.
We can now make the notion of world closeness more informative.
Let us define S's R-constitution as those characteristics of S that, given
the actual laws of nature, are relevant to whether S's actions result in
producing R when S is in conditions C. Then we may say that a world
W is close to the actual world (in the sense relevant to whether S has
the ability to produce R in C) only if S has the same R-constitution in
W as in the actual world, and only if the laws of nature that hold in W
are the same as those that hold in the actual world. (A3) should be read
with this characterization of world closeness in mind.
According to these suggestions, the conditions for S's having an
ability involve four dimensions, three of which are kept constant and
one of which is allowed to vary. To determine whether S has an ability
in question, we keep constant the actual laws of nature, the set of
conditions associated with the ability, and S's R-constitution associated
with the ability. We then vary the other conditions of the environment,
and we look at S's success rates in the worlds where such variations are
kept relatively minor. What kind of environmental variations are important and what makes a variation relatively minor? This is left vague in
(A3), but context will serve to eliminate much of that vagueness.

215

b. What Is a Cognitive Ability?

A cognitive ability has the form of an ability in general, but with the
cognitive end of arriving at true beliefs and avoiding false beliefs.
Therefore a cognitive ability, in the sense intended, is an ability to arrive
at truths in a particular field and to avoid believing falsehoods in that
field, under relevant conditions.
More formally,
(CA)

S has a cognitive ability A(C,F) with respect to proposition p if

and only if there is a field of propositions F and a set of conditions
C such that
i. p is in F, and
ii. Across the range of close possible worlds where S is in C, S
has a high rate of success with respect to believing correctly
about propositions in F.5

This account successfully distinguishes cognitive abilities such as vision,

memory, and reliable reasoning from non-abilities such as dreaming,
wishful thinking, and hasty generalization. Thus under normal conditions vision is a cognitive ability with respect to determining color
properties of middle-sized objects. This is because in the actual world
and across close possible worlds we have a propensity to believe correctly
about such propositions when in normal conditions. In other words,
across these worlds we have a high rate of success. However, we have
no such propensity with respect to beliefs caused by dreaming. Even if
by coincidence someone were to enjoy actual success in forming true
beliefs through dreaming, there would be no high rate of success across
close possible worlds, and so the lucky dreamer would not count as
having a cognitive ability on the present account.
Similarly, (CA) correctly distinguishes between the two Tom Grabit
cases and between the two barn facade cases. Thus in the case where
Tom has no twin brother, S will very likely be correct about Tom
taking the book in the relevant worlds. But in the case where Tom has
an indistinguishable twin brother, S's rate of success will drop significantly in close worlds. Likewise, in the case where there are no barn
facades in the environment, S will have a propensity to believe correctly
about the existence of barns in the relevant worlds. But in the case
where there are barn facades in the environment, S's rate of success will
drop dramatically across close worlds.
5

Compare Sosa, Knowledge in Perspective, Ch. 16.

216

These cases also constitute examples of how our concept of world

closeness sways with our concepts of a cognitive ability and of knowledge. For example, how far off must the nearest barn facade be before S
has a cognitive ability with respect to beliefs about the existence of
barns? If there are facades in the immediate environment, it is obvious
that S has no ability. And if there are no such facades within one
hundred thousand miles, then it seems that S does have the ability. I
suggest that there are many cases in between where our intuitions are
not sharp. The present point is that our intuitions about world closeness
will sway with our intuitions about whether there is a cognitive ability
and whether there is knowledge. If this is correct, then the vagueness in
our accounts is a virtue rather than a vice. This is because our accounts
of a cognitive ability and of knowledge capture a vagueness that already
exists in those concepts and explain that vagueness by locating it at a
deeper level of analysis.
With the present account of cognitive abilities in place we may now
look at how agent reliabilism explains the account of relevant possibility
in Section 2.
4. AGENT RELIABILISM AND RELEVANT POSSIBILITIES
The way that agent reliabilism explains clause (iia) of (RP) is straightforward. According to agent reliabilism, knowledge is true belief grounded
in cognitive abilities. And according to our account of abilities, they are
a function of S's rates of success across close possible worlds. But then
possibilities that are true only in far-off worlds do not affect S's abilities
in the actual world. If I were a brain in a vat in the actual world, then
across close possible worlds I would lack a high rate of success with
respect to forming true beliefs about my environment. But so long as I
am not a brain in a vat, this mere possibility does not affect my success
rate in the actual world and in close possible worlds. Therefore, the
mere possibility that I am a brain in a vat is irrelevant to whether I
know.
This result is obvious even when we grant that the concept of world
similarity is vague, and that this vagueness infects our account of knowledge. This is because it is clear that the world in which I am a brain in
a vat is a very far world from the actual world, given that the actual
world is anything like we think it is. For although our concept of world
similarity is vague, it is not vague in all respects. One stipulation that
was made on world closeness was that a world W is close to the actual

217

world only if S's constitution in W is similar to S's constitution in the

actual world. This stipulation was well motivated in our account of what
it is to have an ability in Section 3, and the same stipulation serves nicely
here as well. For similar reasons, no skeptical scenarios which entail a
change in the laws of nature will turn out to be relevant. Even if such
scenarios count as logical possibilities, they do not count as relevant
possibilities and so do not undermine our capacities for knowledge.
What about clauses (iib) and (iic) of (RP)? These clauses resulted
from our intuitions that knowledge should be subjectively justified as
well as objectively reliable. Put another way, knowledge requires that
one's beliefs be appropriately formed from one's own point of view.
Agent reliabilism recognizes these intuitions and can therefore explain
these additional clauses of (RP). Because our cognitive abilities include
ones governing the evaluation of counter-evidence, a virtuous believer
will discriminate among alternative possibilities that she believes might
be true. Moreover, she will discriminate possibilities that she ought to
think might be true. Let us take a moment to elaborate on this idea.
We are presently working with the following account of knowledge:
(AR)

S knows p only if
i. p is true;
ii. S's believing p is the result of dispositions that S manifests
when S is thinking conscientiously;
iii. Such dispositions make S reliable in the present conditions, with respect to p.

The first point is that the dispositions referred to in (AR) will include
ones that make S sensitive to counter-evidence to what seems to be the
case. So, for example, my perceptual faculties will involve forming
material object beliefs on the basis of sensory appearances, but will also
involve a process of checking those beliefs against background beliefs
that I have. This process need not be a conscious one. What is required
is that appropriate revisions in my belief system are made as new evidence comes in; either the new evidence will issue in new beliefs and
cause conflicting ones to be revised, or it will be resisted and previous
beliefs will be maintained. This grounds clause (iib), which says that an
alternative possibility is relevant if S believes that it might be true.
The second point is that in particular cases of knowledge a person
must manifest the cognitive dispositions that make her reliable in general. Put another way, she must manifest the dispositions that she does
when thinking conscientiously. But this grounds clause (iic), which says

218

that a possibility is relevant if S ought to think that it might be true. In

other words, if a possibility is such that S would believe that it is likely
if S were thinking conscientiously, then S's evidence must rule it out for
S to have knowledge. Only then would S be appropriately reliable in
the particular case. Accordingly, the nature of our cognition together
with the requirements of subjective justification and reliability explain
the various clauses in (RP).
Finally, it is the nature of our cognition that explains the clauses it
is not an abstract truth about knowledge that (RP) has the content that
it does. It is possible, for example, that some beings have perfectly
reliable perception without any process for checking counter-evidence.
But that is not how human cognition works, and so the need for clauses
(iib) and (iic) in an account of what makes a possibility relevant for us.

219

Moral and Religious Epistemology

In this last chapter I want to sketch how the methodology of strong

particularism can be extended to moral and religious epistemology. I do
this by way of three illustrations. In Part I of the chapter, I look at work
by Plantinga and Alston in religious epistemology, and I show how these
authors effectively engage in an indirect application of the methodology.
Both authors argue that objections to the rationality of religious belief
trade on assumptions that, if true, would lead to skepticism in the
empirical realm as well. As a result, the objections against religious belief
are determined to be unsound. Also as a result, positive suggestions for
the epistemology of religious belief emerge. We will see how Plantinga
defends the general idea that beliefs about God might be noninferentially justified, and how Alston develops this line of thought in a
theory of religious perception. The general structure of Alston's discussion is as follows: Objections to religious perception trade on an inadequate understanding of perception in general; such objections, if sound,
would make even empirical perception impossible. Once an adequate
theory of empirical perception is in place, however, the possibility of
perceiving God becomes a live option in religious epistemology.
In Part II of the chapter, I want to do for moral perception what
Alston does for religious perception. Drawing on the results of Chapters
2 through 8, I argue that traditional objections to moral perception
misunderstand the nature of perception in general. Again, if such objections were sound, then analogous ones would show that even empirical
perception is impossible. However, a theory of empirical perception that
is both epistemologically and psychologically adequate quite naturally
suggests the possibility of moral perception. Drawing on some recent

220

developments in empirical psychology, I argue that a particular model

of empirical perception is especially suggestive along these lines.

/. Religious Epistemology
I have said that strong particularism can be applied to moral and religious
epistemology either directly or indirectly. A direct application of the
method requires us to assume that moral (or religious) beliefs have some
kind of positive epistemic status. We might assume, for example, that
some moral beliefs amount to knowledge, or that some religious beliefs
are at least rational. We may then reject assumptions about knowledge
and rationality that are inconsistent with the assumptions we are willing
to make. If a skeptical argument leads to the conclusion that no one
knows that any action is wrong, for example, then there must be something wrong with that argument. Progress in the epistemology of moral
belief is made if we can identify what that mistake is. Further progress is
made if we can replace it with something better.
As I have said, this seems to me to be a perfectly reasonable way to
proceed in moral and religious epistemology. On the one hand, it seems
outrageous to think that no one knows anything about what is right or
wrong, or that no one's religious beliefs are even rational. On the other
hand, some of the assumptions operative in discussions about moral and
religious epistemology have these consequences. The implausibility of
such assumptions is exposed by making their skeptical consequences
explicit. That is, we want to make those consequences fully explicit,
pressing the logic of a skeptical line of reasoning so as to display the full
force and range of its skeptical conclusions. Moreover, we want to stress
the counterintuitiveness of the more radical forms of skepticism that
result. It is easy enough to say in the classroom that no one "really"
knows right from wrong, but almost no one actually believes this.
Clearly no such belief is manifested in the way that people actually make
moral judgments, or in the way that people actually live.
It seems to me, therefore, that strong particularism can be extended
directly to moral and religious epistemology. But the methodology can
be applied indirectly as well. In this case we do not assume that moral
or religious beliefs have some positive epistemic status, but that empirical
beliefs do. We then look for arguments against moral (or religious) belief
such that, if they were sound, would entail unwanted skeptical conse-

221

quences for empirical belief as well. By rooting out the mistaken assumptions in such arguments, and by replacing them with something
better, we again make advances in moral and religious epistemology. A
paradigm example of this methodology is Plantinga's treatment of the
evidentialist objection to belief in God. I turn to that now to illustrate
the methodology I have in mind.

1. PLANTINGA'S REJECTION OF THE EVIDENTIALIST

OBJECTION TO RELIGIOUS BELIEF

According to the evidentialist objection, belief in God is not rational

unless it is based on good evidence.1 But since there is no such evidence,
belief in God is not rational. This kind of objection to the rationality of
religious belief is widespread. In turn, many theists have tried to answer
it, most often by trying to satisfy the demand for evidence. Recent work
by Plantinga takes a different approach, however. Rather than focusing
on the second premise of the argument, which states that there is no
good evidence for belief in God, Plantinga addresses the first premise.
Why should one accept the premise that belief in God is rational only
if it is based on good evidence? Proponents of the evidentialist objection
do not require that all rational beliefs be based on other beliefs for their
evidence, since this would issue in an infinite regress. They must be
thinking, therefore, that beliefs about God are of a kind that need
evidence. But this last is a substantive assumption, Plantinga argues. It
depends on a theory of rationality which divides rational beliefs into
those that need evidence and those that do not, and which places belief
in God squarely in the former camp.
According to Plantinga, the evidentialist objection assumes some form
of classical foundationalism. Classical foundationalism is a particularly
strict version of foundationalism. Specifically, it is particularly strict about
what kinds of belief are allowed to be foundational. According to ancient and medieval classical foundationalism, only beliefs that are selfevident or evident to the senses are foundational, with all other rational
belief requiring support from these. Modern classical foundationalism is
even more strict, allowing as foundational only self-evident beliefs and
incorrigible beliefs about one's own experience. It is because beliefs
1

The discussion that follows is based on Alvin Plantinga, "Reason and Belief in God," in
Alvin Plantinga and Nicholas Wolterstorff, eds., Faith and Rationality (Notre Dame: University of Notre Dame Press, 1983).

