ON EPISTEMIC AGENCY

A Dissertation Presented

by

KRISTOFFER AHLSTROM

Submitted to the Graduate School of the

University of Massachusetts Amherst in partial fulfillment
of the requirements for the degree of

DOCTOR OF PHILOSOPHY

September 2010

Philosophy
© Copyright by Kristoffer Ahlstrom 2010

All Rights Reserved

 ON EPISTEMIC AGENCY

A Dissertation Presented

by

KRISTOFFER AHLSTROM

Approved as to style and content by:

_______________________________________
Hilary Kornblith, Chair

_______________________________________
Jonathan Vogel, Member

_______________________________________
Christopher Meacham, Member

_______________________________________
Adrian Staub, Member

____________________________________
Phillip Bricker, Department Head
Philosophy
To Radha
 ACKNOWLEDGMENTS

Thanks to Richard Feldman, Mikkel Gerken, Alvin Goldman, Stephen Grimm, Klemens

Kappel, Christoph Kelp, Michael Lynch, Anne Meylan, Erik Olsson, Nikolaj Jang Lee

Linding Pedersen, Casey Perin, Duncan Pritchard, James Simmons, Robert Talisse, and J.

D. Trout for helpful feedback on different parts of the present material. Thanks also to

Christopher Meacham, Adrian Staub, and Jonathan Vogel for valuable comments and for

agreeing to serve on my committee. A special thanks to Hilary Kornblith. I could not

have asked for a better philosophical mentor. Last but in no way least, thanks to Radha—

for everything.

v
 ABSTRACT

ON EPISTEMIC AGENCY

SEPTEMBER 2010

KRISTOFFER AHLSTROM, B.A., UNIVERSITY OF GOTHENBURG

Ph. D., UNIVERSITY OF MASSACHUSETTS AMHERST

Directed by: Professor Hilary Kornblith

Every time we act in an effort to attain our epistemic goals, we express our epistemic

agency. The present study argues that a proper understanding of the actions and goals

relevant to expressions of such agency can be used to make ameliorative

recommendations about how the ways in which we actually express our agency can be

brought in line with how we should express our agency. More specifically, it is argued

that the actions relevant to such expressions should be identified with the variety of

actions characteristic of inquiry; that contrary to what has been maintained by recent

pluralists about epistemic value, the only goal relevant to inquiry is that of forming true

belief; and that our dual tendency for bias and overconfidence gives us reason to

implement epistemically paternalistic practices that constrain our freedom to exercise

agency in substantial ways. For example, we are often better off by gathering only a very

limited amount of information, having our selection of methods be greatly restricted, and

spending our time less on reflecting than on simply reading off the output of a simple

algorithm. In other words, when it comes to our freedom to express epistemic agency,

more is not always better. In fact, less is often so much more.

vi
 TABLE OF CONTENTS

ACKNOWLEDGMENTS ...................................................................................................................... V

ABSTRACT .......................................................................................................................................... VI

CHAPTER

INTRODUCTION ................................................................................................................................... 1

1. WHAT IS EPISTEMIC AGENCY?.................................................................................................... 9

1.1. Agency and Amelioration ................................................................................................... 9

1.2. Doxastic Agency............................................................................................................... 13
1.3. Agency and Rational Authority ........................................................................................ 17
1.4. On the Epistemic Merits of Reflection ............................................................................. 23
1.5. Is Epistemic Agency a Myth? ........................................................................................... 28
1.6. Conclusion ........................................................................................................................ 32

2. I CAN’T MAKE YOU LOVE TRUTH (AND YOU CAN’T MAKE ME LOVE EVIDENCE)..... 35

2.1. Goals and Values .............................................................................................................. 35

2.2. Feldman on Epistemic Value ............................................................................................ 39
2.3. Is Feldman Mistaken About Value? ................................................................................. 43
2.4. True Belief and Human Flourishing ................................................................................. 46
2.5. Epistemic Value and the Ethics of Belief ......................................................................... 52
2.6. Conclusion ........................................................................................................................ 58

3. IN DEFENSE OF EPISTEMIC VALUE MONISM......................................................................... 60

3.1. Varieties of Value ............................................................................................................. 60

3.2. On the (Alleged) Surplus Value of Knowledge................................................................ 67
3.3. Meno and the Monist ........................................................................................................ 75
3.4. Justification and Non-instrumental Value ........................................................................ 86
3.5. The Trials and Achievements of Understanding .............................................................. 96

3.5.1. Objectual Understanding ................................................................................ 96

3.5.2. Understanding-Why ..................................................................................... 102

3.6. Conclusion ...................................................................................................................... 107

4. A FRAMEWORK FOR EVALUATION........................................................................................ 108

4.1. Defining Epistemic Betterness........................................................................................ 109

4.1.1. Questions and Informative Answers ............................................................ 111

4.1.2. Problems of Individuation ............................................................................ 115
4.1.3. Shared Agendas ............................................................................................ 119
4.1.4. On Incomparability....................................................................................... 120

7
 4.2. On the Specialized Role of Epistemic Evaluation .......................................................... 123

4.2.1. Degrees of Interest........................................................................................ 125

4.2.2. Banal and Harmful Pursuits.......................................................................... 128

4.3. Conclusion ...................................................................................................................... 133

5. EPISTEMIC AGENCY AND PATERNALISM: WHEN LESS IS MORE................................... 134

5.1. Reasons for Mandating Compliance............................................................................... 134

5.2. What is Epistemic Paternalism?...................................................................................... 140
5.3. Two Cases for Expansion ............................................................................................... 147

5.3.1. Prediction Models in Diagnosis and Prognosis ............................................ 148

5.3.2. Prediction Markets as an Alternative to Social Deliberation ....................... 152

5.4. Countervailing Reasons .................................................................................................. 158

5.4.1. Reasons for Not Relying on Prediction Models ........................................... 159

5.4.2. Reasons for Not Relying on Prediction Markets.......................................... 162

5.5. Conclusion ...................................................................................................................... 167

BIBLIOGRAPHY ............................................................................................................................... 168

8
 INTRODUCTION

When considering the last fifty years of epistemology in the analytical tradition—filled

with far-fetched thought experiments involving evil demons, reliable clairvoyants, and

disguised mules—it is easy to forget that there is an important sense in which

epistemology is an utterly practical discipline. Indeed, according to a long-standing

tradition, one of epistemology’s main missions is to provide hands-on advice, aiding the

epistemic inquirer in her pursuits. Understood thus, epistemology is not only normative,

in the sense that it concerns itself with specifically epistemic goods, but also

ameliorative, in that it attempts to say something constructive about how our odds of

actually attaining the relevant goods may be increased.

Ameliorative epistemology is not a new thing, nor is it “a departure from the

central issues taken up in epistemology,” as has been suggested by Richard Feldman

(1999, p. 172). To the contrary, it was practiced by two of the most central historical

figures in Western epistemology, i.e., John Locke and Rene Descartes. The ameliorative

ambition of the latter is most obvious in his Rules for the Direction of the Mind and

Discourse on the Method—two works explicitly in the business of epistemic prescription.

Similarly, one of Locke’s main concerns was the reformation of people’s intellectual

lives from one based on tradition to one based on reason. Part of the motivation for such a

reformation was, as Locke (1706/1996) writes in the posthumously published Of the

Conduct of the Understanding, simply that “there are a great many natural defects in the

understanding capable of amendment” (§2). In addition, bringing about the relevant kind

1
of amendment would, as Locke saw it, serve as a protection against the social conflicts

resulting from the breakup of medieval Christendom’s intellectual consensus.1

As the last fifty years of experimental psychology makes clear, however,

amelioration is a worthwhile endeavor even in the absence of an imminent threat of large-

scale, social turmoil. The often banal ways in which our dual tendencies for bias and

overconfidence constantly get in the way of accuracy and informed action—

unfortunately, often with far from banal consequences—is a sufficient reason for trying

to ameliorate our epistemic conduct and practices. On a more positive note, the very same

body of empirical research provides us with something that our epistemological ancestors

lacked, namely robust data on the systematic and often predictable ways in which we go

astray, as well as actionable evidence regarding what kinds of ameliorative

recommendations are likely to be effective.

This is not to suggest that the epistemologist should be expected to do the job of

the empirical psychologist.2 Instead, it is to seek a middle ground between those who

would suggest that epistemology just is empirical psychology (as some uncharitable

readings of Quine would have him say) and those who wish to engage in amelioration

without at all engaging with the relevant empirical literature. As for the latter, consider

Robert Roberts and Jay Wood’s recent defense of what they call regulative

epistemology—a kind of epistemology with explicit, ameliorative ambitions:

[…] to say that our virtue epistemology is regulative is not to deny that it’s
analytic. In fact, what we call analysis is our chief expedient of regulation. By the

1
See Wolterstorff (1996).
2
That is, unless the philosophers in question are trained to do so. In this context, consider
the interesting and in many ways commendable cross-scientific research associated with
so-called experimental philosophy.

2
 conceptual work that is distinctive of philosophical discourse, we propose to
facilitate the improvement of intellectual character. If conceptual analysis is done
right, it clarifies the character of the intellectual life in a way that can actually
help people live that life. Conceptual clarification is an important part of
education, and the improvement of intellectual character is a kind of education.
Conceptual clarification is not the whole of education [but] it is at least
something, and it is what the philosopher is well suited to contribute (Roberts and
Wood 2007, pp. 27-28).

Roberts and Wood’s commitment to epistemic improvement is commendable. But the

manner in which they restrict their contribution to that of conceptual analysis (as

evidenced by their book at large, not solely by the above passage) seems unnecessarily

limiting. Granted, philosophers receive ample training in conceptual clarification, which

is, clearly, something that comes in handy in all forms of inquiry. And it would probably

be a bad idea to assume that the skill-set of the typical philosopher naturally translates

into an aptitude for experimental work. But it is one thing to suggest that philosophers are

not well-suited for contributing empirical work of their own, and quite another to fail to

consult the vast amount of empirical work that is of high relevance to the multitude of

empirical bets—perhaps most pertinently, regarding what means of regulation will be

conducive to the relevant ends, given the surprising ways in which cognition sometimes

works—made in the course of any attempt to regulate epistemic behavior or shape

epistemic institutions. In fact, to be in the business of telling people how to think without

consulting the relevant empirical work borders on the irresponsible.

