The Value of Cognitive Values

Heather Douglas
To be presented at PSA 2012 and published in the proceedings
Word Count: 4914
Abstract (100 words):
Traditionally, the cognitive values have been thought to be a collective pool of
considerations in science that frequently trade against each other. I argue here that a finer
grained account of the value of cognitive values can help reduce such tensions. I separate
the values into three groups, minimal epistemic criteria, pragmatic considerations, and
genuine epistemic assurance, based in part on the distinction between values that describe
theories per se and values that describe theory-evidence relationships. This allows us to
clarify why these values are central to science and what role they should play, while
reducing the tensions among them.
Introduction
The value of cognitive values (also called theoretical virtues or epistemic values) has
been underdeveloped in philosophy of science. They have largely been considered
together in one group, and when examined in this light, they seem to trade off against one
another, creating as much tension as guidance for scientific inference. Although some
work has examined a particular value in greater depth and attempted to ground a
justification for its importance in an epistemic argument (e.g. Forster & Sober 1994),
for the most part, the values have been justified collectively and historically, i.e.,
that some set of values is (by and large) what has been important to scientists in their
practice, and that that should be good enough for philosophers of science (e.g., Kuhn
1977).
This paper will attempt a more robust justification. Through the tactic of organizing the
conceptual terrain of cognitive values, I will argue that there are at least three distinct
groups of values that normally get lumped together. Once the values are divided into
these groups, it is clearer why the values are important and what their value to science
and to scientists is. Justifications, clarifying the value of cognitive values, then follow.
Creating these divisions requires finer grained appraisals of the values than has been
customary. For example, internal consistency will be considered distinct from external
consistency. Simplicity has two distinct aspects as well, as does scope. This paper does
not make the claim that the terrain mapped here provides a complete account of these
values, but the kind of complexity presented can be a starting point for further
discussions and amendments.
Another benefit of clarifying the terrain is that the supposed tensions among the values
prove to be far less common and problematic than is often presumed. Once the bases for
the values becomes clearer, their functions in science become clearer, and thus which
should be important when is clarified. In addition, as we will see, the values within a
group are shown to often pull together rather than against each other.

Finally, organizing the terrain and mapping the value of cognitive values will also enable
us to address the criticisms raised concerning the canonical distinction between
epistemic/cognitive and non-epistemic/non-cognitive values (e.g. Rooney 1992) and
criticisms over what should count as a cognitive/epistemic value (e.g. Longino 1996).
First, I will provide a brief overview of how the standard view on cognitive values
developed. Then, I will offer a more nuanced terrain for those values than has been
traditionally offered. I will proceed to show how both tensions among the values are
reduced (albeit not eliminated) and how the justifications for the various values are
clarified. Finally, I will draw implications from this re-organization of the terrain.
A Brief History of Cognitive Values
Philosophers of science have long referred to and discussed various qualities of scientific
claims deemed important in science. In the 20th century, philosophers such as Duhem
(e.g., 1906, 171, 217), Popper (e.g., 1935, 61-73, 122-128) and Levi (1960, 354; 1962,
49) famously described a range of qualities (and sometimes provided reasons for the
importance of those qualities). But it was not until Kuhns 1977 paper that these qualities
became widely known as values, and the discussion was framed in terms of values
internal to science. For Kuhn (1977), McMullin (1983), Laudan (1984), and Lacey
(1999), the values were a collective (if evolving) set. And there were clear tensions and
tradeoffs among the various values or virtues thought relevant at any given time. One
might gain scope in a theory, but lose precision. One might gain simplicity, but lose
scope. Understanding the history of science meant understanding how scientists made
those trade-offs (or shifted their interpretation of those values) in the course of scientific
debate.
But the collective pool of these values turns into a problematic swamp when one attempts
to find a grounding for the values. This problem was worsened by the tendency of
philosophers, in an attempt to make the values appear less overwhelming, to collapse
various attributes together. Thus, although some distinguished internal consistency
(minimal logical consistency of a theory) from external consistency (broader
considerations of whether a theory fit with prevailing scientific views), other
philosophers collapsed the two, and considered consistency tout court (e.g., Kuhn 1977,
357 vs. McMullin 1983, 15) This makes it harder to see how to justify consistency.
While internal consistency can be viewed as a minimal requirement of empiricism
(Duhem 1906, 220; Popper 1935, 72), external consistency is nothing of the sort, and is
valuable only insofar as ones confidence in the rest of scientific theory is high. Or
consider how explanatory power can be viewed either as an ability of a theory to
elucidate particular pieces of evidence with great detail or as an ability of a theory to
bring under one conceptual umbrella multiple disparate areas (which can also be
conflated with scope). Both are clearly valuable, but for quite different purposes and
reasons.

