Philosophy and

Phenomenological Research
Philosophy and Phenomenological Research
Ó 2012 Philosophy and Phenomenological Research, LLC

Reverse Engineering Epistemic

Evaluations*
sinan dogramaci
The University of Texas at Austin

1. A Puzzle: What Is the Function of Epistemically

Evaluative Language?
What is the function of words like ‘irrational’ as used in ordinary
epistemic evaluations? I’m thinking of simple evaluations: criticism like
‘Smith’s belief that Obama’s a Muslim is irrational’ or praise
like ‘Green’s belief that all humans are mortal is rational’. We don’t
make such claims just for the sake of it, just for fun, or for no reason
at all. So what is the real point or purpose of this epistemically evalua-
tive aspect of our linguistic practice? It must have some utility, but
what? What would we lose if epistemically evaluative words suddenly
disappeared from our vocabulary?1
The question requires some motivation, since discovering a word’s
function doesn’t necessarily call for armchair philosophy. Some words
might serve a hidden psycho-sociological function that only empirical
science can uncover. This might be true of words like ‘hello’ and
‘thanks’. And for many words, their basic function isn’t particularly

* This paper won the Young Epistemologist Prize for the Rutgers Epistemology
Conference in 2011.

I pronounce it, when speaking English, like this: sinÆon dorÆuhÆmudgeÆuh.
1
The question of this paper thus contrasts with the timeworn questions of how to
give necessary and sufficient conditions for this or that philosophically interesting
property. There are, however, a few excellent philosophical explorations of the func-
tion of this or that philosophically interesting word or concept.
In philosophy of logic and language, Quine offered an elegantly simple insight
about the utility of the truth predicate which precipitated the contemporary devel-
opment of deflationism about truth; see Quine (1970), Leeds (1978), Horwich (1990/
98), and Field (1994).
In metaphysics, a series of recent papers on modality have broached the fasci-
nating issue of the function of modal concepts; see Kment (2006), chapter 5 of
Williamson (2007), and Divers (2010).

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 1

mysterious. For example, we use ‘water’ to be able to communicate
important, sometimes life-saving, information about where there is (or
isn’t) water. It’s obvious that if ‘water’ were expunged from our vocab-
ulary (by an Orwellian government, suppose), life would be harder
until we could recover or replace it.
There is, however, a distinctly philosophical puzzle about the func-
tion of our epistemically evaluative language. The puzzle I have in
mind begins with the following intuitive line of thought. Whatever the
utility of our epistemically evaluative practice is, it can’t just be that it
promotes rational beliefs and suppresses irrational beliefs: that outcome
just isn’t valuable for its own sake. The utility of the practice must be
instrumental utility. Instrumental for what, though? Intuitively, truth.
As inquirers, what we are ultimately interested in are not rational
beliefs, but true beliefs (about topics of interest to us).2 I thus take the
following as an intuitive premise: the utility of our epistemically evalua-
tive practice is just an instrumental utility it has by helping us get true
beliefs.3
By and large, that premise jibes with our ordinary epistemically eval-
uative practice. We apply ‘irrational’, by and large, to someone’s beliefs
when he is veering from truth, and we apply ‘rational’ when the truth
is being well-tracked. (Or ‘[un]justified’, ‘[un]reasonable’, even ‘[il]logi-
cal’: our topic is evaluations made using ordinary language, so we can
treat these all as pretty well equivalent. I’ll stick with ‘[ir]rational’ in
this paper just for uniformity, and because it strikes my ear as the most
commonly used in ordinary conversation.) A puzzle arises because

In epistemology, there is Edward Craig’s wonderful and underappreciated book,

Knowledge and the State of Nature, Craig (1990). Some recent work by Steven
Reynolds, brought to my attention only after this paper was given at the Rutgers
Epistemology Conference, is closely related to Craig’s project; see Reynolds (2002).
Though Craig and Reynolds write about knowledge rather than rationality or
justification, there are obvious common themes and specific points of contact between
their views and the view I will be developing. The most notable common theme is the
general importance we assign to the role of testimony in the function of epistemic
language. One major difference, since they write about knowledge and not rationality,
is that their views are not designed to explain the puzzle I will be raising shortly.
2
By ‘ultimately interested in’, I don’t mean to claim that true belief is valuable for its
own sake. Maybe it is, but I don’t want to commit myself to that. My claim is that
true beliefs have some kind of utility, and we can appeal to that to explain the util-
ity of our epistemically evaluative practice.
3
Note that I’m not saying that epistemic rationality is instrumental rationality. Kelly
(2003) argues, convincingly in my opinion, that an epistemically rational belief is
not just a belief that is instrumentally rational for the believer. Kelly shows that
there are cases where a belief is rational even though the believer does not have the
goal of believing the truth in this case. But Kelly’s view does not conflict at all with
my claim that we have an interest in believing the truth about topics that are of
interest to us, and that the utility of our epistemically evaluative practice is due to
that interest of ours.