In contrast to the project of Roberts and Wood, the present study places itself

firmly within the ameliorative tradition of Locke and Descartes, as it has been developed

more recently within the naturalist movement, combining a commitment to amelioration

with the acknowledgement that such amelioration needs to be informed by the

3
psychology of cognition in order to be effective.3 While much has been said about this

naturalistic movement,4 the motivation for the particular investigation to be undertaken

below is that one notion in particular has received little to no attention in connection with

the ameliorative project, i.e., the notion of epistemic agency. This omission is unfortunate

since the notion of epistemic agency has the potential of playing a central role in

ameliorative epistemology—or so it will be argued.

But first, what is epistemic agency? As will be suggested in chapter 1, epistemic

agency is the kind of agency we express through actions performed in an effort to attain

our epistemic goals. This characterization begs for clarification on at least two points.

(Further clarifications will be postponed for now.) First, what are the actions by way of

which we express such agency? Second, what are our epistemic goals, or goal, if it turns

out that there is only one? Chapter 1 approaches the first question by considering three

candidate accounts of the relevant actions. The first account identifies epistemic agency

with doxastic agency, or decisions about what to believe. The second account identifies

epistemic agency with the kind of reflective agency or authority that some have

maintained that we have over our reflectively formed beliefs. The third account also

identifies epistemic agency with reflective agency, but motivates it with reference to the

epistemic benefits of reflecting.

All three accounts are rejected, the first one because it is committed to doxastic

voluntarism, the second one because it fails to delimit a sufficiently robust notion of

freedom to be relevant to amelioration, and the third one because it relies on too

optimistic a view of the epistemic merits of reflection. Contrary to other recent critiques

3
See, e.g., Bishop and Trout (2005), Kitcher (1992) and Goldman (1978).
4
See Kornblith (1994).

4
of the relevant accounts, however, it will not be concluded that the appeal to epistemic

agency, thereby, is “nothing more than a bit of mythology.”5 The notion of epistemic

agency is a useful one. More specifically, if we take epistemic agency to be expressed

through actions performed in an effort to attain our epistemic goals, a proper

understanding of the domain of epistemic agency serves to delimit exactly the kind of

activities with which ameliorative recommendations are to be concerned.

In fact, understood thus, the question of what actions are relevant to the

expression of epistemic agency is almost trivial. After all, to perform actions for the

purpose of forming true belief in a manner that furthers our epistemic goals is to engage

in inquiry. In other words, epistemic agency is constituted by all the things we do when

conducting inquiry. We gather information, mull over our data, choose among different

methods of investigation, and so on, and in so far as we are doing all of this in an attempt

to attain our epistemic goals, we are expressing our epistemic agency. At the same time,

the very triviality of such an account of the actions relevant to epistemic agency suggests

that the interesting questions regarding epistemic agency lie elsewhere. One such

question is the ameliorative question, namely: How can the ways in which we actually

express our agency be brought in line with how we should express our agency, given our

epistemic goals? However, this question begs a further one, posed already in the above,

namely: What are our epistemic goals?

This gets back to the observation that epistemology is normative. To talk about

epistemic goals is to talk about epistemic value, in that the goals of inquiry determine

what activities, states, processes, practices, and so on, are epistemically valuable.

5
Kornblith (forthcoming, p. 118; all page references to this work refer to the manuscript).

5
Chapters 2 and 3 develop and defend a particular account of such value that has come

under fire in recent discussions in epistemic axiology, namely that true belief is the only

goal of inquiry and, consequently, also unique in being of non-instrumental epistemic

value. More specifically, the idea that believing truly is valuable thus is defended on two

fronts, both from those denying that true belief is a non-instrumental epistemic value, and

from those who deny that true belief is the only non-instrumental epistemic value.

Chapter 2 takes on the former challenge by considering the case for a kind of

evidentialism that takes believing upon sufficient evidence to be the only goal of inquiry

and, consequently, also the only thing of non-instrumental epistemic value. For reasons

that will be discussed below, the strategy will not be to reject such a take on value, but

merely to protect the idea that truth is of non-instrumental epistemic value from the

evidentialist’s contention that it is not. More specifically, it will be argued that the

prospects for providing someone who does not share one’s love for the relevant non-

instrumental epistemic goods with reasons for starting to do so seem dim. However, the

point applies equally to the evidentialist as to those of us who value true belief non-

instrumentally from an epistemic point of view. Moreover, given that failure to convince

someone to value something is by no means a sufficient reason for ceasing to value that

thing, this dialectical fact renders neither axiological position illegitimate. In light of this,

subsequent chapters assume the position taken by most epistemologists to the effect that

true belief is, indeed, a non-instrumental epistemic good.

Chapter 3 takes up the second challenge, i.e., that of showing that true belief is

not just one non-instrumental epistemic value among many. Here, the strategy will be

less reconciliatory. Recently, epistemic value monism—i.e., the idea that believing truly

6
is unique in possessing non-instrumental epistemic value—has come under attack by

philosophers arguing that we cannot account fully for the domain of epistemic value in

monistic terms. However, it will be argued that the relevant critiques fail to establish any

such thing. For one thing, there is a presumption of monism due to considerations about

axiological parsimony. Granted, such a presumption would be defeated by positive

evidence to the effect that the relevant kind of monism makes us unable to fully account

for the domain of epistemic value. But a proper examination and subsequent rebuttal of

the most promising counterexamples to monism casts serious doubt upon the claim that

there is any such evidence. Consequently, epistemic value monism still stands, and true

belief is not only a but the only good of non-instrumental, epistemic value.

It does not follow from true belief being the only good of non-instrumental

epistemic value that all true beliefs are valuable thus. More specifically, chapter 4 argues

that the epistemic value of a practice is a function of the extent to which it yields true

beliefs that answer questions posed by relevant sets of inquirers. As such, the account

tells us something about what methods or practices inquirers should utilize, in so far as

they want answers to their questions. The remainder of the chapter is primarily concerned

with tracing the limits of this ‘should,’ and suggests that there are very real limits to the

extent to which we can make epistemological judgments about what questions inquirers

should pose (as opposed to about what practices to opt for, given a set of questions).

However, it is also argued that, rather than indicating a flaw in the account, this simply

suggests that it is able to account for the specialized nature of epistemic evaluation.

Armed with a framework for epistemic evaluation, chapter 5 returns to the

ameliorative question by illustrating the manner in which a proper understanding of the

7
actions and goal relevant to epistemic agency can be used to make recommendations

about how the ways in which we actually express our agency may be brought in line with

how we should express our agency. More specifically, a case is made for the claim that

our dual tendency for bias and overconfidence gives us pro tanto reason for mandating

compliance with methods that have been shown to increase our reliability. Moreover, it is

argued that mandating compliance thus would be epistemically paternalistic, and that we

have reason—indeed, not just pro tanto but all-things-considered reason—to practice

such paternalism on a wider scale than we are already doing.

To sum up, when properly understood, epistemic agency has nothing to do with

decisions about what to believe, and far less to do with reflection than some have

suggested. Instead, epistemic agency is constituted by all the things we do when

conducting inquiry. As such, epistemic agency is neither a myth, nor something the

expression of which should be considered epistemically valuable in its own right. After

all, given the plausibility of epistemic value monism and the prevalence of bias and

overconfidence, it turns out that we are often better off from the point of view of attaining

our epistemic goal by having our freedom to exercise agency be constrained by

paternalistic practices. In other words, when it comes to our freedom to express agency,

more is not always better. In fact, less is often so much more.

8
 CHAPTER 1

WHAT IS EPISTEMIC AGENCY?

1.1. Agency and Amelioration

What is it to express epistemic agency? It is, clearly, to do something. But, surely, that is

not sufficient. Someone falling down a set of stairs is doing something, namely falling

down a set of stairs, but not, thereby, expressing any agency, let alone any epistemic

agency. So, perhaps to express epistemic agency is to engage in a kind of goal-directed

action. But the relevant kind of action cannot be directed toward just any goal. A stunt

man deliberately and skillfully falling down a set of stairs on a movie set is not only

doing something but also performing an action directed at the goal of making it seem as if

she is falling down the stairs by accident, while doing so deliberately and (if skillfully)

with no significant risk of harm. So, what is the particular goal—or goals, if there turns

out to be several—toward which the kind of actions characteristic of epistemic agency

needs to be directed? Subsequent chapters will make a case for a particular answer to this

question, namely true belief, and use it to provide the foundation for an account of

epistemic value. For the remainder of this chapter, however, this answer will simply be

taken for granted.

Even if we assume that epistemic agency is expressed by way of actions directed

at the goal of forming true belief, several questions remain. First, what does it mean to be

“directed” at the goal of forming true belief? Does it imply that the agent in question

consciously entertains that goal whenever performing whatever actions are associated

9
with expressions of epistemic agency? That seems too strong. Many paradigmatic

instances of what we would intuitively want to categorize as expressions of epistemic

agency, such as the everyday inquiry of a scientific researcher, involve subjects that do

not consciously keep that goal in mind at all times. Surely, they could bring it to mind, if

properly prompted, and if we had to put any constraint on agency it would be that the

agent in question at least has the ability to entertain (and endorse) the relevant goal

consciously, although there is no need for her to actually do so in order to qualify as an

epistemic agent. Such a constraint would rule out what seem less controversial instances

of non-agency, such as that of the infant crawling across the floor in proto-inquiry, or an

ant foraging. This is not to suggest that epistemic agency is a prerogative of adult

humans. Some adults might lack the conceptual sophistication required to think of

themselves as engaging in a project directed at forming true belief, and some non-human

animals may be sophisticated enough to qualify as agents.