It is time to extricate ourselves from this swamp. Laudan (2004) made the first steps in
this direction when he divided theoretical virtues into those that were genuinely epistemic
(truth indicative) and those that were cognitive (valued by scientists for other reasons).
He suggested that few of the traditional theoretical virtues (construed as the swampy
collective described above) have genuine epistemic (that is, truth-indicative) merit. Two
that did (on his view) were internal consistency and empirical adequacy. Laudans
distinction is a good start on the problem, but I will go further here, dividing up the
terrain of cognitive values further in an attempt to elucidate their strengths, their
purposes, and their justifications.
The Terrain of Cognitive Values
Two distinctions will help further our project. First, following both Laudan (2004) and
Douglas (2009), we can distinguish between ideal desiderata and minimal criteria. We
might prefer one grand, simple, unified theory of great scope that explains everything, but
in practice we are willing to settle for less. (Indeed, some arguments for pluralism
suggest we should be happy with a complex plurality of perspectives. See, e.g., Kellert,
Longino & Waters 2006; Mitchell 2009.) In contrast, there are some virtues or values
that any acceptable scientific theory must instantiate (e.g. internal consistency). We
might accept a theory that falls short on these criteria out of shear desperation, but we
would know something was wrong and work furiously to correct it.
Second, it is important to note that in discussing the set of cognitive values, philosophers
have lumped together two different kinds of things in science to which cognitive values
can apply. By apply, I mean that which the values are thought to describe, or the object
of instantiation for the value (i.e., what has the value). The object of instantiation can
either be a theory per se or the theory in relation to the evidence thought to be relevant to
it. There are thus two different directions for assessment when using cognitive values:
are we describing the theory itself or the theory in relation to the available evidence?
To see how crucial these two different targets for cognitive values can be, consider the
value of scope. If we are talking about a theory with scope (and just the theory), the
theory might have the potential to apply to lots of different terrain or to wide swaths of
the natural world (i.e. the claims it makes are of broad scope), but whether it in fact does
so successfully can still be up in the air. Any proposed grand unified theory can be
considered to have scope in this senseit has broad scope, but not in relation to any
actual evidence yet gathered under that scope. Contrast that with a theory that already
does explain a wide range of evidence and phenomenaso that the scope applies to a
theory in relation to broadly based evidence (e.g. evidence from different phenomena or
evidence gathered in different ways). Here the value of the cognitive value is quite
different, and brings with it an epistemic assurance from the diversity of evidence
supporting the theory.
A similar point can be made with regards to simplicity. A simple theory (that is, just a
simple theory, and not where simplicity is describing a relation to evidence) might be
prima facia attractive, but unless we think the world actually is simple, we have little