2 SINAN DOGRAMACI
there appears to be a range of cases that stand as stark and inexplica-
ble exceptions. These are cases where we criticize as irrational the
beliefs of someone we know is tracking the truth very well, and we
praise as rational the beliefs of someone we know is tracking the truth
very badly!
The clearest such cases are the old counterexamples to the simplest
and most intuitive versions of reliabilism. A simple reliabilist theory of
justification, such as Goldman (1979), is just the sort of theory we’d
have expected to be right if our evaluative practice is a means for
getting at the truth. But then BonJour (1980) gave us the unwittingly
reliable clairvoyant, Norman, whose beliefs we criticize, calling them
‘irrational’. The thing that’s puzzling me is, if we know Norman tracks
the truth so well, why on earth do we criticize his beliefs? What pur-
pose could it serve? That’s a puzzle case for anyone wondering what
function ‘rational’ serves.4
Some people hesitate to call Norman’s beliefs irrational.5 Perhaps
this compromises Norman’s status as a counterexample to reliabilism,
but it doesn’t remove the puzzle I’m concerned with. Why do so many
people feel any temptation at all to criticize the beliefs of someone
whom we know, by stipulation, enjoys an excellent, robustly reliable
connection to the truth? That’s a puzzle, as I said, for anyone wonder-
ing what function ‘rational’ serves.
The other classic counterexample to reliabilism also generates the
puzzle. Cohen (1984) pointed out that Dupe, the victim of a Cartesian
evil demon, uses an unreliable process to form perceptual beliefs, but
we apply ‘rational’ to Dupe’s beliefs. Why do we give epistemic praise
to someone we know to be so badly disconnected from the truth?
Cohen himself was gripped by a puzzle different from ours but closely
related. He wondered about, as he labeled it, ‘‘the truth-connection’’:
surely, Cohen said, rational beliefs are somehow connected with truth,
but if reliabilism is wrong, then what could that truth-connection be?
My puzzle is not in the first instance about the property of being rational,
but is rather about our use of the word ‘rational’ and similar terms of
epistemic evaluation. I’m puzzled by why we use these words in ways
that appear to be downright counterproductive(!) to our ultimate episte-
mic goal, namely believing just the truth (about topics of interest to us).

4
Goldman, BonJour and other contributors to this literature mostly used the term
‘justification’. Of course, they were explicitly using it to pick out a notion present in
ordinary thought and talk. As noted, I treat ‘justified’ and ‘rational’ as meaning
pretty much the same thing as far as ordinary conversation goes.
5
See, for example, Weinberg et al. (2001). As the authors emphasize, variation in
responses varies with cultural background. This important point has supportive
implications for the larger project advertised in the final section of this paper.

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 3

 There’s no dismissing the puzzle by saying Norman and Dupe are
isolated and unrealistic cases. BonJour and Cohen extracted these vivid
examples from real standards that tacitly guide our everyday usage.
Our criticism of Norman is a distillation of a feature of our standards
that also generates a complaint you can hear in everyday discourse:
‘What you believe happens to be true, but you didn’t know it!’ And for
a realistic version of Dupe, think of an unwitting schizophrenic, like
John Nash was portrayed in the film A Beautiful Mind. We aren’t
tempted to call his beliefs ‘irrational’.
Puzzle cases even arise for deductively inferred beliefs. Let Rami be an
unwitting logical or mathematical clairvoyant: in just the way we some-
times infer q from p and if-p-then-q, Rami infers Fermat’s Last Theorem
immediately from some elementary axiom. (The example is drawn from
Boghossian (2003).) Again, even if some people feel ambivalence, a puzzle
arises just from the fact that many of us share Boghossian’s strong temp-
tation to call Rami’s inferred conclusion irrational. Why do many of us
feel any inclination at all to criticize Rami’s conclusion when we ourselves
know his transition is as valid as our own basic deductive transitions?6
For Dupe’s deductive analog, consider Frege. Frege drew intuitive infer-
ences concerning sets, oblivious to the proof (provided by Russell’s para-
dox) that he was following an invalid rule (his notorious Basic Law V).
Many will share the temptation to say Frege’s unwittingly invalid conclu-
sions were, like Dupe’s conclusions, rational (at least at the time). Why,
then, do we call Frege’s conclusions rational and Rami’s irrational?
These puzzle cases can lead you to worry that our epistemically eval-
uative practice is full of glitches! Are we so badly designed that every
so often we make evaluations that are counterproductive to our interest
in truth? Or might there be some other non-obvious interest at play
that explains our behavior in these cases? But even if it were to turn
out that, in these cases, our evaluative practice is serving some other
valuable function, that would still be disappointing. Either that other
function is unavoidably in tension with our interest in truth, or the ten-
sion is avoidable and we’ve been serving our interests very inefficiently.
An ideal solution would provide a clear and plausible function
that all our epistemically evaluative practice serves in a uniform way,
and which vindicates that intuitive premise that the utility of our

6
I hope it is clear why we cannot say our inferences are better than Rami’s because
we make ours with the recognition of their validity. Three familiar problems: (1)
Most people don’t give any thought to the validity of their inferences. (2) Between
Carroll (1895) and Boghossian (2003), it should be very clear that recognition of
validity doesn’t help to explain why our basic deductively valid inferential transi-
tions are rational. And (3) Rami is in as good a position as we are to argue, by giv-
ing a standard rule-circular soundness proof, that his basic transitions are valid.

4 SINAN DOGRAMACI
epistemically evaluative practice is just an instrumental utility it has by
helping us get true beliefs. I think there is such a solution to the puzzle,
and I will use the rest of this paper to sketch it.
The task is a one of reverse engineering. First, bracketing specula-
tion about what it is for, let’s look at what our epistemically evaluative
practice actually is, and what some of its significant effects on us are.
Once we have a picture in hand, we can try to come up with an expla-
nation of how this practice benefits us: what exactly is it for, and how
does it accomplish that? I’ll propose an answer that vindicates the intu-
ition that it is all ultimately for the sake of having true beliefs.