222

about God fall into none of these categories, Plantinga argues, that the
objection to religious belief demands evidence. Beliefs about God are
neither self-evident, evident to the senses, nor incorrigible, and so if
they are rational at all they must be based on other beliefs that have such
status.
The next move that Plantinga makes is the one that is most relevant
for our purposes. Having identified classical foundationalism as a crucial
assumption of the evidentialist objection, he goes on to reject the assumption precisely because it leads to broader skeptical consequences. If
classical foundationalism were true, this would have implausible skeptical
consequences for other kinds of belief, such as beliefs about the past and
beliefs about other minds. And so classical foundationalism must be false.
We should note first that... if these claims are true, then enormous quantities of what we all in fact believe are irrational. One crucial lesson to be learned
from the development of modern philosophy Descartes through Hume,
roughly is just this: relative to propositions that are self-evident and incorrigible, most of the beliefs that form the stock in trade of ordinary everyday life are
not probable. . . . Consider all those propositions that entail, say, that there are
enduring physical objects, or that there are persons distinct from myself, or that
the world has existed for more than five minutes: none of these propositions, I
think, is more probable than not with respect to what is self-evident or incorrigible for me.
. . . . And now suppose that we add to the foundations propositions that are
evident to the senses, thereby moving from modern to ancient and medieval
foundationalism. Then propositions entailing the existence of material objects
will of course be probable with respect to the foundations, because included
therein. But the same cannot be said either for propositions about the past or
for propositions entailing the existence of persons distinct from myself; as before,
these will not be probable with respect to what is properly basic.2
Having rejected classical foundationalism's criteria for proper basicality, Plantinga goes on to suggest that belief in God might be properly
basic. For example, such beliefs might be grounded in various kinds of
experiences of God.
Upon having experience of a certain sort, I believe that I am perceiving a tree.
In the typical case I do not hold this belief on the basis of other beliefs; it is
nontheless not groundless. My having that characteristic sort of experience . . .
plays a crucial role in the formation of that belief. It also plays a crucial role in
its justification. . . .
2

Ibid., pp. 59-60.

223

Now similar things may be said about belief in God. When the Reformers
claim that this belief is properly basic, they do not mean to say, of course, that
there are no justifying circumstances for it, or that it is in that sense groundless
or gratuitous. Quite the contrary, Calvin holds that God "reveals and daily
discloses himself in the whole workmanship of the universe," and the divine art
"reveals itself in the innumerable and yet distinct and well ordered variety of
the heavenly host." God has so created us that we have a tendency or disposition
to see his hand in the world about us.3

The idea is that beliefs about God might be evidentially grounded in

religious experience, much as beliefs about material objects can be evidentially grounded in sensory experience. The logical space for this kind
of approach to the epistemology of religious belief is created by a
rejection of classical foundationalism. That position is not rejected in an
ad hoc manner, however. Rather, it is rejected by showing that the
position has unacceptable skeptical consequences that is, consequences
that are unacceptable even for the person who thinks that beliefs about
God are not rational.
The position that there might be a perception of God is developed
further by Alston. Before turning to Alston's discussion, however, I want
to consider some remarks that Plantinga makes about methodology in
epistemology. According to Plantinga, the proper way to proceed in
epistemology is by means of Chisholm's particularism. We are to consider a range of cases that seem to be instances of beliefs having some
positive epistemic status, and a range of cases that seem not to be. We
then look for theories of knowledge and evidence that explain our
intuitions about these cases. A question arises, however, regarding cases
where our intuitions conflict with those of others. What happens if
religious believers think that a range of beliefs about God are rational,
or even amount to knowledge, but non-believers refuse to grant those
same beliefs about God any such status? Plantinga's response is to adopt
a kind of methodological pluralism.
And hence the proper way to arrive at such a criterion is, broadly speaking,
inductive. We must assemble examples of beliefs and conditions such that the
former are obviously properly basic in the latter, and examples of beliefs and
conditions such that the former are obviously not properly basic in the latter.
We must then frame hypotheses as to the necessary and sufficient conditions of
proper basicality and test these hypotheses by reference to those examples. . . .
But there is no reason to assume, in advance, that everyone will agree on
3 Ibid., pp. 79-80.

224

the examples. The Christian will of course suppose that belief in God is entirely
proper and rational; if he does not accept this belief on the basis of other
propositions, he will conclude that it is basic for him and quite properly so.
Followers of Bertrand Russell and Madelyn Murray O'Hare may disagree; but
how is that relevant? Must my criteria, or those of the Christian community,
conform to their examples? Surely not. The Christian community is responsible
to its set of examples, not theirs.4
It seems to me that we are not forced into any such pluralism,
however. This is because a good theory of knowledge ought to explain
why our intuitions disagree, when they do. That is, all parties to the
dispute ought to look for a theory of knowledge that (a) explains clear
and uncontroversial cases, (b) explains cases where our intuitions are
vague, and (c) explains the cases where our intuitions disagree.
In a different essay, Plantinga himself provides us with a nice illustration of how this could work. There he defends a general theory of
knowledge which is a version of agent reliabilism. According to that
theory, a belief has the kind of positive epistemic status required for
knowledge just in case, roughly, it is produced by reliable faculties
functioning properly in an appropriate environment. On this view,
someone who believes in God on the basis of religious experience can
have knowledge thereby, so long as the belief is the product of some
properly functioning faculty, in this case a sensus divinitatus, so to speak.
But of course individuals will disagree over whether a given belief is
such a product, or even could be. Religious believers, or at least some
of them, will think that human beings are so designed and that therefore
such beliefs are possible. Freudians and Marxists, or at least some of
them, will think that we are not so designed, and that therefore such
beliefs are not possible. Rather, beliefs about God must be the result of
a cognitive ma/function. But notice that this is a disagreement over the
non-epistemological facts. Both the atheist and the theist could agree on
Plantinga's agent reliabilism, and precisely because it explains everyone's
intuitions involved, including such disagreements as arise.
Plantinga ends his discussion by noting that disputes over the rationality of religious belief are not merely epistemological they involve an
ontological or metaphysical dimensions as well:
What is rational depends upon what sort of beings human beings are; and what
you properly take to be rational, at least in the sense in question, depends upon
4

Ibid., pp. 76-77.

225

what sort of metaphysical and religious stance you adopt; it depends upon what
kind of beings you think human beings are, and what sorts of beliefs their noetic
faculties will produce when they are functioning properly.5
I fully agree, but we must keep in mind exactly where the metaphysics
comes in. It is not in determining which intuitions are to be explained,
and it is not in determining which theory of knowledge and evidence
best explains those intuitions. It does not even come in when we apply
the theory to individual cases, so long as those cases are hypothetical
ones that we can fully describe. Even a staunch atheist might agree that
belief in God would have positive epistemic status in cases where the
universe is as the theist describes it. And even a committed theist might
agree that belief in God would lack positive epistemic status in cases where
the universe is as the atheist describes it. Rather, one's metaphysics comes in
only when one decides whether particular, actual beliefs satisfy the
conditions that the preferred theory lays down. But again, that will not
be an epistemological dispute, except in an extended sense. Therefore,
people with quite different metaphysical positions can engage in strong
particularism and can thereby come to agree on the essentially epistemological questions. Methodological pluralism is not warranted by a
plurality of intuitions, or even by a plurality of metaphysical positions.6
2. ALSTON'S DEFENSE OF RELIGIOUS PERCEPTION
Alston develops systematically the idea that a perception of God is
possible.7 One objection to this idea that immediately arises is that
religious experience is, in itself, a merely subjective phenomenon. In
order for such an experience to justify one's beliefs about God, one must
have some good argument for believing that the experience is veridical.
In other words, one must have some good reason for believing that the
experience is of what one takes it to be. But no such argument is
5

Plantinga, "Epistemic Probability and Evil," Archivio di Filosofia 56 (1988): 557-584, at

p. 583. Reprinted in Daniel Howard-Snyder, ed., The Evidential Argument from Evil (Bloomington: Indiana University Press, 1996).
6 The point is less straightforward with respect to subjective justification, but I think that it
can be made even here. So, e.g., disputes about what beliefs are subjectively justified might
be explained in terms of different opinions regarding the availability and strength of counterevidence, or about what norms are typically countenanced, or what dispositions are typically
manifested, in conscientious thinking.
7 In William Alston, "Christian Experience and Christian Belief," in Plantinga and Wolterstorff (1983), and Perceiving God.

226

forthcoming, the objection goes, since a putative experience of God

could be explained in any number of other ways.
Alston responds to this objection by noting an analogous problem for
empirical perception.
Around about the seventeenth century someone had the bright idea of
remarking that whereas we had all along been supposing that any normal adult
human being could get plenty of empirical data of the form "This log is
burning" or "I am seated before the fire in a dressing gown" just by opening
her eyes, these supposed data are really hybrids. There is indeed a real datum of
the form "I seem to see a log burning" or "A burning-logish sense datum is in
my visual field" or "I am being appeared to log-burningly." This datum gets
conjoined with an explanation, namely, that it is generated, by a psychological
process we will pass over in discrete silence, by an actual physical log that is
actually burning, to produce the hybrid we uncritically take as the datum.
Hence, before we can use our experience as a ground for supposing that there
are such things as burning logs, or that a particular log is burning at a particular
time, we must find some way of justifying this particular way of explaining the
occurrence of sense data.
It is obvious that this familiar move in the epistemology of sense perception
is exactly parallel to the challenge we are considering to the claims of religious
experience. If we can figure out what to do in this case, it may help us with
our original problem.8
Alston notes that attempts in the history of philosophy to meet the
challenge head-on have failed miserably. A better approach in the epistemology of empirical perception is to hold that we perceive material
objects directly.
[O]ur only hope is to reject the bifurcation alleged in the challenge and seek to
justify the claim that we do encounter independently existing physical objects
in our experience. Put the matter in a somewhat different way. The question
concerns the justifiability of a certain practice the practice of forming physicalobject beliefs directly on the basis of perception rather than as an explanation of
what is perceived or experienced.9
An adequate (i.e., non-skeptical) theory of empirical perception requires
that we think of perception as non-inferential: quite immediately and
without anything like an inference, we take sensory experience to be
8 Alston, "Christian Experience and Christian Belief," p. 108.
9 Ibid., p. 109.

227

the experience of physical objects, and we form our beliefs accordingly.

Alston suggests that we consider this model for religious perception as
well. Just as we form beliefs about physical objects directly on the basis
of sensory experience, some people form beliefs about God directly on
the basis of religious experience.
The next question is whether this practice gives rise to beliefs that are
epistemically respectable. Granting that Alston has correctly described
the psychological process by which some people form some of their
beliefs about God, does the practice of forming beliefs in that way give
rise to justification and/or knowledge? Alston does not think that the
analogy to empirical perception in itself shows that it does. Rather, it
opens up the possibility. We have seen that thinking of empirical perception as inferential dooms one to skepticism about empirical perception. Likewise, thinking of the perception of God as inferential dooms
one to skepticism about religious perception. In both cases, this is because there are unanswerable skeptical objections to the possibility of an
adequate inference. But denying that there need be an inference from
religious experience to divine reality opens up the logical space. As with
empirical perception, it does not show that religious perception is actual,
but it removes a skeptical obstacle that would otherwise show that it is
impossible.
To take mystical experience as a form of perception is not to beg the question
as to the upshot of the inquiry. It still could be true that sense perception is
genuine perception of its putative objects, whereas mystical perception is not,
and that sense perception provides justified belief and knowledge about its
objects whereas mystical perception yields no such results for any beliefs about
anything. The point is only that the problems, both as to the nature of the
perception and as to the epistemic status of the perceptual beliefs stemming
therefrom, arise in the same form for both. Whereas on the widespread view
that mystical experience is to be construed as purely subjective feelings, sensations, and the like, to which supernaturalistic causal hypotheses are added, the
issues concerning the two modes of experience will look very different (unless
one is misguided enough to treat sense perception in the same fashion).10
So far so good. But is the perception of God actual, in an epistemic
sense of perception? In other words, do some people actually perceive
God, and on this basis have justified beliefs and knowledge about God?
In general, one perceives an object only if that object exists and one is
appropriately related to it. For this reason it is impossible to argue that
10

Alston, Perceiving God, pp. 6667.