reason to think it true. A simpler theory, all other things being equal, is not more likely to
be true. Contrast this with a theory that is simple with respect to the complex and diverse
evidence that it captures. The simpler theory, in relation to the evidence it explains, is
more likely to not be overfit to the evidence and thus more likely to be predictively
accurate. (Forster & Sober 1994) In such a case, simplicity has genuine epistemic
import.
With these two distinctions in mind1) what we want our values for (minimal criteria
vs. ideal desiderata) and 2) to what the value applies (the theory per se vs. the theory with
respect to evidence)we can turn to the terrain for such values. There are three groups
into which we can divide the cognitive value terrain:
Group 1: Values that are minimal criteria for adequate science
There are values that are genuinely truth assuring, in the minimal sense that their absence
indicates a clear epistemic problem. If a claim or theory lacks these values, we know that
something is wrong with our empirical claim. Thus, these are truly minimal criteria,
values that must be present if we are to be assured we are on the right track. These values
include internal consistency (which is about the theory per se) and empirical adequacy
(as measured against existing evidence, not all possible evidence, and thus is about the
theory with respect to evidence). Philosophers as diverse as Duhem (1906), Popper
(1935), Laudan (2004), and Douglas (2009) have noted these values as minimal criteria.
This group could be divided along the lines of Group 2 and 3 below using the second
distinction (regarding the instantiation of the value), but because it is so small, I leave
them together here. Because both of these minimal criteria have clear epistemic import
(theories failing these criteria are not good candidates for our beliefs), keeping them in
the same group helps clarify their function.
Group 2: Values that are desiderata when applied to theories alone
There are values that, when instantiated solely by the theory or claim of interest, give no
assurance as to whether the claims which instantiate them are true, but give us assurance
that we are more likely to hone in on the truth with the presence of these values than in
their absence. As such, these might be considered strategic or pragmatic values. Douglas
(2009) emphasizes the term cognitive values, as an aid to thinking; Dan Steel has called
them extrinsic epistemic values (2010). These include scope, simplicity, and (potential)
explanatory power. When theories (or explanations or hypotheses) instantiate these
values, they are easier to work with. Simpler claims are easier to follow through to their
implications. Broadly scoped claims have more arenas (and more diverse areas) of
application to see whether they hold. Theories with potential explanatory power have a
wide range of possible evidential relations. (I say potential because if the theory has
actual, known explanatory power, that implies that evidence is already gathered under its
umbrella and this would bring us to the next category of values.) It is easier to find flaws
in the claims and theories that instantiate these values. It is easier to gather potentially
challenging (and thus potentially strongly supporting) evidence for them. In this sense,
all of these values fall under the rubric of the fruitfulness of the theory.

Group 3: Values that are desiderata when applied to theories in relation to evidence
Finally, we should consider values that might sound similar to pragmatic cognitive values
(group 2), but because they qualify the relationship between theory and evidence, rather
than just theory itself, they provide a different kind of assurance. Whereas group 1
assured us that we have a viable scientific theory (genuine epistemic assurance), and
group 2 assured us that if we were on the wrong track, we should find out sooner than
otherwise, group 3 provides a particular kind of genuine epistemic assurance. It provides
assurance against ad hocery, and thus assures us that we are not making a particular kind
of mistake. One of our most central concerns in science is that we have made up a
theory that looks good for a particular area, but all we have done is make something that
fits a narrow range of evidence. If our theories are ad hoc in this way, they will have
little long term reliability or traction moving forward. Instantiation of these values in the
relation between the theory and the evidence that supports it provides assurance that we
have not just made something up. If a diverse range of evidence can be explained, or the
theory fits well with other areas of science (and, crucially, the evidence that supports
them), or the theory makes successful novel predictions, we gain precisely the assurance
we need. For this reason, these values have genuine positive epistemic import. These
values include unification (in terms of explanatory scope, simplicity, external consistency,
and coherence), novel prediction, and, modifying these values with an additional layer,
precision. (I discuss this group further below.)
What does this map of the terrain clarify? First, with this map we can see that the values
do have justifications independent of scientists historical reliance on them. We can
articulate reasons why a scientist should care about these values and clarify what they are
good for. There are clear epistemic reasons (independent of any particular objectives of
science at any particular period) for demanding that scientific theories be internally
consistent and empirically competent. And there are good epistemic reasons for
preferring scientific theories which have a broad range of evidence that support them or
that instantiate other values in group 3 (more on this below). Finally, there are good
pragmatic reasons for scientists to run with a simpler, broader, or more fruitful theory
first (group 2) if one is trying to decide where to put research effort next.
Second, as I will argue below, the idea that the values are in a collective pool and pull
against each other is misguided. Having this map makes it clearer what the purposes of
the values are, and shows that the tensions among the values are not as acute or
problematic as they appear when they considered as a collective pool.
Reducing the Tensions among the Values
There are two possible sources of tensions within the terrain I have mapped above. The
first arises from tensions among the groups of values. The second arises from tensions
within each group. I will address each of these in turn as I argue that tensions with this
map have been reduced, albeit not eliminated.