2. Our Use of Epistemically Evaluative Language

2.1. The Simple Use: Evaluations of Epistemic Rules

We use ‘[ir]rational’ in interpersonal, evaluative discourse. With the
simplest such assertions, we praise or, far more often, we criticize a
belief by asserting that the belief is rational or that it is irrational,
respectively. I am taking it that assertions such as,

Smith’s belief that Obama’s a Muslim is irrational.

Jones’s belief that 9/11 was an inside job is irrational.
Brown’s disbelief in global warming is irrational.
Green’s belief that all humans are mortal is rational.

have the simplest logical form of any of the assertions we make using
‘rational’ or ‘irrational’. In attempting to reverse engineer a function
for epistemically evaluative language, the aspect of our overall use that
I will focus on is the use exemplified by the indented assertions above.
In other words, my focus is on our practice of calling beliefs rational
and, the more common case, calling beliefs irrational. I label this the
simple use of epistemically evaluative terms like ‘[ir]rational’.
(The simple use is not the only way we use ‘rational’. We embed
‘rational’ into non-indicative and/or logically complex structures. Also,
not only do we use the word in speech, we use the associated concepts
in thought, for instance if I were to believe that Smith’s belief is irratio-
nal, perhaps never making an assertion that expresses that belief. For
now, I am setting these uses aside, returning to them only at the very
end of the paper.)
In any instance of the simple use, like the indented claims above, the
surface form of the assertion indicates that the object of the evaluation
is a belief or a disbelief. However, two observations about ordinary dis-
course reveal that what is being implicitly evaluated are belief-forming

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 5

rules. (‘Implicitly’ here just means it isn’t reflected in the surface form
of the evaluation.)
The initial observation reveals that the evaluation is sensitive to
more than just the attitude itself. The observation is that we ordinarily
allow for the belief that p to be rational in one person and irrational in
another, or even in the same person at different times, when the basis
of the belief differs or changes over time. We can allow that while
Brown’s disbelief in global warming is irrational, the same attitude held
by his very young and very sheltered child is rational. So, in our ordin-
ary practice, we consider more than just the belief when deciding
whether or not it is rational: what we implicitly evaluate includes the
belief together with its basis.
The next observation is that the object of our evaluations is not just
something particular but something general. To see this, notice that
we’d not count Smith as having responded to our criticism if the only
thing she does is revise her belief that Obama’s a Muslim. She could
form future beliefs that have different contents, but which invite what
we’d intuitively count as the same criticism. Smith has not responded to
our original criticism if she merely replaces her old belief with a new
belief that Obama is the antichrist. Or imagine a frustrating gambler
who needs to be repeatedly criticized each time he believes of this or
that string of bad luck that it is increasingly due to turn around. Even
if he revises his present belief on each occasion he is criticized, he is
not fully responding to the criticism; the same criticism is being
repeated. Such cases suggest that what we implicitly criticize is the way
beliefs are formed from their bases.
The resulting picture is a familiar one in contemporary epistemology,
though labels vary: the fundamental objects of epistemic evaluations
are belief-forming methods, processes, norms, or rules.7 I’ll use ‘rules’,
but the other labels are fine too.
Note that we don’t need explicit knowledge of the rules to engage in
the simple use. Like with the rules of syntax, implicit knowledge can
guide our belief-forming practice and our evaluations of particular
belief-basis pairs. Epistemologists can try to work out the implicit gen-
eral rules by following the standard procedure of pursuing reflective
equilibrium. The data are our evaluations of particular beliefs, and per-
haps even a few intuitions we may have about the rules or their general
structure. While there isn’t yet any consensus about exactly what any of
our rules are, we have plausible ideas about the rough form of many of
our various observational, inductive and statistical, and deductive rules.

7
See, for example, Pollock and Cruz (1999), Goldman (1986, 2009), Wedgwood
(2002), Field (2000), Peacocke (2004), Boghossian (2008).

6 SINAN DOGRAMACI
 Note also that some authors distinguish our basic (observational,
inductive, etc.) rules from non-basic rules, such as believing what the
Times reports. Whenever I refer to rules, I mean basic rules (unless I
explicitly describe them as non-basic rules).
So then, in a critical instance of the simple use, like the criticism of
Smith’s belief about Obama, the evaluator is (implicitly) criticizing
Smith’s failure to follow the correct rules. Maybe Smith accepts the
correct rules, but she failed on this occasion to follow them, i.e., she
committed a performance error, like what happens with syntax. More
likely in this case, Smith followed bad rules that take fear and prejudice
as bases for beliefs. (Instances of the simple use that involve epistemic
praise are slightly more complicated, since we call a belief ‘rational’
only if the belief was formed by correct rules and the believer had no
irrational beliefs in her basis. To avoid wordy formulations, and in any
case to stick with the more common instances of the simple use, I’ll
continue to focus on critical instances.)