Second, do the actions directed at the goal of forming true belief need to be

successful, i.e., actually terminate in the attainment of true belief, in order to qualify as

expressions of epistemic agency? In one sense, clearly not. A researcher has not failed to

express epistemic agency just because she gets something wrong. That said, our intuitions

get less clear when we consider the other limiting case, i.e., someone sincerely pursuing

truth but doing a terrible job at it. Consider, for example, an astrologist who sincerely

believes that her study of the movements and relative positions of celestial bodies

provides her with predictive data about human behavior. Could she (qua astrologist) still

qualify as an epistemic agent? In so far as we are inclined to say “no,” perhaps on

account of how her ideas about what is truly predictive of what are so far off the mark,

10
we might want to put the following success constraint on epistemic agency: in order to

qualify as an epistemic agent, one’s methods of inquiry must be generally reliable with

respect to the relevant subject matters. Let us refer to a notion of agency invoking this

constraint as a strong notion of agency.

I hesitate to embrace such a strong notion because it places those in greatest need

of intellectual advice outside the scope of agency. Moreover, rejecting above success

constraint and, in effect, opting for what we may call a weak notion of agency, does not

serve to disqualify considerations about success from questions that may be posed in

terms of epistemic agency. On such a weak notion, aforementioned astrologist could

qualify as an epistemic agent; she would just be a terrible epistemic agent. Given that

diagnosis, we could admit her into the domain of epistemic agency, and then determine

how she could start doing things differently, in order to make more successful predictions

about human behavior. After all, if you are interested in providing useful intellectual

advice, what you are interested in doing are two things. First, you want to determine what

people engaging in actions directed (successfully or not) at the goal of forming true belief

are actually doing. Second, to the extent that they are unsuccessful in their pursuits, you

want to say something constructive about what the agents in question should be doing

differently in order to actually attain that goal.

This provides the first motivation for an inquiry into the notion of epistemic

agency: Contrary to what its absence in previous epistemological treatments might lead

us to believe, the notion of epistemic agency has an important role to play in ameliorative

epistemology in that a proper understanding of the domain of epistemic agency serves to

delimit exactly the kind of activities with which ameliorative recommendations are to be

11
concerned. One way to illustrate this point in more detail is by thinking of the domain of

actions relevant to epistemic agency in terms of a simple Venn diagram, as follows:

C
B

A
D

A encompasses the ways in which we actually express our agency. These actions are

relevant because, as noted above, our actual epistemic practices provide the point of

departure for any ameliorative advice. C encompasses the ways in which we could

express our agency, and C–A the ways in which we could express our agency differently

from what we are presently doing. Furthermore, A∩B designates the ways in which we

are doing things right from the point of view of attaining true belief, and B–A the ways in

which we could be doing things differently and for the better. Finally, everything that

falls outside the scope of C pertains to the ways in which we could not express our

agency, designated above by D. As will become clear in the next section, D is relevant

for a purely negative reason in that it tells us something about the ways in which we

cannot do things differently, and what we, consequently, have to work around rather than

with, as far as intellectual advice is concerned.

The second motivation for an inquiry into the notion of epistemic agency is what

most likely also provides the reason for the notion’s absence from the relevant

discussions, namely a series of accounts about what epistemic agency amounts to that are

12
of no help in so far as we are interested in amelioration. It is the burden of the remainder

of this chapter to consider some of the most prominent accounts of such agency,

demonstrate the ways in which they fall short as far as the ameliorative project is

concerned, and then show how the notion of epistemic agency can be re-established so as

to play an important role within that project, in line with what was just outlined.

1.2. Doxastic Agency

Above, it was argued that to express epistemic agency is to engage in actions directed

(consciously or not) at the goal of forming true belief. What are the relevant actions? One

way to approach this question is by considering one of the most fundamental objects of

epistemic evaluation, i.e., belief. Beliefs can be justified, rational, and constitute

knowledge. In addition, our beliefs make up the “map of neighbouring space by which

we steer,” as pointed out by Frank P. Ramsey (1931). Given that some ways of steering

are more successful than others (epistemically or otherwise), one straightforward way to

understand the actions of epistemic agency is in terms of decisions about what beliefs to

form (or not to form). Let us refer to the relevant kind of agency—if there is, indeed, such

agency—as doxastic agency.

The main problem with the suggestion that epistemic agency is to be identified

with doxastic agency is that it amounts to assuming doxastic voluntarism. That is, it

amounts to assuming that belief-formation is under our voluntary control and that we, in

effect, are able to form beliefs at will. This view seems doubtful, to say the least.

“Believing, we feel, is not something one can in any sense decide to do or refrain from, it

13
seems rather a condition in which one finds oneself,” as John Heil writes (1984, p. 56).

William Alston elaborates on a similar point:

Can you, at this moment, start to believe that the Roman Empire is still in control
of Western Europe, just by deciding to do so? If you find it incredible that you
should be sufficiently motivated to even try believing this, suppose that someone
offered you $500 million to believe it, and that you are much more interested in
the money than believing the truth. Could you do what it takes to get at the
reward? Remember that we are speaking of believing at will. No doubt, there are
things you could do that would increase the probability of your believing this
[but] can you switch propositional attitudes toward that proposition just by
deciding to do so? It seems clear to me that I have no such powers. Volitions,
decision, or choosings don’t hook up with propositional attitude inaugurations,
just as they don’t hook up with the secretion of gastric juices or with metabolism
(Alston 2005, p. 63).

At this point, the voluntarist may object that the aforementioned line of reasoning

presupposes a far too strong reading of “voluntary control.” More specifically, it might be

objected that it is being assumed that the relevant kind of voluntarism requires direct

control over our belief-forming processes, where having direct voluntary control implies

being able to bring about the object of control by a mere act of will, intention or volition.1

This is the kind of control we might have over mental imagery—i.e., our ability to bring

about mental pictures—but that we do not seem to have over our belief-forming

processes. However, the voluntarist may insist that there are other forms of control, most

pertinently various forms of indirect voluntary control, located on a continuum of

decreasingly direct control. For example, certain things may be brought about not by a

direct act of will, but by way of a chain of events, making up a single, uninterrupted act.

This is the sense in which opening my door or adjusting the temperature in my office is

1
The distinction between direct and indirect voluntary control is borrowed from Alston
(2005), as are the further distinctions between different kinds of control discussed below.

14
within my voluntary control; while I may not be able to open the door or alter the

temperature merely by an act of will, the opening of the door and altering of the

temperature is nevertheless within my (indirect) voluntary control, by virtue of it being

possible for me to bring about the necessary chain of events that is me getting up,

walking across the room, and opening the door or turning the knob on the thermostat.

Even further out on the continuum of indirect voluntary control are the kinds of

acts that do not constitute single, uninterrupted acts, like walking over and opening a door

or turning a knob, but rather a series of inter-related actions spread out over an extended

period of time. This is the sense in which looking for more evidence, taking steps to

engage in more thorough consideration and weighing of evidence, and deliberating over

how to direct my inquiry in a more strategic manner is within my long-range (indirect)

voluntary control. This is also the kind of control Roderick Chisholm seems to be getting

at in the following passage:

If self-control is what is essential to activity, some of our beliefs, our believings,

would seem to be acts. When a man deliberates and comes finally to a conclusion,
his decision is as much within his control as is any other deed we attribute to him.
If his conclusion was unreasonable, a conclusion he should not have accepted, we
may plead with him: “But you needn’t have supposed that so-and-so was true.
Why didn’t you take account of these other facts?” We assume that his decision is
one he could have avoided and that, had he only chosen to do so, he could have
made a more reasonable inference. Or, if his conclusion is not the result of a
deliberate inference, we may say, “But if you had only stopped to think”,
implying that, had he chosen, he could have stopped to think. We suppose, as we
do whenever we apply our ethical or moral predicates, that there was something
else the agent could have done instead (Chisholm 1968, p. 224).

As pointed out by Alston, however, any attempt to frame doxastic voluntarism in terms of

indirect voluntary control (long-range or otherwise) needs to ignore a crucial distinction,

15
namely the distinction between doing C to voluntarily bring about an E, and doing C to

voluntarily bring about a definite E. Alston elaborates:

In order that the phenomenon of looking for more evidence would show that we
have voluntary control over propositional attitudes, it would have to be the case
that the search for evidence was undertaken with the intention of taking up a
certain attitude toward a specific proposition. For only in that case would it have
any tendency to show that we have exercised voluntary control over what
propositional attitude we come to have (Alston 2005, p. 70).

In other words, the fact that we do have voluntary control over many actions that might

give rise to beliefs does not show that we have voluntary control over what beliefs we

form as a result of those actions. The same goes for various forms of indirect voluntary

influence—a kind of influence even further out on the continuum of decreasing

directness—where we may take steps that either bring to bear on candidates for belief or

on our general belief-forming habits and tendencies. That we may have such influence

over our beliefs does not show that doxastic voluntarism is true, i.e., that we may

voluntarily choose what specific propositions to believe, anymore than our ability to open

our eyes and look out a window on a sunny day shows that we may choose whether or

not to believe that it is sunny.

The implications of this are far from trivial as for how we are to understand

epistemic agency, since the failure of doxastic voluntarism suggests that the domain of

belief-formation not only does not but cannot coincide with the domain of epistemic

agency. Indeed, above considerations suggest that the two domains are completely

distinct, since one pertains to something that we do (i.e., agency), and the other to

something that simply happens to us (i.e., belief-formation). More than that, if the two

domains are incorrectly identified, the prescriptive import of epistemic agency, as it

16
relates to epistemic amelioration, will be reduced to a set of unhelpful intellectual

recommendations in terms of belief and its suspension—recommendations that, to

borrow Alston’s analogy, are no more helpful than dietary recommendations framed

exclusively in terms of the mechanisms of our digestive apparatus.