Among the groups, one reduction in tension should be immediately clear.

Minimal criteria do not (or at least, should not) pull against pragmatic fruitfulness
concerns of group 2 or the epistemic assurance concerns of group 3. Minimal criteria
come first, and both must be met. Indeed, one cannot tell whether one has an empirically
competent theory without minimal internal consistency. Now, in practice, scientists may
still choose to pursue the development of a theory with characteristics of group 2 even in
the face of failings in group 1 (minimal criteria). But this must be done with the full
acknowledgement that the theory is inadequate as it stands, and that it must be corrected
to meet the minimum requirements as quickly as possible. Although philosophers like to
quip that every scientific theory is born falsified, no scientist should be happy about it.
Once the remaining values are divided into the pragmatic cognitive values (instantiated
by theories onlygroup 2) and the epistemic anti-ad hocery assuring values (instantiated
by the relations between theories and evidencegroup 3), the two groups have less
problematic tension within each than has been generally thought.
Consider the possible tensions within the pragmatic cognitive valuesgroup 2. Recall
that within this group, the values describe theories or claims on their own, independent of
the evidence which may or may not support them. In this group, all of these values are
ultimately about the fruitfulness of the theory, the ease with which scientists will be able
to use the theory in new contexts (not necessarily successfully), to devise new tests for
the theory, and thus refine, revise, or if need be overhaul completely, the theory. It is true
that some scientists will find scope an easier handle with which to further test a theory, as
they will find it more amenable to apply the theory in a new arena to which the broadly
scoped theory is applicable, and some scientists will find simplicity an easier handle with
which to devise further tests. So some tensions may remain around the issue of what will
be fruitful for different scientists. But this need not create any epistemic worries, for
three reasons. First, the proof will be in the pudding for fruitfulness, and the pudding is
relatively straightforward to assess. If the theory cannot be used to devise additional
tests, if the scientists are unable to use the aspects of the theory that instantiate the value
they prefer, then the value is of no further use in that case. We will be able to tell readily
if the instantiation of a pragmatic-based value in fact proves its worth. Second, because
this category of values does not provide direct epistemic warrant, but is instead focused
on the pragmatic issue of the fruitfulness of a theory, there is little reason to be concerned
about divergent scientific perspectives on these values. None of these pragmatic values
provides a reason to accept a theory as well-supported or true or reliable at the moment.
Group 2 values are simply not epistemic. Third, social epistemological approaches to
science (e.g. Solomon 2001, Longino 2002) have made it quite clear that having diverse
efforts in scientific research is a good thing for science. It has been argued that diversity
of efforts in science is crucial for the eventual generation of reliable knowledge. So
having diverse views about what makes a theory fruitful is likely to be good for science.
In sum, the values in this group are pragmatic, they are easily assessable by external
criteria (are more new tests being produced?), their diversity supports a diversity of
epistemic effort, and yet, they do not have direct epistemic import. Whatever tensions
arise here can play out in diverse efforts of scientific practice.