2.2. The Simple Use Promotes Coordination

I’ve highlighted a simple and central part of our use of ‘rational’, but
before we can tell what function this simple use has, we need to say
something about what effects that use has on us beyond its effects on
speech. Making assertions has little point for its own sake; a word’s
real function must depend on how it influences non-linguistic life.
To see what this influence could plausibly be, I look to metaethicists
for some clues. Here is R. M. Hare summing up a line of thought
whose force he says was recognized by Socrates, Aristotle, Hume,
Kant and Mill. (Of course, it also motivated the specific theory Hare
developed.)

The prescriptive meaning is the function all normative and evaluative

statements have of guiding our actions. This shows up in the fact that
someone who makes one about his own proposed actions but does
not act accordingly exposes himself to a charge of speaking insin-
cerely, as also does someone who makes one about other people’s
actions but does not will them [...] so to act.8

Hare makes two claims here, both of which I claim are plausible when
they are transplanted into the epistemic domain. Let’s examine them.
The first claim is this: there’s a robust correlation between the rules a
speaker (implicitly) evaluates positively/negatively and the rules that she

8
See Hare (1998), section 2. My ellipsis deletes ‘(in the above broad sense)’. Hare
describes that as ‘a broad sense in which it covers Kant’s rational will and
Aristotle’s boulesis or rational desire.’

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 7

herself does/doesn’t follow. This first claim is a modest form of judg-
ment internalism for epistemic rationality, a plausible thesis in light of
the intuitive general tie between evaluation and practice. Note that I
am not insisting on any kind of necessary relationship here (though I’ll
record my sympathy for the view that there’s a conceptual necessity
here). There are many cases where the correlation can fail. For one
thing, the assertion may not be sincere; it may be a lie. Also, since our
knowledge of our own rules is implicit, a speaker can form a mistaken
explicit belief about her rules.
The second claim in the quote from Hare is something like this:
when we make our evaluations of others’ beliefs, we are intending for
them to follow the correct rules. Actually, I’ll claim something slightly
stronger; I claim this: our evaluations have an overall tendency to influ-
ence our audience to follow the endorsed rules.
Why think that second claim is right? Properly understood, it is not
any kind of particularly bold claim. I am not committed to any claim
about the essence of evaluative concepts. I am not committed to any
claim about any metaphysically necessary connections. I am making a
contingent claim. Thus, thought experiments about aberrant creatures
who are invulnerable to the influence of our evaluations are irrelevant
here. (Indeed, the function I’ll be proposing wouldn’t apply to such
aberrant creatures, and I myself have no idea what function epistemi-
cally evaluative notions like ours could have for them.) I am also decid-
edly not claiming that interpersonal evaluations are the sole cause or
explanation of why each of us follows the particular belief-forming
rules we do.
What I am claiming is just a commonsensical truth about our being
built to respond to each others’ evaluations in accommodating ways.
Each of us knows from personal experience how we are built to
respond to the sting of criticism, or the pleasantness of praise. As
Adam Smith put it:

Nature, when she formed man for society, endowed him with an origi-
nal desire to please, and an original aversion to offend his brethren.
She taught him to feel pleasure in their favourable, and pain in their
unfavourable regard. She rendered their approbation most flattering
and most agreeable to him for its own sake; and their disapprobation
most mortifying and most offensive.9

The claim is consistent with the observation that, often, the person crit-
icized (or praised) by an assertion is not the audience, not the person
the assertion is made to. I may tell you that Smith’s belief is irrational,

9
See Smith (1759/1976), p.116.

8 SINAN DOGRAMACI
even though I believe or suspect that you follow all and only the cor-
rect rules. In such cases, my evaluation still influences my audience: it
serves to reinforce a good practice, as well as to encourage the prosely-
tization of third parties. Or I may tell you Smith’s belief is irrational,
even though I believe or suspect that you are equally guilty of Smith’s
mistakes. In such cases, my evaluation will be especially effective in
influencing my audience because I am being diplomatic! Indeed, it’s
plausible that such diplomatic third-personal evaluations are the most
common instances of the simple use.
The scope of the claim (that our evaluations tend to influence our
audience’s rule-following) is all our rules, down to our most basic.
There is no restriction, even if cognitive science finds that some rules,
e.g., very basic deductive or inductive rules, are innate.10 Two points
show why. First, what cognitive science could find, at best, is that, like
with syntax, we innately accept certain rules; our competence is innate.
But, our success in following those rules, our avoidance of performance
errors, can still be improved by the social influence of the evaluative
use. Consider, syntax is innate, but some professional athletes’ inter-
view transcripts read like a sputtering of barely interpretable fragments,
while some public speakers are trained to always speak in well-formed
sentences. Second, it’s likely that even our most deeply accepted rules
are subject to modification. After all, many promising solutions to the
semantic paradoxes call on us to revise one or another of our most
dearly held deductive rules of reasoning.11 And with induction, while it
may be hard to imagine actually adopting a ‘‘gruesome’’ rule, it’s easy
and common enough for us to revise the setting of the parameter for
the trade-off between speed and caution in our inductive rules. I criticize
Jones’s belief in a 9/11 conspiracy theory because she is being unrea-
sonably quick to draw conclusions from a small amount of data. And,
I criticize Brown’s global warming skepticism because he is being
unreasonably cautious when the data is overwhelming. The simple use
can bring people like Jones and Brown to settle the trade-off between
power and caution in a more reasonable, moderate way.12
An important point requires emphasis here: it’s not my view that we
criticize Smith, Jones and Brown as a way of reasoning with them (or
with our audience, if the audience is someone else). We cannot hope to
modify their basic rule-following behavior by providing them with new

10
See Carey (2009).
11
See Field (2008).
12
The inescapable trade-off between speed and power was something William James
was apparently aware of and moved by; see James (1897/1956). See Goldman
(1986) for a more contemporary discussion.