228

some people actually perceive God without assuming that God exists
and is experienced by people in the required way. But even without
making such assumptions, there is still a lot that can be said about the
epistemology of religious perception. We can look at the practice of
forming beliefs about God directly on the basis of religious experience
(call this RP, for "religious practice"), and we can compare this to the
practice of forming beliefs about material objects directly on the basis of
sensory experience (call this SP, for "sensory practice"). Alston's strategy
throughout is to argue that the two kinds of perceptual practice are in
the same epistemic boat.11
Consider first the question of whether the two practices are objectively reliable, or truth conducive. In both cases this will be a matter of
what the facts are: it will depend on whether God (or the material
world) exists, and whether beliefs formed on the basis of the practice
hook up with the relevant reality in a reliable way. But in both cases, if
the relevant reality exists, and if that reality is related to us in the right
way, then the relevant practice is objectively reliable. Moreover, engaging in the relevant practice could constitute a reliable perception of the
reality in question.
Consider next whether we can prove or establish that the practices
are objectively reliable. Alston argues that, again, RP and SP are in the
same boat. In both cases it is impossible to establish the reliability of the
practice without engaging in a particular kind of circularity. Specifically,
we cannot establish the reliability of the practice without using premises
that are deliverances of the practice in question. Thus we can establish
the reliability of empirical perception, but only by using empirical perception. If we want something more than this for example, an argument that is independent of any empirical premise then we will be
disappointed. Alston takes it that the history of philosophy teaches us
this lesson; we need only look at the litany of failed attempts to establish
the reliability of empirical perception in this fashion.
The point is significant, because it blocks an otherwise plausible
objection to the possibility of perceiving God. RP cannot be shown to
be a reliable process, unless, of course, one is allowed to use beliefs
resulting from the practice itself as premises. But this kind of epistemic
circularity cannot count against the epistemic efficacy of RP. For if it
11 Alston tends to talk in terms of "Christian practice" and "Christian mystical practice," but
as he notes himself, there is nothing particular about Christianity that figures into his account of religious perception. I therefore use the term "religious practice."

229

did, an analogous objection would be sound against SP as well. Since

skepticism with regard to empirical perception is assumed to be false, we
may draw the conclusion that epistemic circularity of the sort just described cannot ground a skeptical conclusion with regard to religious
perception.
The reliability of SP cannot be established without epistemic circularity. On the other hand, there is no good reason to think that SP is
unreliable. Alston argues that RP is once again in the same boat epistemically. His reasons for this are somewhat involved, and going into
them in detail is beyond the scope of the present discussion. The important point for our purposes is the main structure of Alston's argument.
Once again, his main idea is that an adequate understanding of religious
perception should be based on an adequate understanding of empirical
perception. In particular, we cannot reject the idea of religious perception on the basis of considerations which, if sound, would show empirical perception to be impossible as well.
The implicit assumption of Alston's entire discussion, of course, is
that perceptual knowledge and justified belief in the empirical realm are
possible. In other words, Alston's methodology is an instance of strong
particularism: we assume for the sake of our investigation that empirical
knowledge and justified belief are possible, and we quite appropriately
reject assumptions about knowledge and evidence which are inconsistent
with this.
Finally, throughout Alston's discussion he considers several objections
to RP, all to the effect that the practice has some epistemically crippling
defect. Some of these he classifies as instances of "epistemic imperialism." These objections reject RP for not having some feature of SP,
thereby making the features of SP normative for all epistemically legitimate practices. Other objections, however, are classified as instances of
a double standard. Here the idea is that RP is rejected for having some
feature (or lacking some feature) that SP has (or lacks) as well.12 The
way that Alston answers these latter objections again demonstrates the
methodology of strong particularism. We begin by assuming that perceptual knowledge and justified belief are possible in the empirical realm.
On this basis, we can safely conclude that SP's having (or lacking) some
feature does not have skeptical consequences. But then RP's having (or
lacking) the same feature cannot have skeptical consequences for religious belief either.
12

Alston, Perceiving God, p. 249.

230

Of course, this line of reasoning does not establish that RP is an

epistemically efficacious practice, even when granted the assumption that
SP is. No matter how many features RP and SP share, there might be
some further, relevant feature that the practices do not share, and in such
a way that R P ends up being epistemically subpar. But the methodology
that both Alston and Plantinga adopt sets up an interesting and difficult
dialectic for the skeptic about the rationality of religious belief: When
raising objections to the rationality of religious belief, such a skeptic
must be careful not to issue in a more general, unacceptable skepticism.
That is, the religious skeptic must be careful not to issue in a skepticism
that is unacceptable even to himself. Avoiding this pitfall ends up being
far more difficult than one might first expect, as the discussions by Alston
and Plantinga testify.13

//. Moral Epistemology

We have seen that one major motivation for thinking of perception as
non-inferential is to avoid skepticism. If perceptual knowledge requires
some good inference from a relevant kind of experience to the relevant
kind of reality, then skepticism about perceptual knowledge is correct.
This is because there are unanswerable skeptical arguments which show
that such an inference is impossible. However, we have also seen that
thinking of perception as non-inferential poses a different kind of problem. Specifically, it would seem that our perceptions are theoretically
loaded, in the sense that they seem to depend on background beliefs,
prior assumptions, expectations, special training, and the like. The problem, then, is this: How can we account for the theoretical character of
perception, if perceptual beliefs are not inferred from other beliefs that
act as their evidence? This problem is nowhere more pronounced than
in the case of moral perception. Even in cases that are plausibly described
13 This same dialectic is also manifested in recent discussions of the problem of evil. See, e.g.,
various essays in Howard-Snyder, The Evidential Argument from Evil, cited in n. 5. For yet
other instances of the same dialectic, see Nicholas Wolterstorff, "Can Belief in God Be
Rational Even If It Has No Foundation?", in Plantinga and Wolterstorff (1983); Wolterstorff, "Epistemology of Religion," in Greco and Sosa (1999); and various works by John
Hick, esp. An Interpretation of Religion: Human Responses to the Transcendent (New Haven:

Yale University Press, 1989). Some relevant selections from this work are reprinted in
"The Rationality of Religious Belief," in R. Douglas Geivett and Brenden Sweetman,
eds., Contemporary Perspectives on Religious Epistemology (Oxford: Oxford University Press,

1992).

231

as "seeing" that some action is wrong or "feeling" that some action is

right, it would be hard to deny that such moral experiences are shaped
by prior moral judgments, background assumptions, and so forth.
Perhaps this kind of consideration more than any other has made
theories of moral perception unpopular. Nevertheless, in Part II of the
chapter I want to argue that an adequate theory of empirical perception
opens up the possibility of moral perception. Such a theory helps us to
address exactly the present concern about the theory-laden character of
perception in general. It also helps us to answer some other objections
that have traditionally been raised against moral perception. Before we
can turn to that argument, however, it will be necessary to engage in
some preliminaries. In particular, I want to say a bit more about what I
mean by moral perception.
1. THREE KINDS OF MORAL EPISTEMOLOGY

Theories in moral epistemology can be divided into three kinds. Since

the publication of John Rawls' A Theory of Justice, the most prominent
position in moral epistemology has been evidentialism.14 According to
this position, a moral belief is justified (or perhaps amounts to knowledge) only if it is supported by adequate evidence. Such evidence is
construed as other justified beliefs or knowledge, and the general shape
of the position is usually coherentist. That is, most evidentialists think
that all justified moral beliefs require other moral beliefs for their evidence, and so to avoid a regress of justification they adopt the position
that moral beliefs can lend each other mutual support. A second position
in moral epistemology, now less popular than it once was, is intuitionism. Sometimes intuitionism is characterized as any position which holds
that moral beliefs can be non-inferentially justified. Similarly, the term
"moral sense" has often been used as a synonym for "moral intuition."
But it seems to me that this terminology is inadequate, for it fails to
distinguish between two very different ways that a belief might be noninferentially justified. We may see this by considering four distinctions:
14 John Rawls, A Theory of Justice (Cambridge, MA: Harvard University Press, 1971). Rawls'
book has had a tremendous influence on moral epistemology, despite the fact that it is not
a work in the epistemology of moral belief. Rawls' account of "reflective equilibrium"
concerns criteria for moral theory construction, rather than criteria for justification or
knowledge. For a persuasive argument that this is the case, together with an excellent
discussion of relevant issues, see Michael DePaul, "Reflective Equilibrium and Foundationalism," American Philosophical Quarterly 23 (1986): 59-69, and Balance and Refinement:
Beyond Coherence Methods of Moral Inquiry (London: Routledge, 1993).

232

a. A belief is non-inferential if and only if it is not inferred from other beliefs.

A belief is inferential if and only if it is inferred from other beliefs.
b. A belief is a priori if and only if it does not depend on experience for its
evidence.
A belief is a posteriori if and only if it does depend on experience for its
evidence.
c. A belief is necessary if and only if it is logically impossible that it is false.
A belief is contingent if and only if it is logically possible that it is false and it
is logically possible that it is true.
d. A belief is general if and only if it does not refer to any particular existing
thing.
A belief is particular if and only if it does refer to a particular existing thing.

The first distinction is psychological; it concerns how a belief is formed.

The second is epistemological; it concerns the kind of evidence that a
belief has. The third and fourth distinctions are logical; they refer to the
logical properties of a beliefs content. Now it seems to me that intuitions are properly understood as non-inferential, a priori, necessary and
general. This accords with the idea that simple mathematical truths and
simple logical truths are paradigm cases of contents for possible intuitions. It also accords with the common understanding that intuited truths
are self-evident, in the sense that understanding them is sufficient for
being justified in believing them, or knowing them. On the other hand,
perceptual beliefs are properly understood as non-inferential, a posteriori,
contingent, and particular. When one perceives that something is the
case, this involves some sort of experience as one's evidence, and it
concerns some contingent matter of fact about a particular existing thing
or things.
One might think about moral knowledge on either of these very
different models. One might hold that there are moral intuitions, and
cash this out on an analogy with mathematical and logical intuitions, or
one might hold that there are moral perceptions, and cash this out on
an analogy with empirical perceptions.15 In the remainder of this chapter
I want to pursue the latter alternative. What I mean by "moral perception," then, is a process by which moral beliefs concerning contingent
truths about particular existing things are formed directly (i.e., noninferentially) on the basis of moral experience. Moreover, in moral
perception such experience serves as the evidential grounds of the moral
15 One might also hold both views, characterizing the foundations of moral knowledge as
including both moral intuitions and moral perceptions.

233

beliefs it produces, thereby giving rise to justification or even knowledge.16 An example of moral perception, on this model, would be seeing
that a particular action for example, some boys torturing a cat - is
wrong. Such a belief is particular rather than general, in that it refers to
particular existing things; in this case the boys, the cat, and the action
they are involved in. The belief also expresses a contingent truth rather
than a necessary one, in that its content is that this individual action is
wrong, and it is only a contingent truth that this action took place at all.
Finally, I will argue that such a belief can be formed directly on the basis
of a moral experience of the action, and that the belief can be thereby
justified, or can even amount to knowledge.
One reason for pursuing this kind of position is that it seems to be
the most adequate phenomenologically. Seeing that an action is wrong,
or that a man is trustworthy, or that a child is innocent seems to be an
adequate description of a fairly common sort of occurrence. On the
other hand, it seems to be phenomenologically inadequate to characterize many moral judgments in terms of an inference from evidence. At
least in a broad range of cases, we precisely do not infer that some action
is wrong from a general rule. What would such a rule be, and who but
a philosopher even thinks in terms of such rules? Neither do we typically
judge that some person is trustworthy by means of an implicit theory
about what trustworthy persons look like, which we then use to infer
that this particular person, who looks so and so, must be trustworthy.
Rather, a person often "strikes" us as trustworthy; she has a certain look
to her, which we probably could not articulate, on the basis of which
we form our beliefs and act. Or perhaps a voice sounds insincere, or
caring, or threatening. But who among us, except perhaps the odd
phenomenologist or stage actor, has beliefs about what such a look or

16 Intuitionism in moral epistemology has recently been defended by Robert Audi in a number of places. Audi's work also contains an excellent treatment of a broad range of relevant
issues in moral epistemology. E.g., much of what he says in defense of intuitionism goes
toward a defense of non-inferential moral knowledge in general. See his "Intuitionism,
Pluralism, and the Foundations of Ethics," in Walter Sinnott-Armstrong and Mark Timmons, eds., Moral Knowledge? (Oxford: Oxford University Press, 1996); and "Moral Knowledge and Ethical Pluralism" in Greco and Sosa (1999). For some recent defenses of moral
perception views, see Michael DePaul, "Argument and Perception: The Role of Literature
in Moral Inquiry," Journal of Philosophy 85 (1988): 552-565, and Balance and Refinement;

John McDowell, "Values and Secondary Qualities," in Ted Honderich, ed., Morality and
Objectivity (London: Routledge and Kegan Paul, 1985); and William Tolhurst, "On
the Epistemic Value of Moral Experience," Southern Journal of Philosophy 29, supplement
(1990):67-87.