1.3. Agency and Rational Authority

At this point, it might be objected that the previous section invoked a too narrow notion

of our doxastic activities, as one involving mere responses to external stimuli. After all,

as human beings, we not only form first-order beliefs, such as perceptual belief; we also

reflect on our first-order beliefs, as a consequence of which we form certain reflective,

second-order beliefs, or beliefs about our beliefs.2 More than this, unlike belief, reflection

is not simply a condition that we find ourselves in; it is something that we do. For one

thing, we may choose when and on what to reflect—a fact that will be considered in the

next section. For another, the phenomenology of reflection surely suggests that we have

some say when it comes to what beliefs we end up with as a result of reflecting.

The latter idea is central to Richard Moran (2001).3 According to Moran,

reflective belief-formation, or what he calls deliberative belief-formation, is in an

important sense up to us. However, he explicitly rejects the idea that he is, thereby,

committing himself to any form of doxastic voluntarism: “The agency a person exercises

2
‘Reflection’ is sometimes used in a manner that does not (necessarily) involve any
second-order beliefs, as when we talk about reflecting on our data. That is not how the
notion of reflection will be used here.
3
Similar themes to those found in Moran can be found also in Korsgaard (1996), on
which parts of Moran’s investigation rest. However, given that Moran is more explicitly
concerned with belief-formation than Korsgaard is, I will focus on the former presently.

17
with respect to his belief and other attitudes,” he writes, “is obviously not like that of

overt basic actions like reaching for a glass” (p. 114). Still, we have a very real authority

over what beliefs we form, and this authority is what makes it possible for the agent to be

responsible for her beliefs:

Beliefs and other attitudes […] are stances of the person to which the demand for
justification is internal. And the demand for justification internal to the attitudes
involves a sense of agency and authority that is fundamentally different from the
various forms of direction or control one may be able to exercise over some mind
or other (Moran 2001, p. 114).

It should be noted that Moran’s concern with a sense of agency and authority, albeit,

clearly, inspired by the phenomenology of reflection, should not be interpreted merely as

a point about phenomenology. Granted, at times, Moran talks as if the authority in

question is simply a matter of the agent assuming responsibility for her actions (e.g., on

p. 131)—a way of framing the issue that is compatible with such assumptions being

metaphysically mistaken. However, the great majority of passages makes clear that

Moran is committing himself to a stronger claim, in taking it that reflective belief-

formation not only seems to be but really is up to us. Indeed, Moran’s book is fraught

with locutions to the effect that we make up our minds (p. 92) by stepping back from

some content of consciousness (p. 141) in order to call into question what we are inclined

to believe (p. 143) and decide or commit ourselves to believing something (p. 145).

So, how are we to interpret these locutions, without either committing Moran to

doxastic voluntarism, or taking them to be mere comments on the phenomenology of

reflection? Consider the following passage from Moran, succeeding a discussion of

Anscombe’s notion of intention:

18
 The stance from which a person speaks with any special authority about his belief
or his action is not a stance of causal explanation but the stance of rational
agency. In belief as in intentional action, the stance of the rational agent is the
stance where reasons that justify are at issue, and hence the stance from which
one declares the authority of reason over one’s belief and action. Anscombe’s
question “why?” is asking not for what might best explain the movement that
constitutes the agent’s action, but instead is asking for the reasons he takes to
justify his action, what he is aiming at. It is as an expression of the authority of
reason here that he can and must answer the question of his belief or action by
reflection on the reasons in favor of this belief or action. To do otherwise would
be for him to take the course of his belief or his intentional action to be up to
something other than his sense of the best reasons, and if he thinks that, then
there’s no point in his deliberating about what to do. Indeed, there is no point in
calling it “deliberation” any more, if he takes it to be an open question whether
this activity will determine what he actually does or believes. To engage in
deliberation in the first place is to hand over the question of one’s belief or
intentional action to the authority of reason (Moran 2001, p. 127).

Understood thus, there might, at first glance, seem to be a tension between the rationality

and the agency of rational agency. After all, how do we square the idea of agency or

freedom with the restrictive demands of reason and rationality? To understand Moran’s

answer, it might help to consider Descartes’ answer to a question not too different from

this one. In his Meditations on First Philosophy, Descartes argues that God has bestowed

us with free will. At the same time, Descartes takes it that the purest form of freedom is

the one exercised when the clarity and distinctness of a truth is so apparent that it prompts

more or less automatic assent, and the will is completely overpowered, so to speak, by the

psychological force of the intellect.4 At first, this might seem incompatible with freedom,

but Descartes denies that it is:

[…] the will simply consists in our ability to do or not to do something (that is, to
affirm or deny, to pursue or avoid); or rather, it consists simply in the fact that
when the intellect puts something forward for affirmation or denial or for pursuit

4
See Descartes (1641/1984, p. 48; AT VII 69).

19
 or avoidance, our inclinations are such that we do not feel we are determined by
any external force. In order to be free, there is no need for me to be inclined both
ways; on the contrary, the more I incline in one direction—either because I clearly
understand that reasons of truth and goodness point that way, or because of a
divinely produced disposition of my inmost thoughts—the freer is my choice
(Descartes 1641/1984, p. 40; AT VII 57-58).

In other words, to be free in the relevant sense is not to be able to opt out of the demands

of rationality, but rather to reason in ways that are sensitive to the demands in question.

In fact, according to Descartes, freedom takes its purest form when perfectly aligned with

those demands. Moran seems to be working with a relevantly similar notion of freedom,

as made clear in the following passage, where he discusses the relevant notions of

control, activity and responsibility, as they pertain to desires and belief (which, for

present purposes, are perfectly analogous, according to Moran):

The person’s responsibility […] is to make his desire answerable to and adjustable
in the light of his sense of some good to pursue. It is not a responsibility that
reduces to the ability to exert influence over one’s desires, and that is why the
idiom of “control” is misleading in this context. When the desire is (already) the
expression of the person’s reasons, there is no need for exerting any control over
it. As in the case of ordinary theoretical reasoning, which issues in a belief, there
is no further thing the person does in order to acquire the relevant belief once his
reasoning has led him to it. At the beginning of his practical reasoning he was not
aiming to produce a particular desire in himself (as he might with respect to
another person), but rather holding open his desire to how the balance of reasons
falls out (Moran 2001, pp. 118-119).

In other words, the relevant kind of authority, i.e., rational authority (p. 124), is one on

which we are not so much choosing what to reflectively believe—as in: choosing what

particular propositions to believe (or refrain from believing)—as simply leaving it up to

our reflective faculties to lead us toward the proper conclusion, under the assumption that

they are sensitive to the demands that rationality places on us. That is why Moran’s

20
account of agency is compatible with doxastic involuntarism; the kind of freedom or

agency associated with rational authority does not require that we can decide what beliefs

to form on the basis of reflection.

That said, if this is Moran’s take on what it is for the beliefs formed on the basis

of reflection to be “up to us,” his account of the kind of authority at issue is not relevant

to epistemic agency, in so far as the latter notion relates to epistemic amelioration. To see

why, return to the first motivation for an inquiry into the domain of epistemic agency

discussed above. What we noted was that, in so far as we are interested in providing

intellectual advice, we are interested in finding out two things, namely what people

engaging in actions directed at the goal of forming true belief are actually doing, as well

as what they should be doing differently, in so far as they could be doing better with

respect to attaining that goal. Given that the objective of amelioration is to provide

intellectual advice, that someone should do something differently presupposes that the

target activity is one over which she has effective voluntary control—which was exactly

what disqualified doxastic agency as a plausible account of agency. Again, in want of any

voluntary control over our beliefs, ameliorative recommendations cannot be formulated

in terms of decisions about what to believe, any more than dietary recommendations can

be framed exclusively in terms of the mechanisms of our digestive apparatus.

If Moran is right, we do have a kind of authority over our reflective beliefs, in so

far as our beliefs are formed under the rational authority of reason. As we have also seen,

however, Moran makes it clear that the kind of authority manifested under such

circumstances neither requires nor typically involves any voluntary control (let alone any

effective voluntary control) on part of the deliberating subject. While this

21
acknowledgement has him steer clear of doxastic voluntarism, it also serves to insulate

the relevant reflective processes from ameliorative recommendations. After all, to have

authority over what you believe upon reflection is to hand over your belief-formation to

the authority of reason (as Moran puts it), and to do so is exactly to not be able to believe

otherwise—nor, consequently, be able to be effectively advised to believe otherwise—

unless your reasons require you to believe otherwise.

Although this suggests that there is no agency of the kind relevant to amelioration

to be found in what Moran refers to as rational authority, it is important to note that this

does not serve to discredit his account, as far as his (non-ameliorative) project is

concerned. When encountering evidence that things are not going my way, I may have a

strong desire to believe contrary to what the evidence suggests. If I could believe (and not

believe) at will upon reflection, resisting the force of the evidence would be an easy feat.

That, however, does not change the fact that there is a sense in which I would not be

believing as I should, were I to resist the evidence thus. Differently put, as far as

epistemology is concerned, our belief-forming tendencies—reflective or not—should not

be sensitive to what we want to believe, but rather to what we have reason to believe. So,

perhaps it is not such a bad thing that our belief-forming tendencies are to a certain extent

(albeit not completely, as instances of wishful thinking go to show) insulated from our

desires and the influence of the will—which, of course, is exactly in line with Moran’s

account of rational authority.

22
1.4. On the Epistemic Merits of Reflection

The previous section suggested that, in so far as there is a sense in which what beliefs we

end up with as a result of reflection is up to us, this does not serve to delimit a notion of

agency relevant to amelioration. However, above discussions largely ignored a perfectly

legitimate domain of agency associated with reflection, identified in passing above when

it was noted that we may choose voluntarily when and on what to reflect (although, as we

have seen, not what beliefs to form as a result of such choices). To this extent, reflection,

clearly, is an action relevant to epistemic agency, since we can engage in it in a manner

that is directed at the goal of forming true belief. The questions to be discussed in this

section, however, is whether it is the only thing that we can do thus, or even something

that it generally is a good thing to do, given the goal in question. Let us attend to the

latter issue first.