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 9

data or evidence (such as the fact that we believe their rules are
unreliable). Reasoning with someone and providing her with new
evidence can only effect changes in non-basic rules. Maybe you could
get someone to stop trusting the Enquirer by asserting: that newspaper
isn’t reliable. But bringing about changes in someone’s basic epistemic
rule-following (e.g., changing the caution/speed parameters in Jones
and Smith’s basic inductive rules) requires a more brutish kind of force
than reasoning and evidence are capable of. It requires the kind of
force Adam Smith describes flattery and disapprobation to have in the
above quote. And this is why evaluative terms like ‘[ir]rational’ and
‘[un]justified’, normatively loaded as they are, can play a role that non-
normative terms like ‘[un]reliable’ cannot. This is why the simple use
can promote a deep kind of coordination, coordination among basic
epistemic rules.
Okay, I’ve presented and hopefully made plausible my two claims
about the simple use. This leads to important conclusions about how
our epistemically evaluative assertions have an substantive effect on
our lives. Instances of the simple use serve to pressure others to con-
form to the belief-forming rules of the evaluator. If that’s right, we get,
as an immediate consequence of everyone’s participation in the simple
use, a significant result. The result is that the simple use promotes the
coordination of epistemic rule-following across the linguistic commu-
nity. I pressure you to follow my rules, you pressure me to follow your
rules: together we push toward an equilibrium in which we follow
shared rules.
To be clear, I don’t claim the simple use is the sole cause or explana-
tion of why there is coordination among our belief-forming rules. What
I claim is that the simple use promotes coordination. The simple use is
the social influence on our belief-forming patterns; I allow other non-
social factors, e.g., some kind of evolutionary biological factor, to
share responsibility for why we follow the coordinated rules we do. As
mentioned, empirical science may find there are certain rules of reason-
ing we accept innately. The simple use cannot, of course, be responsible
for any degree of coordination those innate rules start out with, but
the simple use can help us stay on the shared track. As a matter of
actual fact, we often do veer off track, and we do benefit from interper-
sonal help to get back on track. Another empirical finding seems to be
that we each have an incurably asymmetric tendency to accurately
detect the cognitive flaws and biases in others’ reasoning while remain-
ing blind to those in our own reasoning.13 If so, then by allowing others

13
See Pronin et al. (2002).

10 SINAN DOGRAMACI
to help police our reasoning, we receive a more objective evaluation of
it.
Now, this view that evaluative language serves to promote coordi-
nation across the community is not entirely new. Gibbard (1990)
proposed that the function of evaluative discourse is to foster coor-
dination. Gibbard was, however, primarily focused on practical ratio-
nality; he spoke of coordination of actions and feelings, not belief-
forming rules.14 So, the question we still have to answer is: what
could this proposed coordination of belief-forming rules be for?
What we need now is an account of why we use evaluative language
to coordinate belief-forming rules. I offer my answer in the next
section.

3. Epistemic Communism
3.1. Coordination Makes Testimony Trustworthy
We each want true beliefs, and no false beliefs, about the topics that are
important to us. This requires collecting lots of evidence, evidence that
will serve as the bases for the beliefs we want to acquire. We each have
our individual faculties of evidence collection. Each individual’s
perceptual faculties give her experiences that can serve as the bases of

14
See especially chapter 4. For instance, here Gibbard says: ‘Shared evaluation is
central to human life, I suggest, because it serves biological functions of rehearsal
and coordination.’ (p.72) Also: ‘Here, then, in brief, is the proposal. Normative dis-
cussion might coordinate acts and feeIings if two things hold. First, normative dis-
cussion tends toward consensus. The mechanisms here, I shall propose, are two:
mutual influence, and a responsiveness to demands for consistency. Second, the
consensus must move people to do or feel accordingly.’ (p.73)
Also see chapter 12, specifically pp.223–6, for good summaries of the relevant
theme of the book. Also see the rest of pp.72–3, where Gibbard separates out two
aspects of normative discourse paralleling the two I used the Hare quote to intro-
duce. And see pp.77-8, where Gibbard emphasizes a point I agreed with earlier,
namely that the connections between evaluation and practice, though robust, may
not be necessary.
A very significant difference I have with Gibbard is that he ends up invoking
the coordinative effect to argue for the anti-realist conclusion that there are no
moral facts. See, especially, chapter 6, for the full presentation of his argument.
Much later, in Gibbard (2003), he backed away from his anti-realist views, saying
that it is difficult to see how to coherently draw the realist/anti-realist distinction.
(See, e.g., p.x of the preface.) I’m sympathetic to the view that there is no good
way to draw a coherent distinction (such that neither side becomes a non-starter).
Field (2009) outlines a view of the evaluative role of the linguistic and psycho-
logical use of ‘rational’ very much like the view in Gibbard (1990), and Field does
focus on the epistemic domain. Field, though, does not discuss coordination, and
goes into few details about how use of the word/concept results in its performing
any function for us. The most he says about function is on p.286, where he says
the function of normative discourse ‘is to give advice, to oneself and others’. I
certainly agree with that much.