Consider next the possible tensions within group 3. Because these values do have
genuine epistemic import, tensions among them would be central to the problem of
scientific inference and the epistemic assessment of scientific theories. But when
examining these values as instantiated by the relation between theories and the evidence
that supports them, there is less tension among these values than might be initially
supposed. For example, while simplicity, scope, and explanatory power are often thought
to pull against each other when considering theories alone (group 2), they pull together
when considering a theory in relation to evidence (group 3). A theory that has broad
scope over diverse evidence is also simple with respect to that diverse evidence, unifies
that diverse evidence, and has explanatory power over that evidence. Indeed, it is this set
of relations that Paul Thagard has formalized under his conception of coherence.
(Thagard 2000) Scientists might disagree over which evidence is more important to
unify or explain under a particular rubric, either because of different purposes or because
of different views on the reliability of the evidence under consideration. But that is a
disagreement over which instantiation of a cognitive value is more important, not a
disagreement based on tensions among values.
Yet there are still some tensions in group 3. For example, predictive accuracy (or the
value of the novel prediction) might pull against the considerations captured by
coherence. And indeed, when faced with such a tension, scientists can legitimately
disagree, some scientists finding greater epistemic assurance in the successful novel
prediction and other scientists finding greater epistemic assurance in the successful
unification of evidence or the explanatory power/coherence of a theory. When we have
both together, both successful explanation of the available evidence and a surprising
prediction (use novel or temporally novel), we have Whewells consilience (Fisch 1985),
which is perhaps the strongest epistemic assurance we have available to us. When
consilience is on the table, it is hard for other theories to compete. But we are not always
so lucky. Hence genuine epistemic tension is possible here.
There is an additional qualifier for the value considerations of group 3. Whether we are
considering the relation between theory and evidence that is some form of coherence or
some form of prediction, the precision or tightness of fit between the theory and evidence
also matters. The more precise the explanatory relations between theory and evidence, or
the more precise the prediction and the evidence that tests it (having just one or the other
is not helpful), the more we gain the epistemic assurance of group 3. This assurance is
that we have not just made our theories up, that they have some empirical grip on the
worldthey are fundamentally anti-ad hocery assurance. The more precision we have in
the relations between theory and evidence, the more assurance we get. The more
successful predictions we have, the more assurance we get. The more coherence or
explanatory power over diverse evidence we have, the more assurance we get. Because
there are these different sources of this kind of assurance, there will be tensions among
them in practice. But hopefully why these tensions arise, and what should be done about
them, will be clearer.
So what of tensions between the values of group 2 and group 3? These two groups aim at
different purposes, and thus any apparent conflict can be managed. It is particularly

important to note that group 2, the pragmatic cognitive values, have no bearing on what
should be thought of as our best supported scientific knowledge at the moment. Just
because a theory looks fruitful (whether because of its innate simplicity, scope, or
potential explanatory power) is no reason to think it more reliable now than any other
narrower or more complex theory. If one needs epistemic assurance, particularly for an
assessment of our best available knowledge at the moment, group 3 is where one should
look (after the requirements of group 1 are met). When one needs to figure out what
should be said about the state of knowledge now, pragmatic fruitfulness (group 2)
concerns have no bearing. When one wants to justify future research endeavors, such
pragmatic concerns are central.
In sum, there are no tensions among the groups: group 1 trumps groups 2 & 3, and
groups 2 & 3 have different purposes. Within the groups, there are no tensions within
group 1, there are productive tensions within group 2, and there remain some tensions
within group 3. Thus, while tensions among values remain, they are much reduced from
the traditional view. With a clearer account of the bases for such values, we can see their
function more clearly, and thus their purposes.
Implications
In earlier accounts of the theoretical virtues, the tensions among them were thought to
explain how scientists at any given moment could rationally disagree with each other
different scientists focused on different virtues. Does my organization of the theoretical
virtues dissolve this ready-made explanation for rational disagreement? No-- there are
still resources we can draw upon to explain disagreement. So, for example, one can still
see a tension between the explanatory scope of a theory (with respect to available
evidencegroup 3) and the predictive precision of its competitor. Such a tension will
likely continually arise in scientific practice. Or, consider the tension between a wellsupported theory (with group 3 values supporting it) and an underdeveloped theory (with
lots of group 2 values and thus lots of potential). The explanations of divergent choices
that we give, scientists being risk-takers with new theories or with staying with the older,
more developed theories, still hold in the account given here, but with a sharper
understanding of the source of the divergent choices. Indeed, we should help scientists
distinguish an epistemic assessment from a pragmatic fruitfulness assessment in their
commitments to scientific theories. Finally, one could also use the account of the place
of social and ethical values given in Douglas 2009 to show how concerns over the
sufficiency of evidence (driven by social or ethical values) could generate rational
disagreement among scientists (as Douglas argues ethical values in the assessment of
evidential sufficiency is a rational role for those values).
So what has been gained by organizing and explicating the various values of cognitive
values? First, we can see more clearly where and why such values are indeed valuable.
The justification need no longer rest on the contingency of the history of science
(although it is certainly illuminated by the history of science). This allows us to note why
these values have seemed so central. Groups 1 & 3 have genuine epistemic import, and
thus do not bleed across the epistemic/non-epistemic boundary (although their