Why is it a good thing to reflect? According to Ernest Sosa,

[…] reflection aids agency, control of conduct by the whole person, not just by
peripheral modules. When reasons are in conflict, as they so often are, not only in
deliberation but in theorizing, and not only in the higher reaches of theoretical
science but in the most ordinary reasoning about matters of fact, we need a way
holistically to strike a balance, which would seem to import an assessment of the
respective weights of pros and cons, all of which evidently is played out through a
perspective on one’s attitudes and the bearing of those various reasons (Sosa
2004, p. 292).

Notice that Sosa’s claim that reflection aids agency depends, crucially, on certain

assumptions about the epistemic merits of reflection. After all, it would make little sense

to talk about reflection aiding agency, if it turned out that reflection generally did not

weigh pros and cons properly, nor strike an epistemically appropriate balance between

reasons. If that turned out to be the case, I highly doubt that Sosa would still be inclined

23
to consider reflection aiding agency, in the relevant sense.5 In short, in so far as reflection

aids agency, it does so in virtue of the epistemic merits of reflection.

This, moreover, is completely in line what Sosa (1991) says in relation to his

account of reflective knowledge. According to Sosa, reflecting on your beliefs, and,

thereby, not only demonstrating a “direct response to the fact known but also [an]

understanding of its place in a wider whole that includes one’s belief and knowledge of it

and how these come about” (p. 240) will, at least in a preponderance of cases, make you

epistemically better off than not reflecting on your beliefs.6 Distinguishing reflective

from animal knowledge, i.e., the kind of knowledge resulting from a cognitive faculty

that is reliable with respect to a particular field of propositions, Sosa writes:

Since a direct response supplemented by such understanding would in general

have a better chance of being right, reflective knowledge is better justified than
corresponding animal knowledge (Sosa 1991, p. 240).

If Sosa is right, it would seem reasonable to have reflection play an important role in our

account of epistemic agency, since reflection would not only pertain to something we can

do but also to something the exercise of which serves the goals of such agency well. And

as we have seen, Sosa does, indeed, take reflection to be central to epistemic agency.

5
This is further evidenced by the fact that Sosa (in conversation) denies that that the kind
of empirical evidence to be discussed below, suggesting (it will be argued) that reflecting
does not generally make one better off from an epistemic point of view, gives us any
reason to re-evaluate the epistemic merits of reflection. Denying that such a re-evaluation
is called for would have been unnecessary, were the role of reflection in agency taken to
be independent of the epistemic merits of the former.
6
It is not completely clear on Sosa’s account whether the “understanding” in question
needs to be a result of conscious reflection, or whether a certain counterfactual sensitivity
to contrary evidence is sufficient. Some passages seem to suggest the latter (e.g., Sosa
1991, p. 240 and 2007, p. 111), others the former (e.g., 2007, p. 117). See Kornblith
(forthcoming) for a discussion.

24
However, if Sosa is not right in assuming that we in general are better off for reflecting as

opposed to not reflecting, then it would seem that, although epistemic agents may still

express their agency by reflecting, we lack good reason to endow reflection with a

particularly central role, at the expense of other cognitive tools or strategies with perhaps

more impressive epistemic credentials.

How do we determine the epistemic merits of reflection? According to Sosa,

cognitive faculties are field-specific.7 In other words, what is relevant for the epistemic

evaluation of a faculty is not simply reliability, but reliability with respect to a certain set

of propositions. And how are such fields individuated?

Just how fields are defined is determined by the lay of interesting, illuminating
generalizations about human cognition, which psychology and cognitive science
are supposed in time to uncover. […] Human intellectual virtues are abilities to
attain cognitive accomplishments in “natural” fields, which would stand out by
their place in useful, illuminating generalizations about human cognition (Sosa
1991, p. 236).

In other words, if we want to know how to individuate faculties for the purpose of

evaluating the reliability of those faculties, we need to look to cognitive science. So are

we, as a matter of empirical fact, in general better off from an epistemic point of view for

reflecting?

Relevant experimental evidence suggests that the answer is “no.” For one thing, it

is well known that we suffer from a variety of cognitive biases when reasoning under

uncertainty.8 This is not a problem for reflection per se, since many of the biases in

question pertain to non-reflective cognition. But keep in mind that the idea is supposed to

7
See Sosa (1991, p. 236).
8
See Gilovich et al. (2002).

25
be that, by reflecting on our first-order beliefs, formed non-reflectively, we will be made

better off from an epistemic point of view, due to the corrective influence of reflection.

Such an optimistic take on the power of reflection simply does not hold up in light of the

evidence, one important reason being that we systematically underestimate the extent to

which we (as opposed to everyone else) suffer from cognitive bias.9 More specifically,

when reflecting on our grounds for belief, we have a strong tendency to conclude that our

beliefs are well supported, even when they are not.10 For this reason, reflecting on our

grounds for belief does not invariably or even in general make us epistemically better off

than not reflecting.

In fact, we are sometimes worse off for reflecting, as opposed to not reflecting.

The last fifty years of predictive modeling suggest that we may often increase our

reliability significantly by rejecting labor intensive strategies that intuitively strike us as

epistemically responsible—such as collecting large quantities of data, weighing all

available evidence, and defecting from general rules in so far as the situation is perceived

to be relevantly unique—and instead opting for often extremely simple predictive

models.11 In fact, the greatest challenge for those developing such models is no longer to

identify reliable models, but to convince people that they will be better off for using

them. In other words, it is not lack of reliability but defection from policies urging that

we use the models in question that is the main challenge. Moreover, as we will discuss at

length in chapter 5, one of the primary activities driving up the defection rates is subjects

9
See Pronin et al. (2002).
10
This finding should be considered against the background of a large body of
experimental research suggesting that we have a general tendency to overestimate our
abilities (cognitive or otherwise). See Taylor (1989) for an overview of relevant research.
11
See Dawes et al. (2002).

26
reflecting on their reliability and coming to believe—falsely and as a result of

overconfidence—that they will be better off by not relying on the models in question.12 In

other words, reflecting on our grounds for belief sometimes makes us epistemically

worse off than we would have been, had we opted for strategies that do not involve

reflection.

What is the implication for how we ought to think about the relation between

reflection and epistemic agency? It is not that reflection is a useless intellectual tool, or

that reflecting always leaves us worse off than not reflecting. Nor is it that decisions

about when or on what to reflect may not constitute expressions of epistemic agency.

Reflection has just turned out to have a mixed epistemic track-record—like most of our

intellectual tools—and there is, as such, no reason to consider it to be more closely

related to epistemic agency than is any other goal-directed attempt to form true belief. In

short, from the point of view of agency, reflection is not special.

This gets back to a question raised above: Is reflecting the only action relevant to

epistemic agency? It will be suggested below that the answer is “no,” and if we want to

understand what agents can actually do to increase their chances for forming true belief,

and, thereby, identify the ways in which questions about agency may be relevant to

questions about amelioration, we need to broaden our understanding of the domain of

agency, rather than focus exclusively on reflection.

12
See Sieck and Arkes (2005).

27
1.5. Is Epistemic Agency a Myth?

In a recent investigation into some of the same philosophers discussed above, Hilary

Kornblith (forthcoming) expresses some skepticism about whether there really is a

legitimate notion of epistemic agency to be found. Concluding his discussion of Sosa’s

take on the connection between reflection and agency, Kornblith writes that “the appeal

to epistemic agency seems to be nothing more than a bit of mythology” (p. 118). About

Moran’s account of rational authority, Kornblith suggests that the considerations

introduced by Moran “do not provide us with convincing reasons to believe that there

really is such a thing as epistemic agency” (p. 126). As made clear above, there are

certainly things to be said about the particular ways in which Sosa and Moran are

characterizing epistemic agency, at least in so far as it pertains to epistemic amelioration.

Still, Kornblith’s skeptical conclusion regarding the very phenomenon of epistemic

agency seems premature.

The reason why this is so has been hinted at several times above, but let us revisit

the relevant issues in the context of Kornblith’s complaints. One of his complaints is that

moving from the unreflective to the reflective level of belief-formation does not introduce

any epistemic agency; in both cases, what beliefs I form is not up to me. Kornblith makes

his point in relation to an imagined scenario in which he is serving on a jury. While

serving on a jury, you may be doing several things, such as focusing your attention on

various pieces of evidence, as well as posing questions about their probity. Focusing your

attention thus, however, is no different as far as agency is concerned from you turning

your head or opening your eyes when forming various perceptual beliefs. Kornblith

writes:

28
 Whether I turn my head is determined by my choice, but once my head is turned
in a certain direction, with my eyes open, and the lighting just so, my perceptual
mechanisms will simply operate in me in ways which have nothing at all to do
with the fact that I am an agent. The fact that I focus my attention, and question
the relevance and probity of the evidence, thus show no more agency when I
reflect than goes on in unreflective cases. Indeed, these activities not only show
no more epistemic agency than goes on in unreflective cases in human beings;
they show no more epistemic agency than goes on in lower animals when they
form perceptual beliefs. But this is just to say that these features of reflectively
formed belief do not exhibit epistemic agency at all (Kornblith forthcoming, p.
117).

The presupposition here seems to be that there is no epistemic agency if there is no

doxastic agency, i.e., agency with respect to what beliefs to form. Only under this

presupposition does it follow from both the reflective and the unreflective case involving

an element of non-agency, in that what beliefs we form in response to the relevant stimuli

is not within our voluntary control, that neither case involves any agency whatsoever.

However, there is a perfectly good sense in which both the reflective and unreflective

scenario may involve a kind of epistemic agency. The best argument to this effect is that

only by acknowledging that this is so can we distinguish clearly between the kind of

ameliorative advice that is bound to be unhelpful, and the kind of advice that may be

effectively prescribed.