234

sound amounts to, or believes that such and such an experience is a

reliable indication of such and such a moral property?
A theory of moral perception, on which some moral beliefs are
formed directly on the basis of experience, seems to offer a better
description of the psychology of moral belief formation. What we would
like is a theory that also gives us a plausible account of the epistemology
of such beliefs. In other words, we want a theory that shows how beliefs
so formed can be epistemically respectable.
My strategy for defending such a theory involves three stages. In
Section II.2, I review some of our conclusions regarding empirical perception from previous chapters, and I put forward a positive account of
empirical perception, perceptual justification, and perceptual knowledge.
The account I defend makes use of some recent work in the psychology
of perception. In particular, this work helps us to answer the problem
posed earlier, regarding how perception can be both non-inferential and
theoretically loaded. In Section II.3, I use this account of empirical
perception to put forward a theory of moral perception. I argue that
once we have an account of empirical perception that is both epistemically and psychologically adequate, an analogous account of moral perception suggests itself rather straightforwardly. Finally, in Section II.4, I
address some traditional objections to the possibility of moral perception.
My purpose in doing so is not to give a full-blown defense of moral
perception. Rather, it is to further the case that strong particularism
allows progress in moral epistemology. Once we have a theory of empirical perception that is immune to skeptical objections, we can see
why some traditional objections to moral perception are unsound.
2. A THEORY OF EMPIRICAL PERCEPTION
We already know that empirical perception should be understood noninferentially. We form beliefs about material objects immediately on the
basis of sensory experience, without anything like an inference from a
general rule, or knowledge about the character of the experience, or
knowledge that experience with such and such a character is a reliable
indication of such and such a material object.
How should we think about the sensory experience that is involved
in such a process? First, it is clear that sensory experience involves a
phenomenal content. When we see a material object, it has a certain
look to it phenomenally. Alternatively, an object can have a particular
sound, or smell, or taste, or feel to it. It seems to me, however, that in

235

the typical case sensory experience also has a conceptual or representational content. To use a now familiar phrase, seeing is (at least often)
"seeing as." This can be generalized to the other sensory modalities as
well. At least in the typical case, hearing is "hearing as," smelling is
"smelling as," and so forth. To illustrate this idea we may consider some
atypical cases in which we do not experience a thing as what it is. We
can then contrast these with more typical cases.
First, consider a case where you smell something vaguely familiar but
cannot quite say what the thing is that you are smelling. You might
concentrate carefully on the quality of the odor, trying to figure out for
some time what the source of it is. Contrast this with walking into the
house and smelling tomatoes and garlic cooking on the stove. In this
case there is no awareness of the experience as an isolated aspect of your
thinking you simply smell the tomatoes and garlic cooking as tomatoes
and garlic cooking.17
Consider next the case of hearing a sound coming from outside and
trying to determine its source. Again, one attends to the quality of the
experience so as to figure out whether it is from a car, a truck, a
motorcycle, or something else. Compare this with hearing the voice of
your spouse or your child. Here again there is nothing like an isolated
experience, separated off from other aspects of one's thought. One hears
the voice as the voice of the loved one. Even more likely, one simply
hears the voice as the person, and does not even think of the voice as a
voice.
Based on this kind of investigation into the phenomenology of perception, I assume that when one sees a tree (or smells her dinner, or
hears his child) the sensory experience involved in this has a conceptual
as well as a phenomenal content. To perceive that there is a tree bearing
apples by the brook, for example, is to be appeared to phenomenally in
a particular way, and to take this phenomenal appearance as that of a
tree bearing apples by the brook. One then forms a perceptual belief
with this content immediately on that basis. On this way of thinking,
the psychological and evidential grounds of empirical perceptual beliefs
is thick experience rather than thin, and it is the experience itself rather
than beliefs about the experience.
This way of thinking about perception has distinct advantages over
inferential and simple causal accounts. Over inferential accounts, it
17 In the example I am assuming that you are Italian, or at least sufficiently familiar with Italian
cooking. If not, substitute your own example.

236

shows how phenomenal experience can be involved in the evidential

grounds of perceptual beliefs, but not by figuring into some premise that
must now be used in an inference. Over simple causal accounts, it allows
that phenomenal experience really does play an evidential role in perception, and is not merely a causal antecedent. Rather, we may think of
the relation as broadly semiotic, as opposed to either inferential or
merely causal. Sensory appearings are meaningful signs, giving rise immediately to perceptual beliefs with the same contents. These meaningful signs can be analyzed into phenomenal content and conceptual content, but qua evidential grounds they are always already meaningful.
Again, this explains why bare phenomenal qualia cannot serve as evidence alone, but it also gives them more than a merely causal role in
perception.
Together with the results of Chapters 2 through 8, these considerations suggest the following account of empirical perception, perceptual
justification, and perceptual knowledge.
(P) S perceives X as F if and only if (a) X is F, (b) X appears phenomenally to S as X normally would if X were F, and (c) S takes this
phenomenal appearing as an appearing of X as F.
(PB) S perceives that X is F if and only if (a) S perceives X as F and (b)
S believes that X is F on this basis.
(PJB) S has a justified perceptual belief that X is F if and only if (a) S
perceives that X is F and (b) S's perceiving that X is F is a result
of the cognitive dispositions that S manifests when S is thinking
conscientiously.
(PK) S has perceptual knowledge that X is F only if (a) S has a justified
perceptual belief that X is F and (b) the cognitive dispositions that
S manifests in believing that X is F are objectively reliable in S's
environment.
A few comments are in order. First, I am leaving unanalyzed the somewhat awkward locution "S takes this phenomenal appearing as an appearing of X as F." However, something can be said by way of clarification. Most importantly, such "takings" or "seemings" are intended to
somehow involve representational or conceptual content, namely the
content that X is F. Thus the intentional object of a perceptual taking is
the object being perceived, and not the appearance of the object being
perceived. As I have noted repeatedly, in the typical case people do not
think about appearances, or the way things appear, at all.

237

Second, the present account treats "perceives" and its cognates as

success terms: when we say that one perceives that something is the
case, for example, this implies that what one perceives really is the case.
If we use "perceives" in this way, then cases of perceptual illusion are
not cases of perception. We have to say, for example, that one seems to
perceive (but does not) that there is a pink elephant in the room.
Third, the account allows that there is a phenomenal aspect of perception, but it conceives of that in terms of normality. In other words,
it is not assumed that there is any intrinsic feature of a phenomenal
appearing that makes that appearing, say, the appearance of a tree. What
makes a phenomenal appearance the appearance of a tree is just the fact
that this is how a tree would normally appear phenomenally, relative to
the cognitive agent in question. By means of this same feature, the
account disallows that just any phenomenal appearing can ground a
perception, just so long as the agent takes it a certain way. Rather, the
appearing must be tied into the normal perceptual dispositions of the
agent.
Another feature of the account is that it places no restrictions on what
kinds of things can be perceived. Specifically, there is no implication
that only colors, sounds, smells, and other traditional "perceptual qualities" can be perceived. It leaves it open, for example, that one can see
that his friend is amused or hear that one of the pistons is not firing. It
seems to me that this is as it should be, since it is pre-theoretically
plausible that we can perceive such things.
The account of justified perceptual belief adds the requirement that
the representational aspect of the perception must also be tied into the
cognitive dispositions of the agent. Moreover, the dispositions that are
relevant are the ones that S manifests when properly motivated toward
the truth. This ensures that not just any perception can ground a justified perceptual belief, and it makes the account consistent with the
general account of subjective justification defended in Chapter 7. Finally, it should be noted that the account of perceptual knowledge
states only necessary conditions rather than necessary and sufficient
ones. This is in order to avoid making claims about the adequacy of
the account with regard to Gettier cases. I take it that Gettier cases
raise an entire range of issues which, for present purposes, can and
should be set aside.18
18 I discuss the relationship between agent reliabilism and Gettier problems in Part III of this
chapter.

238

On the present account perception is conceptually loaded. This is

because perceptual takings involve conceptual as well as phenomenal
content. But to say that perception is conceptually loaded is different
from saying that it is theoretically loaded. The latter claim is that perception is influenced by prior assumptions, background beliefs, expectations,
special training, and the like. So, for example, perceiving that some tree
is an elm might depend not only on one's sensory experience having a
certain conceptual or representational content, but also on one's prior
beliefs about trees or one's expectations about where elms can be found.
We still have not addressed this problem for a non-inferential theory of
perception. That is, how can prior beliefs, expectations, and the like
influence perception, even if perceptual beliefs are not inferred from
these?
One way that I have already suggested is through counter-evidence.
In order for our perception to be both reliable and subjectively justified,
cognitive beings like ourselves must be sensitive to evidence that contradicts the way things appear perceptually, and must be able to adjudicate
conflicts appropriately. Accordingly, the production of perceptual justification and knowledge, at least for beings like us, must involve other
beliefs that function in this way. A second suggestion that has already
been made is that background beliefs and the like "shape" our perceptual
dispositions; they function as causal antecedents to perception, influencing us to discriminate some objects rather than others, to pick out salient
information, to filter out the unimportant. But how this is all supposed
to work was left overly vague. If it could work only by having the sorts
of beliefs and inferences that we have deemed to be psychologically
implausible, then saying that the theoretical background shapes perceptual dispositions does not do the job it is supposed to do. What we need
is a model for how perception can be thus theoretically loaded, and yet
not in a way that invokes psychologically implausible beliefs and inferences.
Recent work in empirical psychology provides us with just such a
model. In what follows I will not argue that the model is correct, since
I do not have the expertise to adjudicate competing empirical theories
of perception. Instead, I will argue for a more cautious thesis: that if the
model is correct, then it shows how perception can be non-inferential,
and at the same time be the product of prior assumptions, theory,
background beliefs, special training, and the like. Even if the model is
not correct about how perception actually works, it at least demonstrates
the possibility of non-inferential, theory-laden perception; it demon-

239

strates that this idea is not incoherent or impossible, even if it does not
show h o w such perception actually works in us.
T h e main idea of Schema Theory in empirical psychology is that our
cognition involves "scripts" and "personae" through which information
such as sensory appearances is processed. T h e dispositions of perception
that result are loaded, to be sure. But they are best understood, I want
to argue, as being non-inferential; they ground a process very different
from inferring conclusions from premises. In the following passages
Richard Nisbett and Lee Ross describe h o w scripts and personae are
supposed to work:
To understand the social world, the layperson makes heavy use of a variety of
knowledge structures normally not expressed in propositional terms and possibly
not stored in a form even analogous to propositional statements. In describing
these cognitive structures we shall use the generic designation "schema" and
will comment in detail about only two types of schema event-schemas, or
"scripts," and person-schemas, or "personae."
A script is a type of schema in which the related elements are social objects and
events involving the individual as actor and observer. Unlike most schemas,
scripts generally are event sequences extended over time, and the relationships
have a distinctly causal flavor, that is, early events in the sequence produce or at
least "enable" the occurrence of later events. A script can be compared to a
cartoon strip with two or more captioned "scenes," each of which summarizes
some basic actions that can be executed in a range of possible manners and
contexts (for instance, the "restaurant script" with its "entering," "ordering,"
"eating," and "exiting" scenes).
Social judgements and expectations often are mediated by a class of schemas
which we shall term "personae," that is, cognitive structures representing the
personal characteristics and typical behaviors of particular "stock characters."
Some personae are the unique products of one's own personal experience (good
old Aunt Mary, Coach Whiplash). Others are shared within the culture or subculture (the sexpot, the earth-mother, the girl-next-door, the redneck, the
schlemiel, the rebel-without-a-cause). . . . In each instance the persona constitutes a knowledge structure which, when evoked, influences social judgements
and behaviors. Once the principal features or behaviors of a given individual
suggest a particular persona, subsequent expectations of and responses to that
individual are apt to be dictated in part by the characteristics of the persona.19
19 Richard Nisbett and Lee Ross, "Judgmental Heuristics and Knowledge Structures," in
Kornblith, Naturalizing Epistemology, p. 278, p. 280, and pp. 281-282.