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 11

non-inferential beliefs. And, each individual’s own existing beliefs can
serve as bases for new inferred beliefs. But, a single individual’s
perceptual faculties, and her individual stock of beliefs, are limited
sources of evidence for her. Expanding the pool of accessible evidence
would allow us to acquire more of the true beliefs we want.
That is what the simple use and the ensuing coordination of our
belief-forming rules gives us: our practice with ‘rational’ allows us to
put together every individual’s evidence into a vastly larger communal
pool. This is because, by actively pressuring everyone to share belief-
forming rules, we make testimony trustworthy in our epistemic commu-
nity. Because we are each actively and constantly using ‘rational’ in a
way that promotes that coordination, we can easily recognize that we
are all following the same rules. And if we can recognize such coordi-
nation, then we have reason to trust each others’ testimony.
I am, of course, taking it for granted here that, somehow or other,
we each rationally believe, of our own actual rules, that they are
reliable. If we couldn’t, we would be forced into radical skepticism.
However it is that we are each able to rationally believe of these rules
that they are reliable, we are able to trust each other because we’ve
shared these rules with each other.
So, at last, my proposal is that the function of ‘rational’ is to extend
our common epistemic reach by enabling each person to serve as an
‘‘epistemic surrogate’’ for any other person.15 I can’t have all the per-
ceptual experiences you have, but your experiences can serve as bases
for beliefs that I can acquire by testimony. Likewise, I can’t have all
the beliefs that are in your head, but your own beliefs can serve as
bases for inferred conclusions that I can then acquire by testimony.
And this is all possible because, when we share rules, I can trust that
you will draw the same conclusion from an evidential basis that I
would.
If the resulting view needs a name, call it epistemic communism.
Actually, according to the view itself, we should call everyone an episte-
mic communist. For, in our epistemic community, the pursuit of truth
is a group effort: people depend on the communal support system to
help them follow the same rules, the rules we all agree are correct. And
this answers a question that opened the paper. If ‘rational’ (and ‘justi-
fied’ and all the others) were suddenly expunged from our vocabulary,
and nothing new were introduced to take their place, our communal
support system would be weakened, and the credibility of testimony
would suffer.

15
Thanks to Karl Schafer for the apt label ‘epistemic surrogate’.

12 SINAN DOGRAMACI
 Notice how this form of communism, unlike the political version,
gives rise to a highly efficient (epistemic) economy. There is a division of
epistemic labor, the labor of gathering evidence and forming beliefs
based on it. As we labor, we each do not need to store all our evidence.16
This is because, if you present me with testimony but not your evidence,
I can still trust that, whatever your evidence was, you formed the same
belief that I would on the basis of it.17 Furthermore, not only can I make
use of your unstored evidence, I can make use of your computational
resources for reasoning. I can’t come up with and consider every explan-
atory hypothesis. I can’t go through every chain of deductive reasoning.
But, if you come up with a powerful explanatory hypothesis (for some
known data), or infer some deductive result (from some known pre-
mises), I will trust the belief you form, since it is the same belief I would
myself have formed if I’d been able to devote enough computational
resources to the task. Our system even allows us to exploit the differences
among people’s epistemic strengths. People vary with how fast they can
reason, and with how creative they are at coming up with explanatory
hypotheses. We can thus efficiently divide the intellectual labor according
to principles of comparative advantage. (Or, to sound more communist:
from each according to her epistemic ability!)

3.2. Why Epistemic Communists Aren’t Reliabilists

Sharing rules is not a necessary condition for testimony to be ratio-
nally accepted. The only necessary condition (and the weakest suffi-
cient condition) for testimony to be rationally accepted is that the
audience rationally believes the testifier is likely to be reporting the
truth. So, if we know that Norman is an unwittingly reliable

16
Harman (1986), chapter 4, made the point that we do not and should not store all
our evidence.
17
To be sure, we should not, and of course we in fact do not, discard all our
evidence. Storing evidence may be costly, but it will always have some epistemic
value. This is because, interestingly, there can be cases where two items of evidence,
E1 and E2, individually confirm beliefs B1 and B2, respectively, even though their
conjunction, E1 & E2, disconfirms the conjunction B1 & B2. In these cases, our
cost-saving engineering trick can fail. Example: the early polls confirmed that a
woman will be nominated. The later polls confirmed that a black person will be
nominated. But, the whole batch of polling data did not confirm that a black
woman will be nominated. In such situations, it benefits us to keep track of more
information than just that E1 confirmed B1. In the example, we want to keep track
of the fact that the early evidence confirmed that Hillary Clinton will be nomi-
nated, while the later evidence confirmed that Barack Obama will be nominated.
These kinds of cases illustrate the value of storing more data.
So, sharing rules only allows us to grow the pool of communal knowledge while
storing vastly less information than we would otherwise need to. Again, sharing
rules is what enables me to share your beliefs without sharing all your evidence.