Consider the reflective case first. As we have seen, someone reflecting on her

beliefs does not, thereby, gain any control over what beliefs she form as a result of

reflecting. Consequently, any attempt at improving her reasoning by telling her what

beliefs to form or not to form on reflection are bound to be completely unhelpful. That

said, there are many other kinds of advice that she may be given, e.g., with respect to

what kind of evidence-gathering practices she should involve herself in over others,

29
under what conditions she is better off subjecting her beliefs to reflective scrutiny as

opposed to opting for strategies that involve no such scrutiny, and so on. Providing such

advice is worthwhile exactly because what she does on these scores is within her control.

Moreover, in so far as she chooses one option over the other in a manner that can be

properly characterized as directed at the goal of forming true belief, she is expressing her

epistemic agency. And in so far as she makes choices—perhaps, by heeding to sound

ameliorative advice on the issue—that will actually make her better off with respect to

the formation of true belief, she is expressing her epistemic agency wisely.

Next, consider the unreflective case. In particular, consider a medical student in

the process of learning to read an x-ray plate. While no one has direct influence over

what perceptual beliefs they form as a result of observation, the medical student may,

clearly, benefit from being told exactly how and where to look at the plate. Advising her

thus is worthwhile exactly because what she chooses to do here is within her control.

Moreover, in so far as she looks in the right way and on the right locations, and,

eventually, arrives at an accurate diagnosis, she can be said to not only be expressing her

epistemic agency but to be expressing it wisely. And that is so irrespective of whether her

attention at any point in the process of following her teacher’s instructions moves from

the x-ray plate to a second-order reflection on her first-order perceptual beliefs. As

previous sections have made clear, what is crucial as far as the expression of epistemic

agency is concerned is not that the agent in question reflects on her beliefs, but that she is

doing things (reflectively or not) in a manner that can be characterized as directed at the

goal of forming true belief.

30
 Kornblith may, of course, grant this point, while maintaining that this kind of

epistemic agency is not the kind of agency that the philosophers traditionally writing on

the topic are interested in. On this point, consider what Kornblith has to say about the

passage from Moran quoted above (on p. 26), where he elaborates on the relation

between expressing rational agency, on the one hand, and what it means to (as he says)

hand over one’s belief to the authority of reason. According to Kornblith, there is a

tension between these two claims—a tension that was called attention to above in relation

to the relevant quote. According to Kornblith, the tension is such that we have reason to

give up the former claim (about agency), while accepting a qualified version of the latter

claim (about the authority of reason). However, as called attention to above, the tension

only arises if we take the relevant authority to require an ability to opt out of the demands

of rationality and, in effect, exert control (as opposed to mere authority) over what beliefs

we form as a result of deliberating. As far as Moran’s account of authority is concerned,

no such thing is required. The only thing that is required is that we leave it up to our

reflective faculties to lead us toward the proper conclusion, under the assumption that

they are sensitive to the demands that rationality places on us.

At this point, Kornblith may suggest that Moran is no longer talking about agency

in any robust sense of that term, and here me and Kornblith may be in agreement—at

least if the relevantly robust sense of agency is one that can figure constructively in

ameliorative contexts, in ways that I argued above that Moran’s account of authority

cannot. But as we have seen, this does not suggest that there is no such thing as epistemic

agency; at most, it goes to show that there are theories about agency that fail to account

31
for one important role that a concept of epistemic agency should play, i.e., exactly the

role that the present study attempts to delineate.

The same goes for Kornblith’s final complaint about epistemic agency, i.e., that

the idea of epistemic agency is wedded to an overly optimistic view about the epistemic

merits of reflection. Here, we are in agreement about what the relevant empirical research

shows. But as argued above, showing that reflection cannot carry the epistemic weight

that some philosophers (Sosa, supposedly, being one of them) has wanted to put on its

shoulders, neither shows that the decisions that go into when and on what to reflect may

not constitute expressions of epistemic agency, nor suggests that there is no such thing as

agency. If we acknowledge the weak notion of epistemic agency relied upon here, not

even evidence to the effect that reflecting does not generally improve our reliability

suggests that we may not express our agency by reflecting. At most, it suggests that we

are in need of amelioration with respect to an epistemic activity that we perhaps did not

think would require improvement, and that it will serve us well as epistemic agents to

consider alternative ways to express our agency, in the event that the mechanisms

responsible for the epistemic shortcomings of reflection prove recalcitrant enough.

1.6. Conclusion

Where does what we have seen above leave epistemic agency? Exactly where we left it.

Agency still pertains to actions performed in an effort to attain our epistemic goals and, in

particular, the goal of believing truly. It just turned out that (a) epistemic agency cannot

pertain to decisions about what to believe, since making such decisions is impossible; (b)

32
if there is a sense in which what beliefs we end up with as a result of reflection is up to

us, this does not serve to delimit a notion of agency that is relevant to amelioration; and

(c) reflection, due to its largely mixed epistemic track-record, should neither be

considered central to agency, nor, consequently, taken to be essential to the endeavors

associated with ameliorative epistemology.

Keeping these points in mind, we see that there is a sense in which the question of

what actions constitute epistemic agency is almost trivial. After all, what is it to engage in

actions directed at the goal of forming true belief? It is to engage in inquiry. In other

words, the actions by way of which we express our epistemic agency are those we

perform when conducting inquiry. When conducting inquiry, we sometimes reflect on

our first-order beliefs. But we do so much more than that—and rightfully so, given what

we have found about the ways in which reflecting sometimes leads us astray due to a

combination of bias and overconfidence. We also gather information, choose among

different methods of investigation, and so on. And in so doing, we are expressing our

epistemic agency.

At the same time, the very triviality of this answer to the question of what actions

are relevant to epistemic agency suggests that the interesting questions regarding such

agency lie elsewhere. One question pertains to an assumption that was made at the

beginning of this chapter and relied on throughout, namely that believing truly is our only

epistemic goal and, consequently, also unique in being of non-instrumental epistemic

value. This assumption will be elaborated on as well as defended in the next two

chapters. Another question regarding epistemic agency is an ameliorative one, namely:

How exactly are we to bring the ways in which we actually express our epistemic agency

33
in line with the ways in which we should express our epistemic agency, given the goal of

believing truly? This question will be postponed until chapter 5.

34
 CHAPTER 2

I CAN’T MAKE YOU LOVE TRUTH

(AND YOU CAN’T MAKE ME LOVE EVIDENCE)

2.1. Goals and Values

Why take true belief to be epistemically valuable irrespective of whether believing truly

is conducive to something else of epistemic value? In short, why take true belief to be of

non-instrumental epistemic value? One straightforward answer invokes the idea that true

belief is a goal of inquiry or cognition.1 More specifically, the relevant line of reasoning

can be reconstructed as follows:

(1) True belief is a goal of inquiry.

(2) If something is a goal of inquiry, it is of non-instrumental epistemic value.

(3) Hence, true belief is of non-instrumental epistemic value.

Before turning to (1), consider the rationale for (2). It seems reasonable to believe that

goals of inquiry determine what states, processes, practices, and so on, are epistemically

valuable. Not in the sense of what is actually valued—people fail to value what is

valuable for all kinds of psychological reasons of no direct relevance to axiology—but

rather in the sense of what is worthy of pursuit, given participation in practices that can

1
See, e.g., Lynch (2009a), Alston (2005, p. 30), Haack (1993, p. 199), and BonJour
(1985, pp. 83-84) for four representative statements, and David (2001) for further
references.

35
be plausibly characterized as aiming toward the goals in question. More specifically, if

true belief is a goal of inquiry, then true belief is of non-instrumental epistemic value

and everything that is conducive to true belief is (at least) of instrumental epistemic

value, in virtue of being conducive to the attainment of a goal of inquiry.

Someone might deny (2) by maintaining that goals of inquiry are one thing and

epistemic value another, perhaps on account of the latter being a function of the

fulfillment of epistemic obligations rather than of the attainment of goals of inquiry.2 In

other words, someone may argue that epistemic value is to be measured by something

more akin to effort than to actual achievement, perhaps in a manner mirroring the

distinction in ethics between non-consequentialist and consequentialist accounts of

moral rightness.

There might be theoretical room in epistemology for effort-based conceptions of

epistemic value. For present purposes, however, we may disregard such conceptions, the

reason being that we are concerned with amelioration. Since the purpose of amelioration

is to say something constructive about how to improve our epistemic ways in the

contexts where we are failing to attain our epistemic goals, the kind of success we are

concerned with in amelioration is achievement- rather than effort-based success. Effort,

of any kind, is only of interest to the extent that investing great effort in achieving one’s

epistemic goals is correlated with actual achievement.3

2
Feldman (1988) comes closest to defending such a position. As we shall see below
(§2.2 and n. 14), however, he no longer challenges (3) by denying (2), but by denying (1).
3
Interestingly enough, empirical evidence suggests that there is no strong correlation
between a great effort put into living up to the traditional canons of epistemic obligation
on the one hand, and the formation of true belief on the other. In fact, in some contexts,
the correlation is even negative. See Bishop (2000).

36
 Returning to (1), and setting aside the questions of whether (a) true belief is the

only goal of inquiry, and (b) the goal is to be restricted to truths that are in some relevant

sense significant—questions that will be addressed in chapters 3 and 4, respectively—

why accept the claim that true belief is a goal of inquiry? William Alston finds himself

at a loss for words:

I don’t know how to prove that the acquisition, retention, and use of true beliefs
about matters that are of interest and/or importance is the most basic and central
goal of cognition. I don’t know anything that is more obvious from which it could
be derived (Alston 2005, p. 30; emphasis in original).

Many epistemologists side with Alston here.4 For ease of reference, let us refer to any

philosopher who takes true belief to be non-instrumentally valuable as a veritist.5 As we

shall see in §2.2, however, there are epistemologists who do not consider obvious what

Alston does. Richard Feldman, for one, blocks the inference to (3) by denying (1). The

goal of inquiry, he maintains, is to fulfill one’s epistemic obligations by believing in

accordance with one’s evidence, where true or reliably formed belief is neither necessary

nor sufficient for believing thus. What is particularly interesting about his position is that

Feldman, thereby, rejects (3), while steering clear of two problematic claims,

successfully harpooned in the literature.