240

The application to perception is straightforward. When we are seated

at the restaurant table and a man approaches, we perceive that the waiter
is coming. According to schema theory, this perception is not based on
an inference from other things we believe for example, that waiters
typically dress in such and such a manner, that there appears to be a man
dressed this way, and that therefore this must be the waiter. In other
words, we do not mount an inference from premises about the way the
man appears visually to a conclusion regarding who he is. Rather, we
are operating with a script that disposes us to expect a waiter in the
present time and place, and that invokes an immediate (i.e. noninferential) interpretation of present sensory cues. Or again, suppose that
we are working with the persona of the surly waiter. We will be apt to
interpret a tone of voice as impatient or a facial expression as sarcastic.
On the present proposal, such perceptions are not grounded in anything
like premises about vocal tone or facial expression, together with theories about what these usually signify. And rightly so, since we would be
hard-pressed to find such premises or theories in the typical restaurantgoer. Rather, sensory information is processed in such a way that we
make these interpretations more immediately, aided by some relevant,
currently evoked schema.
If this is right, then perception is non-inferential. As was explained
earlier, scripts and personae are not thought of as beliefs or assumptions,
as premises for inferences would have to be. They are not, for example,
judgments, that serve to entail or probability other judgments. Nevertheless, perception is theoretically loaded on this view. As was also
explained earlier, scripts and personae can be part of our cultural inheritance, the result of special training, or gleaned from previous life experience.
3. A THEORY OF MORAL PERCEPTION

The present account of empirical perception opens up the possibility of

moral perception straightforwardly. That is, it opens up the possibility
that we can sometimes see (quite literally) that an action is wrong, that
a child is innocent, or that a man is dishonest. The basic idea is that we
are equipped with moral scripts and moral personae. If all perception
involves schema-governed interpretation, then moral perception involving this kind of interpretation would be just another kind of perception.
For example, the moral perception that some man is dishonest would

241

not be essentially different from the empirical perception that some man
is a waiter.
We may pursue this idea further by considering the nature of moral
experience. Some moral epistemologists have thought of moral experience as moral emotions. On this view, affective reactions such as indignation, empathy, revulsion, and attraction ground appropriate moral
judgments about the objects of such emotions. For example, a feeling of
revulsion toward some action might ground a moral judgment that the
action is wrong. 20 Other philosophers have thought of moral experience
as moral seemings. Upon witnessing an action or meeting a person, the
object of one's experience appears morally a certain way. For example,
an action strikes one as tragic, or a person strikes one as kind and good.
On this view events and things can have a kind of "moral color" or
"moral feel" to them, and such experience can epistemically ground
moral judgments about the relevant objects.21
Finally, some philosophers have identified the experience involved in
moral perception with sensory experience. On this view moral properties supervene in some way on the natural properties of events and
things, and for this reason it is possible to experience their moral properties by experiencing their natural properties. For example, suppose, for
the sake of argument, that utilitarianism is correct, and that anything that
is pleasurable thereby has moral value. Given the requisite theoretical
background, one could then perceive that something is morally valuable
by perceiving that it is pleasurable.22
Without arguing against other possibilities, I want to defend a conception of moral experience that combines features of the last two
suggestions. Staying close to the analogy with empirical perception, I
suggest that moral experience involves both phenomenal and representational content. The phenomenal content is the same: it involves the
same sensory modalities that are normal for empirical perception. What
makes the experience distinctively moral is its representational content.
On this view, to morally perceive that something has some moral property is to perceive the thing empirically, but to take the phenomenal
appearing involved in a moral way as well. For example, to morally

20 This kind of view is defended by Tolhurst in "On the Epistemic Value of Moral Experience" and by DePaul in Balance and Refinement.
21 At times Tolhurst and DePaul characterize moral experience this way.
22 This view of moral experience is suggested but not defended in Audi, "Moral Knowledge
and Ethical Pluralism," Section 3.

242

perceive that a man is dishonest is to be appeared to phenomenally in a

particular way (a way in which dishonest men normally appear phenomenally), and to take that phenomenal appearing to be of a dishonest man.
To morally perceive that some action is tragic is to be appeared to
phenomenally in a particular way (a way in which tragic actions normally appear phenomenally), and to take the phenomenal appearing
involved to be the appearing of something tragic.
On this account moral experience is conceptually loaded. As was
already suggested, schema theory provides a way in which moral perception might be theoretically loaded a well. Again, the main idea is that
we have moral scripts and personae. In our cast of characters there is the
shifty lawyer, the cop on the take, the schoolyard bully, the vicious drug
dealer, the petty neighbor, and the greedy doctor. There is also the
selfless mother, the devoted teacher, the kind doctor, the courageous
cop, and the crusading lawyer. The list goes on and on. There are also
countless moral scripts. As with personae, many of these are highly
specific to time, place, and culture. Others are more universal, perhaps
showing up in the great literature of several distinct cultures. For example: courageous boy meets innocent girl, boy and girl fall in virtuous
love, boy and girl are separated by immoral parents (or appointed guardian), boy and girl are reunited as the result of selfless devotion to each
other. This is a big picture painted in broad strokes. But scripts can be
loaded with moral details and subtleties as well. Consider the Shakespearean tragedy or the Austenian novel.
We have seen that according to schema theory perceptual processing
can involve scripts and personae. Some feature of a situation activates a
relevant schema, and as a result we are disposed to see or hear things a
particular way. The present suggestion is that these might be moral ways
as well as empirical ways. Moral schema might influence one to see a
movement as aggressive, to hear a voice as threatening, or to feel a touch
as reassuring. If this is indeed the case, then the influences that moral
schema have on cognition are epistemically relevant. Certainly some
schema amount to no more than stereotypes and myths. In such cases
their influence would undermine the reliability of our moral judgments,
thereby having a negative epistemic effect. But it is also possible that
more apt schema contribute positively to the reliability of moral judgments, helping us to see or hear or feel what is actually the case, to
perceive what we otherwise would miss. Again, whether or not moral
schema do function in this way is largely an empirical question. But if
they do, then this explains how moral perception can be theoretically

243

loaded and yet non-inferential. And even if they do not, the present
discussion at least demonstrates that the idea in not incoherent.
If all of this is right, then moral perception is possible. Aside from its
subject matter, moral perception would have little to distinguish it from
perception in general, working pretty much the same way that empirical
perception does. Accordingly, the present account of moral perception
allows us to adopt without qualification the accounts of perception,
perceptual justification and perceptual knowledge that were put forward
in Section II.2. To say that S morally perceives that something is the
case is just to say that S perceives that X is F, where F is a moral
property. As in the empirical case, this will involve a sensory phenomenal appearing, but the relevant perceptual taking will involve a moral
representational content, namely that the thing perceived has (moral
property) F. The accounts of perceptual justification and perceptual
knowledge can then be carried over without qualification. They do no
more than add requirements placing the source of S's perception in S's
relevant cognitive character. But there is no reason why moral perceptions, as characterized above, could not be grounded in S's character in
these ways.
4. SOME TRADITIONAL OBJECTIONS TO MORAL
PERCEPTION

If the theory of perception presented here is correct then a number of

traditional objections to moral perception are removed. I will end this
section by looking at some of these, focusing on ones where it is
especially useful to pursue the analogy with empirical perception.
First, many philosophers have thought that moral perception requires
an occult faculty a mysterious moral sense, or some kind of moral
intuition. But again, if perception in general is understood along the
lines I have suggested, then no special moral faculty is required. Moral
perception would be just like all perception, being distinguished only by
the conceptual content of the judgments it produces, rather than by the
mechanism by which it produces them.
A second objection to moral perception is that moral judgments are
always culturally mediated. We never make moral judgments immediately, it is argued, but always in the thick context of our cultural
inheritance. This objection is easily answered in the present context, for
it is merely a version of the more general idea that moral perceptions
would have to be theoretically loaded. But if all perception, including

244

that of the empirical variety, involves the kind of theoretical background

invoked by the objection, then this cannot count against the possibility
of moral perception. Moreover, we have seen at least one way in which
perception can be theoretically loaded and yet satisfy the conditions on
knowledge regarding subjective justification and objective reliability.
Cultural influence is not incompatible with epistemic respectability.
A third objection claims that moral perception is impossible because
it is impossible to perceive high-level dispositional properties, as moral
properties are properly understood to be. Thus to say that someone is
honest, or kind, or selfish is to attribute a range of dispositional properties that cannot manifest themselves in a moment's experience. On the
current account of perception, however, it is in principle possible to
perceive high-level, dispositional properties. What properties can be
perceived is a matter of empirical fact, concerning the extent of our
cognitive powers. This is as it should be, I have suggested, since it is
pre-theoretically plausible that we can perceive any number of things.
For example, any one of us can see that a given vehicle is a firetruck.
Some of us can hear that the brakes need replacing or smell that the
carburetor needs tuning. Since all of these are high-level, dispositional
properties, it cannot count against the possibility of moral perception
that moral properties are as well.
The current objection might be pressed, however. It might be
claimed that, unlike in the empirical case, it is implausible that we have
the requisite background knowledge to perceive moral properties. The
idea is that the nature of moral properties is highly controversial. Even
if such properties are reliably tied to particular phenomenal cues, no one
understands in what ways this is so, and so no one is in a position to
employ such cues in moral perception. This way of stating the objection,
however, misunderstands how perception in general is supposed to
work. In no case does perception operate by noting the phenomenal
character of one's experience and by using general rules to infer dispositional properties. We have already seen that empirical perception cannot work this way, for lack of the requisite knowledge of one's experience, as well as of the required general rules. There is no reason to think
that moral perception would have to work this way either.
Let us try one more version of the objection. In empirical perception,
it is claimed, the perception of high-level properties such as being a
firetruck or needing brakes is acquired. Moreover, the acquired perception of high-level properties depends on the perception of low-level
properties of the object that are reliable indications of the high-level

245

ones. Perhaps this works by perceiving the low-level properties on

which the high-level ones supervene. In the moral case, however, no
one has the required knowledge regarding which lower-level properties
are a reliable indication of which higher-level properties. Presumably,
some perceivable natural properties are reliable indications of the higherlevel moral properties which are related to them, but not even moral
philosophers have a good idea of what these relations are.
This objection depends on the assumption that one must have knowledge of relevant relations, perhaps supervenience relations, in order to
perceive higher-level properties. But that assumption is implausible regarding empirical perception. It would seem that people can perceive
firetrucks, elm trees, computers, dogs, and much else without having
any knowledge whatsoever about what lower-level perceivable properties are related to these higher-level ones. At least this is so if we are
talking about propositional knowledge. Of course, people do have the
required procedural knowledge. This simply means that they know how
to make reliable perceptual judgments on the basis of the relevant perceptual cues. But there is no reason to think that people cannot have
such procedural knowledge in the moral case as well. It would seem that
some people can reliably judge that a person is kind by the way that he
sounds, and most people know that some things are wrong when they
see them.
The final objection against moral perception that I will consider
trades on the existence of widespread disagreement among moral judgments. If moral perception were a reality, it is argued, there would not
be so much moral disagreement. This objection assumes, however, that
perception must be innate, or at least very widely shared. It thinks of
perception as a natural faculty that, if possessed at all, will be possessed
by more or less everyone and more or less equally. On the theory of
perception that I have proposed, however, even empirical perception
can be the result of highly specialized training and other particularities
of one's circumstances. Perception is not a natural ability to merely
"read off' the properties of the world. Rather, it is at least often an
acquired ability to interpret the way things appear phenomenally. Consider that we have no more a natural ability to see that someone is a
waiter, or to perceive that he is being sarcastic, than we have to see that
someone is dangerous, or dishonest, or kind.
Therefore, our perceptual powers need not be widely shared nor
equally effective. Such a position is in fact required by the idea of expert
perception. Consider that not everyone can see what experts see. This is

246

a generally acknowledged fact for which a good theory of perception

ought to account. The current theory does well with it by allowing us
to think of moral perception as a kind of acquired, expert perception.
Not everyone is a reliable moral perceiver. But this does not entail that
no one is, or that some people are not more reliable than others.
Schema theory in empirical psychology again provides suggestions for
how this might be so. Specifically, those with the best scripts and
personae will be in a position to make the most reliable perceptions. For
example, consider the different schemas available to the tourist and the
seasoned New Yorker. Walking in the city, the tourist is ill equipped
with personae and scripts derived largely from movies and sensationalist
news reports. Accordingly, she might see a particular look as aggressive,
or a particular gesture as threatening, where the native New Yorker
would make no such moral interpretation. There are stories of tourists
being approached for directions or the time of day, and clutching their
purse.
Or consider an unusually affectionate child one who runs up to
hug and kiss strangers with little or no caution or discrimination. Many
people will see this behavior as carefree and innocent. But it can have
another moral color for a psychologist or social worker, for whom
medical evidence or testimony has evoked the persona of a sexually
abused child. The same actions, now understood as manifesting an
absence of normal boundaries, will present themselves as tragic, even
horrifying. In moral cases as in others, the fact that the expert disagrees
with non-experts, or sees differently than they do, does not count against
the knowledge of the expert.
What makes the expert's schemas better than those of the non-expert?
One answer is that the expert's scripts more accurately portray series of
events that we are likely to observe or be involved in, and her personae
more accurately portray characters we are likely to encounter. Another
possibility is that experts tend to have more schemas, and these enable
them to make finer discriminations among types of events and characters. Someone with more available schemas is less restricted by her
expectations than someone operating with fewer. Finally, individual
scripts and personae of experts might be richer, again allowing for finer
discriminations, this time within an invoked script or persona.
But however moral perception works, the important point is that
moral disagreement does not count against the possibility of moral perception, or against the possibility that such perception gives rise to
justified belief and knowledge in the moral realm. This is because per-

247

ceptual powers in general need not be widely shared and need not be
equally effective. Just as there can be acquired, expert empirical perception, this might be the case regarding moral perception as well.