instantiation depends on the available evidence which does depend on cultural values).
The pragmatic group 2 can have clear cultural influences on it. Rooneys concerns
(1992) are thus illuminated. It also allows us to assess proposals for alternative sets of
values (e.g., Longino 1996). We can consider alternative values under the groups
proposed and see if they assist us in reaching our goals.
Second, we can now address the reference often made to these values in other debates
with greater conceptual clarity. For example, when critics of the value of prediction (as
opposed to accommodation) (e.g., Harker 2008, Collins 1994) attempt to reduce the value
of novel prediction to accommodation plus a theoretical virtue (such as unification or
explanatory power), we can see both what might motivate such an attempt (they are
drawn to the power of group 3) and why it is misguided (the value of novel prediction
can be in tension with the value of unification). Finally, if this is indeed a step forward in
the clarity of the terrain, there is perhaps hope for a renewed effort in a qualitative theory
of scientific inference. But that work must await another paper.

References
Collins, Robin. 1994. Against the Epistemic Value of Prediction over Accommodation.
Nous 28,2: 210-224.
Douglas, Heather E. 2009. Science, Policy, and the Value-Free Ideal. Pittsburgh:
University of Pittsburgh Press.
Duhem, Pierre. 1906/1954/1982. The Aim and Structure of Physical Theory. Princeton:
Princeton University Press.
Fisch, Menachem. 1985. Whewells Consilience of InductionsAnd Evaluation.
Philosophy of Science 52: 239-255.
Forster, Malcolm and Elliott Sober. 1994. How to Tell when Simpler, More Unified, or
Less Ad Hoc Theories will Provide More Accurate Predictions. British Journal
for the Philosophy of Science 45: 1-35.
Harker, David. 2008. On the Predilections for Predictions. British Journal for the
Philosophy of Science 59: 429-453.
Kellert, Stephen, Helen Longino, and Kenneth Waters (eds.). 2006. Scientific Pluralism.
Minnesota Studies in Philosophy of Science vol. XIX. Minneapolis: University
of Minnesota Press.
Kuhn, Thomas. 1977. Objectivity, Value, and Theory Choice. In The Essential Tension.
Chicago: University of Chicago Press, 320-339.
Lacey, Hugh. 1999. Is Science Value Free? Values and Scientific Understanding. New
York: Routledge.

Laudan, Larry. 2004. The epistemic, the cognitive, and the social. In P. Machamer & G.
Wolters (Eds.) Science, Values, and Objectivity. Pittsburgh: University of
Pittsburgh Press, 14-23.
Levi, Isaac. 1960. Must the scientist make value judgments? Journal of Philosophy, 57,
345-357.
Levi, Isaac. 1962. On the Seriousness of Mistakes. Philosophy of Science 29: 47-65.
Longino, Helen. 1996. Cognitive and Non-Cognitive Values in Science: Rethinking the
Dichotomy. In Lynn Hankinson Nelson and Jack Nelson, (eds.) Feminism,
Science, and the Philosophy of Science. Dordrecht: Kluwer, 39-58.
Longino, Helen. 2002. The Fate of Knowledge. Princeton: Princeton University Press.
McMullin, Ernan. 1983. Values in Science. In Peter D. Asquith and Thomas Nickles,
(ed.), Proceedings of the 1982 Biennial Meeting of the Philosophy of Science
Association, Volume 1. East Lansing: Philosophy of Science Association, 3-28.
Mitchell, Sandra. 2009. Unsimple Truths. Chicago: University of Chicago Press.
Popper, Karl. 1935/1959/1992/2002. The Logic of Scientific Discovery. London:
Routledge.
Solomon, Miriam. 2001. Social Empiricism. Cambridge, MA: MIT Press.
Steel, Daniel. 2010. Epistemic Values and the Argument from Inductive Risk.
Philosophy of Science 77: 14-34.
Rooney, Phyllis. 1992. On Values in Science: Is the Epistemic/Non-Epistemic
Distinction Useful? In David Hull, Micky Forbes, and Kathleen Okruhlik (ed.)
Proceedings of the 1992 Biennial Meeting of the Philosophy of Science
Association, Volume 2. East Lansing: Philosophy of Science Association, 13-22.
Thagard, Paul. 2000. Coherence in Thought and Action. Cambridge, MA: MIT Press.