The first claim is that true belief is of no value whatsoever.6 Feldman is

committed to no such claim. For one thing, he may reject (3) while consistently

maintaining that believing truly may be of instrumental epistemic value, to the extent

4
See references in n. 1.
5
The term is borrowed from Goldman (1999), who introduces the notion of veritism to
capture the particular brand of epistemology that is concerned with the varieties of
processes and practices involved in the production of true belief.
6
Stich (1990) comes closest to defending such a claim.

37
that believing truly is conducive to believing on the basis of sufficient evidence. For

another, he may also maintain that true belief is of instrumental pragmatic value, in

virtue of how accurate plans for action facilitate desire-satisfaction,7 or of instrumental

moral value, given how having an informed citizenry is conducive to the smooth

working of democratic governance,8 to name but two possibilities.

The second claim is that there are no constraints on what one’s epistemic goals

could be, nor, consequently, anything that stops one from taking virtually any good to be

of non-instrumental epistemic value.9 For reasons that will be discussed in §2.3,

Feldman neither makes nor needs to make any such claim, and the arguments that can be

leveraged against the claim in question, while establishing that there are some

substantive constraints on what one’s epistemic goals could be, neither single out true

belief as the only candidate, nor rule out believing upon the basis of sufficient evidence

as a goal of inquiry. Or so it will be argued.

Does it follow that we should give up on the idea that true belief, as opposed to,

believing upon sufficient evidence, is non-instrumentally valuable? No. As we shall see

below, the dialectical situation between the evidentialist and the veritist is symmetrical,

in that neither can impose their respective notion of value on the other—a conclusion

that will be drawn from the failure of several attempts on the part of both he evidentialist

and the veritist to break this symmetry, discussed in §§2.3-5 below. It is concluded in

§2.6 that, while there is no need to reject the evidentialist’s take on value in order to

protect the idea that truth is of non-instrumental value (although not necessarily for

7
See Kornblith (2002).
8
See Lynch (2004, pp. 174-176).
9
See Field (2001) for an endorsement of this claim, and Lynch (2009b) for a critique.

38
everyone) from the evidentialist’s contention that it is not, the prospects for convincing

someone who does not share one’s love for the relevant non-instrumental epistemic

goods to start doing so seem dim indeed.

2.2. Feldman on Epistemic Value

According to Feldman (2002), “epistemological success amounts to having justified

cognitive attitudes,” which, in turn, “amounts to following one’s evidence” (p. 382).

Does this claim conflict with the idea that true belief is of non-instrumental epistemic

value? That depends on how we spell out what it is to follow one’s evidence, which in

turn needs to be analyzed in terms of having evidence. On one understanding of what it

is to have evidence, something is evidence for something else if and only if the former is

as reliable indicator of the latter.10 That, however, is not how Feldman understands what

it is for something to be evidence, the primary reason being that he wants it to be

possible for someone in an evil-demon scenario to be justified and, consequently, also to

have evidence, despite being subject to massively misleading experiences.11 In an evil-

demon scenario, whatever evidence a subject has in virtue of undergoing certain

experiences, these experiences do not reliably indicate anything about the subject’s

surrounding.

What is it, then, about evidence that justifies belief, if not a relation of truth-

indication? According to Feldman—here, writing with his long-time partner in

evidentialism, Earl Conee—evidence justifies in virtue of a certain coherence relation:

10
See, e.g., Goldman (forthcoming).
11
See Feldman (1985).

39
 […] the general idea is that a person has a set of experiences, including perceptual
experiences, memorial experiences, and so on. What is justified for the person
includes propositions that are part of the best explanation of those experiences
available to the person. […] The best available explanation of one’s evidence is a
body of propositions about the world and one’s place in it that makes best sense
of the existence of one’s evidence. This notion of making sense of one’s evidence
can be equally well described as fitting the presence of the evidence into a
coherent view of one’s situation. So it may be helpful to think of our view as a
non-traditional version of coherentism. The coherence that justifies holds among
propositions that assert the existence of the non-doxastic states that constitute
one’s ultimate evidence and propositions that offer an optimal available
explanation of the existence of that evidence (Conee and Feldman, 2008, p. 98).

In other words, a person’s (ultimate) evidence consists in a set of experiences, and sets

of propositions are justified in so far as they provide explanations of those experiences

that cohere with propositions asserting the presence of the experiences in question.

Coherence is a complex notion with a long history, and Feldman and Conee,

unfortunately, do not spell out the details of the particular notion that they have in mind.

However, there is an emerging consensus in the literature that coherence is not truth-

conducive, meaning that more coherence does not imply a higher likelihood of truth,

even ceteris paribus.12 More than that, such a disconnect between coherence and truth

would seem to be exactly what Feldman needs in order to accommodate the idea that

someone in an evil-demon scenario could be fully justified despite undergoing massively

misleading experiences, as long as the propositions believed by the subject in question

provide explanations of her experiences that cohere with the presence of those

experiences.

This is not to subscribe Feldman to a view on which there is no connection

between following one’s evidence and believing truly. For one thing, Feldman takes it

12
See Olsson (2005).

40
that we can make a good case for most justified beliefs being true in the actual world.13

In other words, Feldman’s position is compatible with it being the case that, in a world

like ours, where experiences as a matter of contingent fact tend to provide pretty good

information about the external world, basing one’s beliefs upon one’s evidence is a good

means toward forming true belief. Crucially, however, Feldman denies that this is

something that holds by necessity. Again, in an evil-demon scenario, no degree of

diligence with respect to basing one’s belief upon one’s evidence will enable one to form

true beliefs about the external world, which is why neither true belief nor reliability is

necessary for justification nor, consequently, for epistemic success, according to

Feldman, given his identification of the two.

Notice that a veritist may grant this. Nothing said so far implies that true belief is

the only goal of inquiry. However, Feldman is not only skeptical about the necessity of

truth or truth-conduciveness for epistemic success, but also about it being sufficient.

Registering his skepticism about the idea that mere true belief is an epistemic goal, and

attaining such belief, consequently, constitutes a form of epistemological success,

Feldman poses the following hypothetical:

[…] suppose a person acquires strong evidence against a proposition he has long
defended in public. Out of stubbornness, the person retains the old belief. And
suppose that, contrary to the new evidence, the old belief is in fact true. If the goal
is simply truth, he’s achieved the goal and is, in this case at least, an
epistemological success. But, as Locke said of a person who does not reason as he
ought, “however he sometimes lights on truth, is in the right but by chance; and I
know not whether the luckiness of the accident will excuse the irregularity of his
proceeding.” […] Setting aside questions of excuses, the idea here seems to be
that the person who reasons badly and stumbles onto a truth is not believing as he
ought, is not achieving epistemological success (Feldman 2002, p. 378).

13
See Feldman (2003, pp. 615-616).

41
Elaborating on a related scenario a couple of sentences further down, Feldman makes it

clear that he sides with Locke on this point. To simply achieve true belief is not to

achieve any goal of ours, nor to be any kind of epistemological success:

Imagine a person who makes an unreasonable and unreliable inference that

happens to lead to a true belief on a particular occasion. It might be fortunate that
he’s got this true belief, but I see nothing epistemologically meritorious about it
(Feldman 2002, p. 379).

Clearly, the idea that there is nothing epistemically meritorious about (mere) true

belief—that is, that believing truly is neither a goal nor one kind of epistemic success—

runs contrary to any account of epistemic value relying on argument (1)-(3) above.14

Granted, few veritists would deny that there is something epistemically lacking about the

person in the scenario Feldman imagines. For example, reliabilists about justification

would deny that she is justified. However, what is at issue here is whether there is

anything of epistemic value about her epistemic situation, and that is where the veritist

parts company with Feldman. Is Feldman in any relevant sense, thereby, mistaken about

epistemic value? The next section considers two arguments—both of which are,

ultimately, rejected—to the effect that he is.

14
Feldman (in correspondence) confirms that there has been a shift in his views on
epistemic value, from his (1988), where he accepts that true belief is a goal of inquiry and
simply denies that true belief has anything to do with value, to his (2002), where he (as
we have seen) denies that true belief has anything to do with value by denying that it is a
goal of inquiry.

42
2.3. Is Feldman Mistaken About Value?

Recently, Michael Lynch (2009a and b) has provided a series of arguments against what

he calls epistemic expressivism, or the view that there are no constraints on what one’s

epistemic goals could be, nor, consequently, on what one may take to be of non-

instrumental epistemic value—a position that he attributes to Hartry Field (2001).

Presently, we will not be concerned with epistemic expressivism. However, as we shall

see, at least two of Lynch’s arguments (if successful) apply equally well to someone

who, like Feldman, simply denies that truth is of non-instrumental epistemic value.

First, consider what we may refer to as the no inquiry argument. The idea is that

someone who is forming belief in a manner that involves no commitment to the idea that

propositions are to be embraced if and only if they are true is not engaging in inquiry.15

Lynch (2009a) makes this point in relation to a hypothetical example including

someone—let us call him Steve—who is committed to the goal of “accepting” all and

only propositions that flatter his cultural origin. Is Steve, thereby, engaging in inquiry?

Lynch suggests that the answer has to be “no,” since the practice Steve is involved in

does not involve a commitment to the idea that propositions are to be embraced if and

only if they are true. Whatever practice Steve is involved in, it is not inquiry.

When applied to Steve, this line of reasoning is highly convincing. But does the

point in question generalize to any practice that does not involve a commitment to the

idea that propositions are to be embraced if and only if they are true, including a practice

15
The norm of embracing propositions if and only if they are true is, of course, not meant
to be action-guiding. Instead, the norm is supposed to tell us something about conditions
under which particular instances of inquiry have to be considered successful or
unsuccessful, namely to the extent that they terminate in true belief or false belief,
respectively.