III. General Conclusions

I will be very brief in drawing some general conclusions from the
foregoing study. First, I have been arguing that the analysis of skeptical
arguments deserves pride of place in the methodology of epistemology.
Specifically, skeptical arguments repay analysis by highlighting plausible
but mistaken assumptions about the nature of knowledge and evidence.
As I said earlier, I have been intending this thesis both descriptively and
prescriptively; I have been arguing that this is how many contemporary
epistemologists actually employ skeptical arguments, and that this is what
epistemologists ought to do.
Second, I have been arguing that taking skeptical arguments seriously
pushes us toward agent reliabilism. That theory explains why various
plausible assumptions of skeptical arguments are in fact mistaken. But
unlike other forms of reliabilism, it also preserves our persisting intuitions
that knowledge requires both (a) non-accidental reliability and (b) subjective justification. The broad picture of knowledge and evidence that
results is an externalist, contextualist foundationalism, which is also a
version of virtue epistemology. According to agent reliabilism, epistemic
notions such as justification, knowledge, and evidence are to be understood in terms of virtuous cognitive character, rather than the other way
around. For example, we do not first define a notion of good evidence,
perhaps in terms of its logical or quasi-logical relations to evidenced
belief, and then define virtuous character in terms of a disposition to
form one's beliefs on that kind of evidence. Rather, good evidence is to
be understood in terms of reliable and properly motivated character; it
is the evidence that a person with virtuous cognitive character would
use to form her beliefs. By reversing the direction of analysis in this way,
an agent-centered reliabilism is able to explain the kind of nonaccidentality required for knowledge, define a relevant sense of subjective justification, and do so in a way that avoids skeptical consequences.
Third, in defending agent reliabilism I have not been claiming to
defend a fully detailed theory of knowledge and evidence. Rather, I
have been arguing for a general direction in epistemology, or a general
framework within which a fully detailed epistemology must work. Ac-

248

cordingly, agent reliabilism is capable of different versions and varieties.

I have already mentioned Alston's social practice view, Plantinga's
proper functionalism, Sosa's perspectivism, and Zagzebski's neoAristotelian approach, and there are other possibilities as well. All of
these positions are versions of agent reliabilism, in that they agree that
knowledge and justified belief are grounded in the reliable dispositions
that make up the knower's intellectual character. Where they differ is
on the next level of analysis down, concerning the nature of those
dispositions. Another way to put the point is as follows: each of these
authors makes his or her position stronger by adding conditions on what
counts as virtuous character and, therefore, on what kind of agent
reliability is involved in knowledge.
The specific features of these more detailed positions might be necessary to address other important epistemological issues. For example,
aspects of Alston's social practice view might be needed to more fully
explain the ways in which knowledge is social. Another possibility,
however, is that by strengthening the conditions for knowledge these
authors make those conditions too strong. In other words, it is possible
that the conditions for knowledge and justification that I have been
defending, and that all of these authors either already share or could
easily endorse, are already sufficient to capture the ways in which knowledge must be objectively reliable and subjectively appropriate. We have
already seen that this is a plausible conclusion regarding Sosa's theory. In
Chapter 7 we noted that Sosa's notion of a perspective on one's faculties
is ambiguous between (a) a set of dispositional beliefs about one's faculties and (b) a set of dispositions which can be identified with one's
faculties. There I argued that requiring a perspective in the former sense
was too strong, because typically we have no such perspective. But
requiring a perspective in the second sense does not add anything to
agent reliabilism, since the requirement of a perspective then amounts
to no more than the requirement of a reliable character.23
A similar dialectic plays out with respect to Alston's and Plantinga's
theories. As we have seen, Alston claims that knowledge and justified
belief must be grounded in reliable social practices. But as with the first
interpretation of Sosa's perspectivism, this seems too strong. Why should

23 Sosa's account does add that the perspective must be formed in a reliable way, but again,
it is not clear that this requirement can be cashed out in such a way that it both (a) adds
something to the requirements already laid down by agent reliabilism and (b) remains a
plausible condition on justified belief and knowledge.

249

we deny knowledge to cognitive agents who are not part of a social

group, and who therefore do not engage in social practices at all? If such
an agent is nevertheless reliable, and if her beliefs are subjectively appropriate in the relevant ways defined, what motivation is there for denying
that she has knowledge among her true beliefs? At one point Alston
considers this kind of objection. He writes,
Why not take all practices to be prima facie acceptable, not just socially established ones? Why this prejudice against the idiosyncratic? . . . It is a reasonable
supposition that a practice would not have persisted over large segments of the
population unless it was putting people into effective touch with some aspect(s)
of reality and proving itself as such by its fruits. But there are no such grounds
for presumption in the case of idiosyncratic practices.24

It is not clear that Alston's supposition is warranted, nor is it clear why

one could not have similar grounds for accepting an idiosyncratic practice. But putting these issues aside, Alston's rationale for distinguishing
social and non-social practices makes sense only in the context of his
discussion of practical rationality. Regarding epistemic justification and
knowledge, Alston embraces a reliabilist account; what matters for justification and knowledge is that one's belief-forming practices are in fact
reliable. In this context Alston explicitly rejects, and quite rightly, any
requirement that one have reasons for believing one's practices are
reliable.
We might conclude, therefore, that the social aspect of social practices
does no work, and as such has no motivation, in Alston's conditions for
epistemic justification and knowledge. Quite the contrary: including a
social aspect in these conditions threatens to make them too strong,
entailing that individuals who do not engage in group practices cannot
have knowledge or epistemically justified belief. On the other hand, by
taking the "social" out of social practices, we effectively remove any
sense in which Alston's conditions add something to agent reliabilism as
I have characterized that position. As we have already seen, to say that
an agent engages in reliable practices is just to say that she manifests
reliable dispositions in forming her beliefs that is, it is to say that she
displays a reliable cognitive character.
Finally, consider Plantinga's claim that knowledge is grounded in
properly functioning faculties, and that proper function is to be understood in terms of functioning according to a design plan. Once again the
24

Perceiving God, pp. 169-170.

250

added conditions seem too strong, and so once again there is pressure to
weaken what one means by them. As it turns out, Plantinga allows that
cognitive faculties might be "designed" by evolution, or by other nonintelligent forces. But this effectively reduces proper function to reliable
function, and so effectively reduces Plantinga's position to agent reliabilism simpliciter.25 All of this suggests that agent reliabilism already lays
down conditions that are sufficient for objective reliability and subjective
appropriateness. It is possible that different versions of the position do
less by trying to do more.
Even if this is so, however, it remains the case that the position I
have been defending does not constitute a complete epistemology. This
is because there are important issues regarding the nature of knowledge
and evidence that the position does not address. In closing I will consider
two of these.
First, the conditions for justified belief and knowledge laid down in
Chapters 7 and 8 are not adequate for addressing Gettier problems. For
this reason, I have been careful to characterize the account of knowledge
only in terms of necessary conditions rather than sufficient ones. To see
why the conditions are not adequate, consider the following Gettiertype example from Zagzebski:
Suppose that Mary has very good eyesight, but it is not perfect. It is good
enough to allow her to identify her husband sitting in his usual chair in the
living room from a distance of fifteen feet in somewhat dim light. . . . Of course,
her faculties may not be functioning perfectly, but they are functioning well
enough that if she goes on to form the belief My husband is sitting in the living
room, her belief has enough warrant to constitute knowledge when true and we
can assume it is almost always true. . . . Suppose Mary simply misidentifies the
chair sitter, who is, we'll suppose, her husband's brother, who looks very much
like him. . . . We can now easily amend the case as a Gettier example. Mary's
husband could be sitting on the other side of the room, unseen by her.26
Of course the point is that Mary does not have knowledge in the case,
even though her belief is true and is also reliably formed and subjectively
appropriate. Accordingly, the conditions that agent reliabilism sets down
25

Warrant and Proper Function, pp. 1314. In truth, the issue with regard to Plantinga is more
complicated than I have presented it. This is because Plantinga argues that the notion of
proper function cannot be given a naturalistic analysis. See esp. Ch. 11.
26 Zagzebski, Virtues of the Mind, pp. 285-287. Zagzebski is discussing Plantinga's proper function view, but the example is potentially a problem for any version of agent reliabilism.

251

as necessary for knowledge are not also sufficient for knowledge; something else must be added.
As far as I can see, there are three ways that agent reliabilism might
handle such cases. First, it is possible that the conditions for objective
reliability and subjective appropriateness laid down by the position can
be interpreted in a particular manner, thereby making them sufficient
for addressing this kind of case and other Gettier-type problems. This
strategy is represented by Sosa, who goes on to analyze agent reliability
in a way that is designed to do just this. The main idea is that reliability
is to be understood, at least partly, in terms of a special kind of tracking.
So understood, Sosa argues, we get a solution to standard sorts of Gettier
problems.27 A second possibility is that an adequate answer to Gettier
problems must add to the conditions for justification and knowledge
already laid down by agent reliabilism, but can do so by continuing to
draw on the special resources of virtue epistemology. This strategy is
represented by Zagzebski, who attempts to address Gettier problems by
means of the notion of an act of intellectual virtue. The main idea is
this: in cases where one's belief results from an act of intellectual virtue,
one believes the truth because one's belief is both reliably formed and
properly motivated. Placing this sort of condition on knowledge, Zagzebski argues, solves the problem concerning Mary, as well as other
Gettier-type problems.28 Finally, a third possibility is that an adequate
answer to Gettier problems is independent of both agent reliabilism and
virtue epistemology. Any number of solutions proposed over the past
four decades fall into this category.
A second issue that is not fully addressed in the preceding chapters
concerns the various ways in which justified belief and knowledge are
sensitive to context. We have already seen that agent reliabilism is
consistent with contextualism. Indeed, it would be incredible if agent
reliability, and therefore justified belief and knowledge, are not partly a
function of contextual features. However, there remain substantive issues
concerning how knowledge is affected by context. For example, there
are questions about which social factors affect reliability, and how, and
on what subject matters. Very little of what has been said in the preceding chapters bears directly on these important issues.
27 See Sosa, "Postscript to 'Proper Functionalism and Virtue Epistemology,' " in Jonathan
Kvanvig, ed., Warrant in Contemporary Epistemology (Lanham, MD: Rowman and Littlefield,
1996).
28 See Linda Zagzebski, Virtues of the Mind, and Zagzebski, "What Is Knowledge?", in Greco
and Sosa (1999).

252

Moreover, recent authors have argued that knowledge (or knowledge

attribution) is sensitive to context in a different sense.29 Specifically, it
has been argued that the standards for knowledge depend on context,
and so whether a knowledge-claim is true is relative to the context in
which the claim is made. It follows from this position that a sentence of
the form "S knows p " can be true when uttered by one person in her
context, and at the same time false when uttered by a different person
in a different context. For example, suppose that I make the claim that
traffic is light on the parkway right now. In a context where it does not
matter greatly whether I am right, the standards for knowledge are low.
Accordingly, you might truthfully say that I know what I claim about
the traffic. But suppose that another person is trying to decide what
route to take to the hospital, and that it is essential that she take the best
one. In this context the standards for knowledge are raised, and the
person might truthfully judge that I do not know what I claim.
Agent reliabilism is at least consistent with this sort of position. One
way to accommodate it is to say that the degree of reliability required for
knowledge changes with context. But another way is to say that the
range of reliability required for knowledge changes. It will be remembered that, in Chapter 8, I defined a cognitive ability in terms of success
across close possible worlds. It is plausible that context affects how far
out into relevant possible worlds one's reliability must extend.30 This
kind of consideration has been put forward to answer a certain type of
skeptical argument. Namely, we noted in Chapter 2 that the skeptical
argument from Descartes could be interpreted in the following way:
(D6)

1. A person can know that p is true only if she knows that every
possibility inconsistent with p is false.
2. The skeptical dream hypotheses are inconsistent with my beliefs
about the world, and I do not know that the skeptical dream
hypotheses are false.
3. Therefore, I do not know anything about the world.

I said in Chapter 2 that interpreting Descartes' argument this way made

it implausible. More specifically, I said that premise (2) of the argument
was implausible, because it seems that I do know that I am not a brain
in a vat, for example. But for those who do find (2) plausible, the
29

For example, Cohen, "How To Be a Fallibilist," and DeRose, "Solving the Skeptical
Problem." My discussion of contextualism in the rest of this section is indebted to both of
these authors.
30 A somewhat similar idea is put forward by DeRose in "Solving the Skeptical Problem."

253

contextualist has an answer. Namely, premise (2) is true in contexts

where the skeptical argument is being considered, and argument (D6) is
therefore sound. This is because that context drives the standards of
knowledge high. For example, the context requires that our cognitive
abilities extend all the way out to far-off worlds where the skeptical
hypotheses are true. But in non-philosophical contexts where knowledge-claims are usually made, the standards for knowledge are not nearly
so high. The result is that we do know many things about the world in
those contexts, and even if it is also true that in "philosophical" contexts
we do not.
Agent reliabilism, therefore, is at least consistent with contextualism
of various sorts. But which versions of contextualism are true, and how
we are to understand the details, depend on a variety of complex issues,
both in the sociology of knowledge and the philosophy of language.
Addressing these and other important issues would require a more complete epistemology than the one I have defended here. This being the
case, however, agent reliabilism nevertheless constitutes the framework
within which such further investigations should take place. This is because, first, rejecting a number of powerful skeptical arguments requires
that we adopt some form of reliabilism. Second, the most plausible
version of reliabilism is agent reliabilism. Further details in our epistemology, I conclude, should take into account these two results.