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 13

clairvoyant regarding some topic, then we can rationally accept what
he reports about that topic. Why, then, do we apply ‘irrational’ to
Norman? If the function of ‘rational’ is to shape those around us into
valuable epistemic surrogates, wouldn’t it make more sense to praise
him with ‘rational’?
To see why we criticize Norman, remember that our task is to reverse
engineer the function of ‘rational’. Thus, we should ask: what’s the easy
way, engineering-wise, to set up a system of trust so that evidence is
pooled? There may be many possible ways of getting us to trust each
others’ testimony, but the easy, the low-cost, way of doing so is to make
it plain to each person that everyone else is using the same methods as
her. Think of the work that has to be done before you can rationally
accept Norman’s testimony. Someone, you or someone you already
trust, must first independently learn a substantial amount about the very
topic that Norman’s reports are about, so that Norman’s reports can be
checked and his reliability rationally established. While such a vetting
procedure may be worth going through under some circumstances, a
community-wide system of testimony that generally worked this way
would be extremely inefficient. (Of course, to the extent that we some-
times actually do have to expend effort to detect liars among us, our
system becomes bogged down. By and large, though, we actually trust
each other without first engaging in costly vetting procedures.) So, we
criticize Norman because the more efficient system is one where we use
‘rational’ just to promote coordination among everyone’s belief-forming
rules.
Of course, Norman’s ‘‘visions’’ are a valuable fund of evidence, evi-
dence that a well-engineered epistemic system wouldn’t let go to waste.
But, it is perfectly possible to epistemically criticize Norman while not
letting his visions go to waste. In our actual system, we don’t let his
visions go to waste because, although we criticize his rule-following, at
the same time we inform him that his visions are reliable, with the
overall result that he can now apply our rules to make proper, praise-
worthy use of his evidence. Remember, epistemic evaluation is evalua-
tion of rule-following. This also addresses the corresponding question
about Dupe, the unwittingly unreliable victim of a Cartesian demon: if
the function of ‘rational’ is to shape others into valuable epistemic
surrogates, wouldn’t it make more sense to criticize Dupe with
‘irrational’? No, because there is nothing to criticize regarding his rule-
following. Dupe’s flaw is that he has been fed misleading evidence, but
‘rational’ is not used to evaluate evidence. With Dupe, the best way to
make him into a valuable epistemic surrogate is to make him aware
that his evidence is misleading. For a slightly more realistic example,
think again of how Nash was portrayed as an unwitting schizophrenic

14 SINAN DOGRAMACI
in the movie A Beautiful Mind. He applies correct perceptual and
deductive belief-forming rules, indeed applies them ingeniously, to
highly misleading inputs about spy codes he sees around him, but
which do not really exist. Thus, we praise Nash’s rule-following, but
must correct his evidence.

Coda: A Keyhole View to a Larger Project

I would like to very briefly advertise, without argument, a perspective
of the bigger picture. This paper has concerned the pragmatics of a bit
of language, the function of epistemic evaluations. Such views don’t
usually license any conclusions about the semantics of that language or
the nature of the properties that language is about. But, there are a few
famous philosophical arguments that revolve around just such a
transition (perhaps most famously a much-debated argument for defla-
tionism about truth).18 I see the communist view as a springboard for a
deflationary view of the property of epistemic rationality. To be only
slightly more specific: I think epistemic communism supports a form of
epistemic conventionalism.
The development of and argument for this view must be presented
elsewhere, but the basic maneuver is intuitive. While our actual episte-
mic community has converged upon one set of moderately reliable and
powerful epistemic rules, there are numerous alternative sets of reliable
and powerful rules that would equally well serve the pragmatic func-
tion that communism claims for epistemic evaluation. These alternative
sets can be radically alien: for a simple example, consider the set of
rules that omits some of our actual deductive rules, but includes some
rules as alien as Rami’s (see above); you can end up with a set of rules
as reliable and powerful as our actual set. So, if the only function of
epistemically evaluative terms is the coordinative one that epistemic
communism describes, then nothing privileges our actual rules over
alien sets that might include Rami’s or even Norman’s rules: it’s an
arbitrary convention that we promote these particular rules. Indeed,
from the conventionalist perspective, there is no reason to expect that
there is any feature ‘‘in virtue of which’’ our actual basic rules are
rational.19

18
See references in footnote 1.
19
Many theories compete to say what epistemic rationality holds in virtue of, includ-
ing reliabilism (e.g. Goldman (1979), developed in Goldman (1986)), conservatism
(e.g. Harman (1986) and Huemer (2007)), metasemantic accounts (e.g. Peacocke
(2004) and Boghossian (2003)), and pragmatic accounts of a Reichenbachian sort
(e.g. Enoch and Schechter (2008), Wright (2004), and Wedgwood (2011)).
As every theory so far proposed faces objections, it makes sense to ask whether
there was ever reason to expect that a successful theory exists.

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 15

 Defending this project requires, among other things, arguing that
epistemic terms and concepts serve no further function beyond the
coordinative one; this requires arguing no such function is served in
either non-simple linguistic uses of ‘rational’ or in our deliberative
thoughts involving the concept of rationality; but this argument can be
made. And any defense of conventionalism also requires a patient dis-
cussion of why Quinean objections don’t apply; but this also can be
done.20