254

Bibliography

Alston, William. "Has Foundationalism Been Refuted?" Philosophical Studies 29

(1976): 287-305.
"Level Confusions in Epistemology." Midwest Studies in Philosophy 5 (1980): 135
150.
"Christian Experience and Christian Belief." In Plantinga and Wolterstorff(1983).
"A 'Doxastic Practice' Approach to Epistemology." In Clay and Lehrer (1989).
Perceiving God. Ithaca: Cornell University Press, 1991.
Anderson, John. The Architecture of Cognition. Cambridge, MA: Harvard University
Press, 1983.
Annis, David. "A Contextualist Theory of Epistemic Justification." American Philosophical Quarterly 15 (1978): 213-219. Reprinted in Moser (1986).
Aristotle. The Nicomachean Ethics. Trans. David Ross. Oxford: Oxford University
Press, 1980.
Posterior Analytics. Trans. Jonathan Barnes, 2nd ed. Oxford: Clarendon Press, 1994.
Audi, Robert. The Structure ofJustification. Cambridge: Cambridge University Press,
1993.
"Dispositional Beliefs and Dispositions to Believe." NOUS 28 (1994): 419-434.
"Intuitionism, Pluralism, and the Foundations of Ethics." In Walter SinnottArmstrong and Mark Timmons, eds., Moral Knowledge? Oxford: Oxford University Press, 1996.
"Moral Knowledge and Ethical Pluralism." In Greco and Sosa (1999).
Austin, J. L. Sense and Sensibilia. Oxford: Oxford University Press, 1962.
Ayer, A. J. Language, Truth and Logic. New York: Dover, 1952.
The Problem of Knowledge. Harmondsworth: Pelican, 1956.
Beauchamp, T. L. and Rosenberg, A. Hume and the Problem of Causation. Oxford:
Oxford University Press, 1981.
Bechtel, William. "Connectionism and the Philosophy of Mind: An Overview." In
William Lycan, ed., Mind and Cognition. Oxford: Blackwell, 1990.
Bender, John, ed. The Current State of the Coherence Theory. Dordrecht: Kluwer,
1989.
Berkeley, George. A Treatise concerning the Principles of Human Knowledge. LaSalle:

Open Court, 1986.

255

Bogen, James. "Coherentist Theories of Knowledge Don't Apply to Enough Outside of Science and Don't Give the Right Results." In Bender (1989).
Bonjour, Laurence. "The Coherence Theory of Empirical Knowledge." Philosophical
Studies 30 (1976): 281-312. Reprinted in Moser (1986).
"Can Empirical Knowledge Have a Foundation?" American Philosophical Quarterly
15 (1978): 1-13. Reprinted in Moser (1986).
The Structure of Empirical Knowledge. Cambridge, MA: Harvard University Press,
1985.
Carnap, Rudolf. Logical Foundations of Probability. Chicago: University of Chicago
Press, 1950.
Cavell, Stanley. The Claims of Reason. Oxford: Oxford University Press, 1979.
Cherniak, Christopher. "Rationality and the Structure of Human Memory." Synthese 57 (1985): 163-186.
Chisholm, Roderick. Perceiving: A Philosophical Study. Ithaca: Cornell University
Press, 1957.
The Theory of Knowledge. 2nd ed. Englewood Cliffs, NJ: Prentice Hall, 1977.
"Comments and Replies." Philosophia 7 (1978): 597-636.
The Foundations of Knowing. Minneapolis: University of Minnesota Press, 1982.
Clay, Marjorie and Lehrer, Keith, eds. Knowledge and Skepticism. Boulder: Westview,
1989.
Cohen, Stewart. "How To Be a Fallibilist." Philosophical Perspectives 2 (1988): 9 1 123.
Davidson, Donald. "The Method of Truth in Metaphysics." In P. A. French, T. E.
Ueling, Jr., and H. K Wettstein, eds. Midwest Studies in Philosophy 2: Studies in
the Philosophy of Language. Morris: University of Minnesota Press, 1977.
Inquiries into Truth and Interpretation. Oxford: Clarendon Press, 1985.

"A Coherence Theory of Truth and Knowledge." In Ernest LePore, ed., Truth
and Interpretation: Perspectives on the Philosophy of Donald Davidson. Oxford:

Blackwell, 1992.
Davis, Wayne and Bender, John, "Fundamental Troubles with the Coherence Theory." In Bender (1989).
DePaul, Michael. "Reflective Equilibrium and Foundationalism." American Philosophical Quarterly 23 (1986): 59-69.

"Argument and Perception: The Role of Literature in Moral Inquiry." Journal of

Philosophy 85 (1988): 552-565.
Balance and Refinement: Beyond Coherence Methods of Moral Inquiry. London: R o u t -

ledge, 1993.
DeRose, Keith. "Solving the Skeptical Problem." Philosophical Review 104 (1995): 152.
"Contextualism: An Explanation and Defense." In Greco and Sosa (1999).
Descartes, Rene. Philosophical Works. Vol. 1. Trans. Elizabeth S. Haldane and
G. R. T. Ross. Cambridge: Cambridge University Press, 1979.
Dewey, John. The Quest for Certainty. New York: Minton, Balch, 1929.
Reconstruction in Philosophy. Boston: Beacon Press, 1948.

Dretske, Fred. "Epistemic Operators. "Journal of Philosophy 67 (1970): 1007-1023.

Dreyfus, Hubert L. Being-in-the-World. Cambridge, MA: MIT Press, 1991.
Edwards, Paul. "Bertrand Russell's Doubts about Induction." In Flew (1951).

256

Fodor, Jerry and Pylyshyn, Zenon. "Connectionism and Cognitive Architecture: A

Critical Analysis." In Brian Beakley and Peter Ludlow, eds., The Philosophy of
Mind. Cambridge, MA: MIT Press, 1992.
Fogelin, Robert. Pyrrhonian Reflections on Knowledge and Justification. Oxford: Oxford
University Press, 1994.
Foley, Richard. "The Epistemology of Sosa." In Enrique Villanueva, ed., Philosophical Issues 5: Truth and Rationality. Atascadero, CA: Ridgeview, 1994.
Flew, Anthony. Logic and Language, First Series. Oxford: Blackwell, 1951.
David Hume. Oxford: Blackwell, 1986.
Fumerton, Richard. Metaepistemology and Skepticism. Lanham, MD: Rowman and
Littlefield, 1995.
Ginet, Carl. Knowledge, Perception and Memory. Dordrecht: Reidel, 1975.
Goldman, Alvin. "Discrimination and Perceptual Knowledge." Journal of Philosophy
73 (1976): 771-791. Reprinted in Pappas and Swain (1978).
"What Is Justified Belief?" In Pappas (1979). Reprinted in Moser (1986).
Epistemology and Cognition. Cambridge, MA: Harvard University Press, 1986.
"Bonjour's The Structure of Empirical Knowledge." In Bender (1989).
Liaisons: Philosophy Meets the Cognitive and Social Sciences. Cambridge, MA: MIT
Press, 1992.
Greco, John. "Internalism and Epistemically Responsible Belief." Synthese 85 (1990):
245-277.
"Virtues and Vices of Virtue Epistemology." Canadian Journal of Philosophy 23
(1993): 413-432.
"Virtue Epistemology and the Relevant of'Relevant Possibility.' " Southern Journal of Philosophy 32 (1994): 61-77.
"Modern Ontology and the Problems of Epistemology." American Philosophical
Quarterly 32 (1995): 241-251.
"Reid's Critique of Berkeley and Hume: What's the Big Idea?" Philosophy and
Phenomenological Research 55 (1995): 279-296.
"The Force of Hume's Skepticism about Unobserved Matters of Fact." Journal of
Philosophical Research 23 (1998): 289-306.
"Perception as Interpretation." In Michael Baur, ed., Texts and Their Interpretation,
Proceedings of the American Catholic Philosophical Association. New York:
American Catholic Philosophical Association, 1999.
"Two Kinds of Intellectual Virtue." Philosophy and Phenomenological Research.
Forthcoming.
Greco, John and Sosa, Ernest, eds. The Blackwell Guide to Epistemology. Oxford:
Blackwell, 1999.
Hacking, Ian. The Emergence of Probability. Cambridge: Cambridge University Press,
1975.
Heidegger, Martin. Being and Time. New York: Harper and Row, 1962.
Henrich, D. ed., Kant oder Hegel. Stuttgart: Klett-Cotta, 1983.
Hick, John. An Interpretation of Religion: Human Responses to the Transcendent (New
Haven: Yale University Press, 1989.
"The Rationality of Religious Belief." In R. Douglas Geivett and Brenden
Sweetman, eds., Contemporary Perspectives on Religious Epistemology. Oxford: Oxford University Press, 1992.

257

Holland, John, Holyoak, Keith, Nisbett, Richard, et al. "Learning Inferential

Rules." In Kornblith (1994).
Horwich, Paul. Probability and Evidence. Cambridge: Cambridge University Press,
1982.
Howard-Snyder, Daniel, ed. The Evidential Argumentfrom Evil. Bloomington: Indiana
University Press, 1996.
Hume, David. Enquiries concerning Human Understanding and concerning the Principles of

Morals. Ed. L. A. Selby-Bigge, 3rd ed. Oxford: Clarendon Press, 1975.

A Treatise of Human Nature. Ed. L. A. Selby-Bigge, 2nd ed. Oxford: Clarendon
Press, 1978.
Klein, Peter. Certainty: A Refutation of Scepticism. Minneapolis: University of Minnesota Press, 1981.
Kornblith, Hilary. "The Unattainability of Coherence." In Bender (1989).
Inductive Inference and Its Natural Ground. Cambridge, MA: MIT Press, 1993.
Kornblith, Hilary, ed. Naturalizing Epistemology. 2nd ed. Cambridge, MA: MIT
Press, 1994.
Kvanvig, Jonathan. The Intellectual Virtues and the Life of the Mind. Savage, MD:
Rowman and Littlefield, 1992.
Lehrer, Keith. "The Coherence Theory of Knowledge." Philosophical Topics 14
(1986): 5-25.
"Coherence and the Truth Connection: A Reply to My Critics." In Bender
(1989).
Lehrer, Keith and Paxson, Thomas. "Knowledge: Undefeated Justified True Belief."
The Journal of Philosophy 66 (1969): 225-237. Reprinted in Pappas and Swain
(1978).
Locke, John. An Essay Concerning Human Understanding. Ed. Peter Nidditch. Oxford:
Oxford University Press, 1975.
McClelland, James, Rumelhart, David and Hinton, Geoffrey. "The Appeal of Parallel Distributed Processing." In Brian Beakley and Peter Ludlow, eds., The
Philosophy of Mind. Cambridge, MA: MIT Press, 1992. Reprinted from James
Rumelhart, David McClelland and the PDP Research Group, Parallel Distributed Processing: Explorations in the Microstructure of Cognition, vol. 1. Cambridge,

MA: MIT Press, 1986.

McDowell, John. "Values and Secondary Qualities." In Ted Honderich, ed., Morality and Objectivity. London: Routledge and Kegan Paul, 1985.
Mind and World. Cambridge, MA: Harvard University Press, 1996.
McGinn, Marie. Sense and Certainty. Oxford: Blackwell, 1989.
Moore, G. E. "Four Forms of Scepticism." In Philosophical Papers. London: Allen
and Unwin, 1959.
Moser, Paul, ed. Empirical Knowledge. Totowa, NJ: Rowman and Littlefield, 1986.
Nisbett, Richard and Ross, Lee. "Judgmental Heuristics and Knowledge Structures."
In Kornblith (1994).
Nozick, Robert. Philosophical Explanations. Cambridge, MA: Harvard University
Press, 1981.
Pappas, George. "Some Forms of Epistemological Scepticism." In Pappas and Swain
(1978).

258

Pappas, George, ed. Justification and Knowledge. Dordrecht: Reidel, 1979.

Pappas, George and Swain, Marshall, eds. Essays on Knowledge and Justification. Ithaca:
Cornell University Press, 1978.
Plantinga, Alvin. "Reason and Belief in God." In Plantinga and Wolterstorff(1983).
"Epistemic Probability and Evil." Archivio di Filosofia 56 (1988): 557-584. Reprinted in Howard-Snyder (1996).
"Positive Epistemic Status and Proper Function." In James E. Tomberlin, ed.,
Philosophical Perspectives 2: Epistemology, 1988. Atascadero, CA: Ridgeview,
1988.
Warrant: The Current Debate. Oxford: Oxford University Press, 1993.
Warrant and Proper Function. Oxford: Oxford University Press, 1993.
Plantinga, Alvin and Wolterstorff, Nicholas, eds. Faith and Rationality. Notre Dame:
University of Notre Dame Press, 1983.
Plato. Dialogues. Trans. B. Jowett. Vol. 3. Oxford: Clarendon Press, 1953.
Pollock, John. Contemporary Theories of Knowledge. Totowa, NJ: Rowman and Littlefield, 1986.
Potter, Vincent G. On Understanding Understanding. New York: Fordham University
Press, 1994.
Prichard, H. A. Knowledge and Perception. Oxford: Clarendon Press, 1950.
Putnam, Hilary. Reason, Truth and History. Cambridge: Cambridge University Press,
1981.
Rawls, John. A Theory of Justice. Cambridge, MA: Harvard University Press, 1971.
Reicher, G. "Perceptual Recognition as a Function of Meaningfulness of Stimulus
Material. "Journal of Experimental Psychology 81 (1969): 275-280.
Reid, Thomas. Philosophical Works. Ed. H. M. Bracken. 2 vols. Hildesheim: Georg
Olms, 1983.
Rescher, Nicholas. Scepticism: A Critical Reappraisal. Totowa, NJ: Rowman and
Littlefield, 1980.
Rorty, Richard. Philosophy and the Mirror of Nature. Princeton: Princeton University
Press, 1979.
Schmitt, Frederick. Knowledge and Belief. London: Routledge, 1992.
Sellars, Wilfred. "Empiricism and the Philosophy of Mind." In Science, Perception and
Reality. London: Routledge, 1963.
Sextus Empiricus. Selections from the Major Writings on Scepticism, Man, and God. Ed.
Philip P. Hallie, trans. Sanford G. Ethridge. Indianapolis: Hackett, 1985.
Sosa, Ernest. "The Raft and the Pyramid: Coherence versus Foundations in the
Theory of Knowledge." Midwest Studies in Philosophy 5 (1980): 3-25. Reprinted
in Sosa (1991).
"Knowledge in Context, Skepticism in Doubt." In James Tomberlin, ed., Philosophical Perspectives 2: Epistemology. Atascadero, CA: Ridgeview, 1988.
Knowledge in Perspective. Cambridge: Cambridge University Press, 1991.
"Philosophical Scepticism and Epistemic Circularity." Proceedings of the Aristotelean
Society (1994): 263-290.
"Virtue Perspectivism: A Response to Foley and Fumerton." In Enrique Villanueva, ed., Philosophical Issues 5: Truth and Rationality. Atascadero, CA: Ridgeview, 1994.

259

"Postscript to 'Proper Functionalism and Virtue Epistemology.' " In Jonathan

Kvanvig, ed., Warrant in Contemporary Epistemology. Lanham, MD: Rowman
and Littlefield, 1996.
Stove, D. C. Probability and Hume's Deductive Scepticism. Oxford: Clarendon Press,
1973.
Stroud, Barry. Hume. London: Routledge and Kegan Paul, 1977.
The Significance of Philosophical Scepticism. Oxford: Clarendon Press, 1984.
"Understanding Human Knowledge in General." In Clay and Lehrer (1989).
"Scepticism, 'Externalism,' and the Goal of Epistemology." Proceedings of the
Anstotelean Society (1994): 291-307.
Strawson, P. F. Introduction to Logical Theory. London: Methuen, 1952.
Stump, Eleonore. "Aquinas on the Foundations of Knowledge." Canadian Journal of
Philosophy, supplementary volume 17 (1992): 125-158.
Tolhurst, William. "On the Epistemic Value of Moral Experience." Southern Journal
of Philosophy, 29, supplement (1990): 67-87.
Tulving, E., Mandler, G. and Baumal, R. "Interaction of Two Sources of Information in Tachistoscopic Word Recognition." Canadian Journal of Psychology 18
(1964): 62-71.
Van Cleve, James. "Foundationalism, Epistemic Principles, and the Cartesian Circle." Philosophical Review 88 (1979): 55-91.
"Reliability, Justification, and the Problem of Induction." Midwest Studies in
Philosophy 9 (1984): 555-567.
"Epistemic Supervenience and the Circle of Belief." Monist 68 (1985): 90-104.
Wheeler, D. D. "Processes in Word Recognition." Cognitive Psychology 1 (1970):
59-85.
Williams, Michael. Unnatural Doubts. Princeton: Princeton University Press, 1996.
"Skepticism." In Greco and Sosa (1999).
Wittgenstein, Ludwig. On Certainty. Oxford: Blackwell, 1969.
Wolterstorff, Nicholas. "Can Belief in God Be Rational Even If It Has No Foundation?" In Plantinga and Wolterstorff (1983).
"Epistemology of Religion." In Greco and Sosa (1999).
Zagzebski, Linda. "Religious Knowledge and the Virtues of the Mind." In Zagzebski (1993).
"Intellectual Virtue in Religious Epistemology." In Elizabeth Radcliffe, and Carol
White, eds., Faith in Theory and Practice: Essays on Justifying Religious Belief. La
Salle: Open Court, 1993.
Virtues of the Mind. Cambridge: Cambridge University Press, 1996.
"What Is Knowledge?" In Greco and Sosa (1999).
Zagzebski, Linda, ed. Rational Faith: Catholic Responses to Reformed Epistemology.
Notre Dame: University of Notre Dame Press, 1993.

260

Index

agent reliabilism, 1, 4-6, 50-1, 177-8

and contextualism, 252-4
and Gettier problems, 251-2
historical roots of, 6n
and problem of strange and fleeting
processes, 174-80
and relevant possibilities, 207, 21719
and subjective justification, 180-1,
190-2, 202
versions of, 248-9
Agrippa, 108
Alston, W., 14, 120n, 178-9,183n, 22631, 249-50
American pragmatism, 77
analytic epistemology, 21, 77-8, 179,
181
Anderson, J., 20n
Annis, D., 114, 123-4
appearance-reality distinction (see also
sensory appearances), 856
Aquinas, 6n, 44, 179
Aristotle, 6n, 7, 44, 109-10, 113, 179,
200-1
Audi, R., 96n, 120n, 160-1, 189n,
234n, 242n
Austin, J., 39, 77-8, 81
Ayer, A.J., 50, 96n
Baumal, R., 170n
Bayesianism, 96n, 156n

Bechtel, W., 194n

Bender, J., 132-3
Berkeley, G., 77-8, 81-3, 89, 105-6
Hindsight, 171n
Bogen,J., 115n
Bonjour, L., 20n, 113-15, 120n,
128-30, 160n, 168n, 181-4,
200n
Calvin, J., 224
Carnap, R., 156n
case of absurd reasoner, 175
case of epistemically serendipitous lesion, 175
case of helpful demon, 1756
Cavell, S., 39
certainty, 3, 69, 111, 120, 145-6
Cherniac, C , 20n
Chisholm, R., 20-3, 85-7, 113, 11617, 182n, 200n, 224
Cohen, S., 53n, 253n
coherentism, 19, 20, 126-35
connectionism, 194-8
consequentialist approach to evidence,
99-101
contextualism
regarding cognitive abilities or virtues, 213-15
and foundationalism, 124-6
regarding knowledge attribution,
253-4

261

contextualism (cont.)
regarding knowledge conditions, 47,
68-9, 252-4
regarding regress argument, 121-6
Continental philosophy, 77-9

Goldman, A., 6n, 20n, 54n, lOOn, 114

16, 166-7, 170n, 179n, 206n,
209n

Davidson, D., 72-3, 75, 168n

Davis, W., 132-3
debate with skeptic, 22, 39-40, 43, 657, 115-16, 210
deduction, see induction
deontological approach to evidence, 99101
DePaul, M , 232n, 234n, 242n
DeRose, K., 18n, 53n, 186n, 253n
DerridaJ., 168n
Descartes' project, 34-6
Descartes, R.
argument for skepticism about
world, 7, 34-8, 54-60, 253
as a contextualist, 126
Dewey, J., 69n, 79, 81, 141-3, 145n
Dretske, F., 206n
Dreyfus, H., 79n, 91n
Edwards, P., 146n
epistemic circularity, 229-30
epistemic norms, 100, 192-200
epistemic responsibility, 2002
epistemology's project, 22, 43-4, 66-9,
120
externalism, 1, 181-2
Flew, A., 146n
FodorJ., 196n, 197
Fogelin, R., 2n, 16-17, 116-17, 108n
Foley, R., 189n
Foucault, M., 168n
foundationalism, 118-21, 170-1
classical, 222-3
and contextualism, 1246
Fumerton, R., 157n, 160n, 161n
Gadamer, H., 168n
Gettier problems, 251-2
Gilson, E., 77
Ginet, C , 182n, 200n

Habermas, J., 168n

Hacking, L, 150n, 156n
Heidegger, M., 72-3, 77-9, 81, 90-1,
168n
hermeneutical school, 168
Hick,J., 231n
Hinton, G., 194-7
Holland, J., 199n
Holyoak, K., 199n
Horwich, P., 156n
Hume, D.
argument for skepticism about future, 27-8
argument for skepticism about
unobserved matters of facts, 7,
138-41, 145, 154
argument for skepticism about
world, 7, 25-34
and deductivism, 146-8
induction
inductive and deductive support, 5860, 149-55, 159-63
as merely contingently reliable, 172
4
problem of, 137
inference from appearance to reality, 868, 95-101, 169-70, 173-4, 23941
infinite regress of reasons, skeptical argument from (IR), 7, 110-11
Jeffreys, H., 150n
justification, subjective, 5, 180-1, 1902, 198-200, 202, 226n
Kant, I., 72-3, 78-9, 200
KeynesJ., 150n
Klein, P., 2n
knowledge (see also agent reliabilism,
contextualism, Gettier problems)
account of, 218
knowing that one knows, 180-4

262

understanding that one knows, 1847

Kornblith, H., 115n, 172n
Kvanvig, J., 179n
Lehrer, K., 130-3, 208n
Locke, J., 64
logical space of reasons, 4, 33-4
McClelland, J., 194-7
McDowell, J., 33n, 234n
McGinn, M., 122-3
MancUer, G., 170n
Moore, G. E., 17, 20, 43
moral epistemology, 232-5
moral perception, see perception

project of epistemology, see epistemology's project

proprioception, 93
Putnam, H., 32n, 72-3
Pylyshyn, Z., 196n, 197
Pyrrho, 108
question begging against the skeptic {see
also debate with skeptic), 39-40,
67, 115-16
Quine, W. V. O., 43-4

Randall, J., 77
rationalism, 69, 126, 148
Rawls,J., 232
regularity principle, 27, 139, 1526,
159-62, 173
Reicher, G., 170n
Nagel, T., 66
Reid, T., 6n, 19-20, 64-5, 105-6,
naturalized epistemology, 43
146n, 179
Nisbett, R., 199n, 240
relevant possibilities approach to skeptino good inference argument (NGI), 84
cism, 56-8, 204-10, 217-19
Nozick, R., 53n
reliabilism {see also agent reliabilism)
as answer to skepticism, 1611A
evidence, 56, 178
Pappas, G., 62n
generic or simple, 1, 4-6, 68, 165
particularism, Chisholm's {see also strong
particularism), 20-3, 224
7
Paxson, T., 208n
method, 5-6, 178
Peirce, C. S., 145n, 146n, 156n
objections against, 174-80
perception {see also inference from approcess, 5-6
pearance to reality, sensory apsocial practice, 1789
pearances)
religious perception, see perception
elements of, 97-9
representationalism, 4, 90-4, 103-6,
empirical, 235-41
141-5
expert, 246-8
Rescher, N., 62n, 63n, 69n, 145n
as merely contingently reliable, 173- Rorty, R., 77, 80-1, 90-1, 124,
168n
4
moral, 231-4, 241-8
Ross, L., 240
as non-inferential, 95-101, 239-41
Rumelhart, D., 194-7
religious, 225, 226-31
RusseU, B., 17
Plantinga, A., 6n, 14, 171n, 172n,
schema theory, 240-1, 243-4, 247-8
175n, 179n, 226, 231, 250-1
Plato, 44, 67, 109, 113, 177
Schmitt, F., 37n
Pollock, J., lOln, 128-9, 192-3, 193n
Sellars, W., 33n
post-modernism, 77, 79-80, 90-1
sensory appearances {see also perception,
Potter, V., 62n, 63n
representationalism)
Prichard, H. A., 182n
as evidence, 95101

263

sensory appearances (cont.)

theories of, 856
thin versus thick conceptions of, 304, 97-9, 135, 174, 236-7
sensus divinitatus, 225

Sextus Empiricus, 108-9

skepticism
as existential problem, 22, 66,
107
as theoretical problem, 22, 107
Sosa, E., 6n, 122n, 124n, 134-5, 167n,
168n, 176-7, 187-90, 192, 206n,
216n, 249, 252
Steup, M , 202n
Stove, D., 146-8
Strawson, P., 66, 146n, 153n
strong particularism, 21-4, 108-18,
198, 220-2, 226, 230-1, 235
Stroud, B., 2n, 37n, 39, 41-4, 53-4,
66, 146, 184-7

Stump, E., HOn

Swain, M., 206n
theory of ideas {see also representationalism), 4, 141-5
Tolhurst, W., 234n, 242n
Tulving, W., 170n
Van Cleve, J., 169n, 172, 184n
virtue epistemology, 1, 4-6, 177, 191
2, 248, 252
virtues, intellectual, 5-6, 96-7, 176-7,
191-2, 216-17
Wheeler, D., 170n
Williams, M., 2n, 36, 44-51, 96n, 125,
168n
Wittgenstein, L., 39, 121-3
Wolterstorff, N., 231n
Zagzebski, L., 179n, 251-2

264