References
Boghossian, Paul (2003). ‘‘Blind Reasoning.’’ Proceedings of the Aristo-
telian Society Supplementary Volume 77(1): 225–248.
—— (2008). ‘‘Epistemic Rules.’’ The Journal of Philosophy 105(9): 472–
500.
BonJour, Laurence (1980). ‘‘Externalist Theories of Empirical Knowl-
edge.’’ Midwest Studies in Philosophy 5: 135–150.
Carey, Susan (2009). The Origin of Concepts. Oxford: Oxford Univer-
sity Press.
Carroll, Lewis (1895). ‘‘What the Tortoise Said to Achilles.’’ Mind
4(14): 278–280.
Cohen, Stewart (1984). ‘‘Justification and Truth.’’ Philosophical Studies
46(3): 279–295.
Craig, Edward (1990). Knowledge and the State of Nature. Oxford:
Oxford University Press.
Divers, John. (2010). ‘‘Modal Commitments.‘‘ In Bob Hale and Aviv
Hoffman (eds.), Modality: Metaphysics, Logic and Epistemology,
189–219. Oxford: Oxford University Press.
Enoch, David and Joshua Schechter, (2008). ‘‘How Are Basic Belief-
Forming Methods Justified?.’’ Philosophy and Phenomenological
Research 76(3): 547–579.
Field, Hartry (1994). ‘‘Deflationist Views of Meaning and Content.’’
Mind 103(411): 249–285.
—— (2000). ‘‘A Priority as an Evaluative Notion.’’ In Paul Boghossian
and Christopher Peacocke (eds.), New Essay on the A Priori.
Oxford: Oxford University Press.
—— (2008). Saving Truth from Paradox. Oxford: Oxford University
Press.

20
For their immense help over the course of the preparation of this paper, I’d like to
thank Ray Buchanan, Hartry Field, John Morrison, Adam Pautz, Jim Pryor, Karl
Schafer, Jonathan Dancy and participants in his spring 2010 graduate seminar at
UT Austin, especially David Enoch and Clayton Littlejohn, participants in the phi-
losophy faculty reading group at UT Austin, and an audience at the 2011 Rutgers
Epistemology Conference.

16 SINAN DOGRAMACI
—— (2009). ‘‘Epistemology without Metaphysics.’’ Philosophical Stud-
ies 143(2): 249–290.
Gibbard, Allan (1990). Wise Choices, Apt Feelings. Cambridge, MA:
Harvard University Press.
—— (2003). Thinking How to Live. Cambridge,MA: Harvard Univer-
sity Press.
Goldman, Alvin (1979). ‘‘What is Justified Belief?‘‘ In George Pappas
(ed.), Justification and Knowledge, 1–23. Dordrecth: Reidel.
—— (1986). Epistemology and Cognition. Cambridge, MA: Harvard
University Press.
—— (2009). ‘‘Internalism, Externalism, and the Architecture of Justifi-
cation.’’ The Journal of Philosophy 106(6).
Hare, R. M. (1998). ‘‘Prescriptivism.’’ In Edward Craig (ed.), Routledge
Encyclopedia of Philosophy. London: Routledge.
Harman, Gilbert (1986). Change in View. Cambridge, MA: MIT Press.
Horwich, Paul (1990/98). Truth. 2nd edition. Oxford: Oxford Univer-
sity Press.
Huemer, Michael (2007). ‘‘Compassionate Phenomenal Conservatism.’’
Philosophy and Phenomenological Research 74: 30–55.
James, William. (1897/1956). ‘‘The Will to Believe.‘‘ In The Will to
Believe and Other Essays in Popular Philosophy. New York: Dover.
Kelly, Thomas (2003). ‘‘Epistemic Rationality as Instrumental Ratio-
nality A Critique.’’ Philosophy and Phenomenological Research
66(3): 612–640.
Kment, Boris (2006). ‘‘Counterfactuals and the Analysis of Necessity.’’
Philosophical Perspectives 20(1): 237–302.
Leeds, Stephen (1978). ‘‘Theories of Truth and Reference.’’ Erkenntnis
13(1): 111–129.
Peacocke, Christopher (2004). The Realm of Reason. Oxford: Oxford
University Press.
Pollock, John and Joseph Cruz, (1999). Contemporary Theories of
Knowl edge. Lanham, Maryland: Rowman & Little field.
Pronin, Emily, Daniel Lin and Lee Ross (2002). ‘‘The Bias Blind Spot:
Perceptions of Bias in Self Versus Others.’’ Personality and Social
Psychology Bulletin 28(3): 369–381.
Quine, Willard Van Orman (1970). Philosophy of Logic. Cambridge,
MA: Harvard University Press.
Reynolds, Steven (2002). ‘‘Testimony, Knowledge, and Epistemic
Goals.’’ Philosophical Studies 110(2): 139–161.
Smith, Adam (1759/1976). The Theory of Moral Sentiments. Oxford:
Oxford University Press.
Wedgwood, Ralph (2002). ‘‘Internalism Explained.’’ Philosophy and
Phenomenological Research 65(2): 349–369.

REVERSE ENGINEERING EPISTEMIC EVALUATIONS 17

—— (2011). ‘‘Primitively Rational Belief-Forming Practices.’’ In Reis-
ner and Steglich-Petersen (eds.), Reasons for Belief, chapter 9.
Cambridge: Cambridge University Press.
Weinberg, Jonathan, Shaun Nichols and Stephen Stich, (2001).
‘‘Normativity and Epistemic Intuitions.’’ Philosophical Topics 29(1
& 2): 429–460.
Williamson, Timothy (2007). The Philosophy of Philosophy. Oxford:
Blackwell.
Wright, Crispin (2004). ‘‘Warrant for Nothing (and Foundations for
Free)?’’ Aristotelian Society Supplementary Volume 78(1): 167–212.

18 SINAN DOGRAMACI