43
that involves (nothing but) a commitment to the idea that propositions are to be

embraced if and only if they are based upon sufficient evidence, as spelled out by

Feldman? It is not so clear that it does. To see why, consider two ways in which the

relevant point about what constitutes inquiry could be pushed. On the one hand, it could

be maintained that the norm according to which we should believe something if and only

if it is true attaches to the relevant notion of inquiry by conceptual necessity. In that

case, the dispute seems to boil down to one about meanings, and the veritist would have

to resort to either suggesting that Feldman is conceptually confused—a claim that is

highly implausible in light of him being one of contemporary epistemology’s most

prominent contributors—or granting that the evidentialist is using a different yet equally

legitimate notion of inquiry.

On the other hand, Lynch makes it clear that this is not how he views matters:

[…] it isn’t at all obvious that the above arguments need to be understood as
being about the meanings of “inquiry” or “belief” or “language”. As we have
noted earlier, (TN) [i.e., the idea that it is prima facie correct to believe <p> if and
only if <p> is true] and what follows from it needn’t be thought of as conceptual
truths. One could take such points to be about inquiry and belief rather than
“inquiry” and “belief”. Cooking aims to produce food; comedy aims to produce
laughter. Likewise, one might think, inquiry aims to produce true beliefs. It is a
fact about such practices that they cannot be separated from their aims; their aims
are partly constitutive of them (Lynch 2009b, pp. 90-91).

In other words, rather than pertaining to conceptual necessities, the relevant truths about

inquiry may, as Lynch suggests, be truths of constitution. However, at this point, the

argument starts to look question begging, at least if leveraged against the evidentialist.

More specifically, the argument against counting above evidentialist practice as a form

of inquiry would, under this construal, require a premise to the effect that inquiry is the

44
pursuit of truth, which is exactly the claim that the evidentialist contests. Clearly, the

evidentialist has no reason to grant the veritist that premise.16

Lynch seems to admit this point when noting that he “cannot give a non-circular

justification of my belief that it is valuable to engage in inquiry [as construed by the

veritist]; for in answering the question I am already committed to the value of the very

practice in question” (2009a, p. 239). Moreover, he grants that, far-fetched examples like

the one with Steve aside, it seems at least prima facie reasonable to say that someone

may engage in inquiry in a context where the latter is spelled out in terms of the goal of

having a coherent belief system, or having epistemically rational or justified beliefs. At

this point, however, the second argument enters, an argument that we may refer to as the

presupposition argument. Lynch writes:

No doubt, people can and do have such goals in pursuing inquiry. The question is
whether […] we can conceive of someone who has distinct epistemic goals that
don’t include the goal of having true belief. Of course, the typical way of
understanding inquiry does require that it have only one goal: truth […]. The
thought here is that the value of achieving, e.g. a coherent belief system lies in the
fact that doing so is instrumental to believing what is true and not believing the
false. […] If we didn’t think this, it seems unlikely that we would care about
whether our beliefs were coherent. But of course if this is how we see the matter,
having the goal of coherence or justification presupposes having the goal of true
belief (Lynch 2009b, p. 89).

The problem is, of course, that this is not how the evidentialist, at least of Feldman’s

stripe, sees the matter. As we saw above, he denies exactly that the value of justification,

16
The same points made here about the futility of discrediting the evidentialist’s account
of value with reference to a constitutive or conceptual connection between truth and
inquiry can be made, mutatis mutandis, against any analogous attempt in terms of there
being a constitutive or conceptual connection between truth and belief, on account of the
former being the norm of the latter, and inquiry being the practice of belief-formation.
See, e.g., Lynch (2009a).

45
i.e., of basing your beliefs upon sufficient evidence, is parasitic upon the value of

believing truly. Suggesting that the evidentialist’s axiology presupposes such a

relationship would be to misconstrue it, at best, and to beg the question, at worst. As

such, we may conclude that neither of Lynch’s arguments establishes that Feldman is

mistaken about epistemic value.

It is crucial to note, however, that the dialectical situation is symmetrical, in that

any evidentialist elaboration on the notion of inquiry is equally unlikely to sway the

veritist, as above reflections on the veritistic notion are to sway the evidentialist—and

rightly so. As such, nothing said so far goes to show that the veritist should stop valuing

true belief non-instrumentally. The point is merely that neither the veritist nor the

evidentialist can impose their respective notion of value on the other, as long as the

relevant values are grounded in nothing but different conceptions of inquiry. The next

section considers an argument on part of the veritist that, if successful, would break the

symmetry, by grounding the value of true belief in the fact that certain true beliefs are

necessary parts of a flourishing life. If successful, the argument would provide reasons

for the evidentialist to start valuing (at least some) true beliefs non-instrumentally,

irrespective of what happens to be her conception of inquiry.

2.4. True Belief and Human Flourishing

According to Lynch (2004), we care about having true beliefs, where “caring for

something entails that we treat it as [an] end” (p. 119). Differently put, we take true

belief to be of non-instrumental epistemic value. But what reasons do we—as in: all of

46
us, irrespective of our particular notions of inquiry—have for valuing truth thus? Lynch

provides two arguments meant to answer this question, both of which elaborate on the

intimate relationship between believing truly and being happy, in the sense of leading a

flourishing life.17 He refers to the first argument as the argument from self-knowledge.

Let us consider Lynch’s formulation of the argument, before turning to his definitions of

the notions invoked (and I quote):

(1) Self-respect and authenticity require a sense of self.

(2) Because it is required for self-respect and authenticity, which are part of
happiness, having a sense of self is an important part of happiness.
(3) Having a sense of self means having true beliefs about what you care about.
(4) Therefore, having some true beliefs—about what you care about—is also
part of happiness, other things being equal (Lynch 2004, p. 127).

“If we grant,” Lynch continues, ”that whatever is part of happiness is worth caring about

for its own sake, from these premises we can deduce that the truth about what you care

about is itself worth caring about for its own sake” (p. 127). In other words, Lynch wants

to infer the conclusion

(5) the truth about what you care about is worth caring about for its own sake,

from (1)-(4), together with the following additional premise:

(AP) If x is part of some y (e.g., happiness) that is worth caring about for its

own sake, then x is worth caring about for its own sake.

17
Actually, Lynch provides six arguments. However, three of these pertain to
instrumental values (see n. 8), and the fourth fails for the same reason as the first and
second (see n. 20), which leaves the two arguments to be discussed here.

47
Let us consider the premises in turn. To have self-respect, as Lynch understands it, is to

have a sense of your own value (p. 124), while authenticity is to identify with the desires

that effectively guide your actions (p. 125). To have a sense of self is to have true beliefs

about what you care about, as per premise (3). Since it is, arguably, impossible to either

have a sense of your own value, or to identify with the desires that effectively guide your

action, without having true beliefs about what you care about, self-respect and

authenticity require a sense of self. Hence, premise (1).

Premise (2) maintains that self-respect and authenticity are parts of happiness.

What does it mean to be a part of something here? Lynch explains:

Being a part of something is crucially different from being a means to it. A means
to an end is ipso facto different from the end itself. A necessary part of something,
on the other hand, helps make the whole of which it is a part the thing that it is.
Change or destroy the part and you change or destroy the whole (p. 128; emphasis
in original).

In other words, for some x to be a part of some y is for x to be a necessary part of y, such

that the absence of x implies the absence of y. Interpreting premise (2) accordingly, and

granting premise (3)—i.e., that having a sense of self means having true beliefs about

what you care about—it follows that having true beliefs about what you care about is a

necessary part of happiness. If that is how we are to understand what it is to be a part of

something, (4) can be reformulated as follows:

(4*) Having true beliefs about what you care about is a necessary part of

happiness, other things being equal.

48
It might seem strange to talk about something being a necessary part of something, other

things being equal. After all, to be a necessary part of something is to be a part of

something by necessity, i.e., irrespective of whether other things are equal. However,

what Lynch means to indicate by the qualification in terms of “other things being equal”

is that the value in question is pro tanto, in that it may be overruled by other

considerations and, hence, is not an all-things-considered good.18 In other words, the

conclusion that we are to draw from Lynch’s argument is the following:

(5*) The truth about what you care about is worth caring about for its own sake,

in the sense that having the true beliefs in question is pro tanto good in

itself.

If Lynch’s argument succeeds in establishing (5*), then we have found an argument to

the effect that at least some true beliefs are (pro tanto) valuable in themselves,

irrespective of what happens to be one’s particular notion of inquiry. The problem with

the argument, however, is (AP). We may see more clearly what the problem is by

reformulating (AP) in light of Lynch’s definition of what it is for something to be a part

of something as follows:

(AP*) If x is a necessary part of some y (e.g., happiness) that is worth caring

about for its own sake—i.e., if x is such that its absences implies the

18
See Lynch (2004, pp. 46-51). This interpretation is confirmed by Lynch (in
correspondence).

49
 absences of that of which it is a part—then x is worth caring about for its

own sake.

Clearly, (AP*) is not true. It does not follow from x being a necessary part of y, that if y

is worth caring about for its own sake then the same goes for x. By way of illustration,

assume that true belief is worth caring about for its own sake. If (AP*) were true, it

would follow that mere belief, being a necessary part of true belief, would be worth

caring about for its own sake. This clearly cannot be right, even if we qualify the good

involved as being a mere pro tanto good. Hence, (AP*) cannot be true. And if (AP*) is

not true, then we do not get the conclusion that Lynch wants to draw from (1)-(4),

namely (5*).19

Lynch’s second argument, or what he refers to as the argument from integrity,

suffers from an analogous problem. This argument attempts to establish the more general

conclusion that the very act of caring about truth as such is non-instrumentally valuable.

The reason, Lynch argues, is that caring about truth as such is essential to intellectual

integrity. To instantiate intellectual integrity

19
Notice that we can run the same argument even if we interpret the relevant relation
between true belief and happiness in terms of supervenience. After all, it does not follow
from y being valuable in itself that, if x is part of the supervenience base of y, then x is
also valuable in itself. By way of illustration, assume (if only for the sake of the
argument) that being a person is valuable in itself. Assume, furthermore, that facts about
person-hood supervene on neurological facts, and that I am the person I am in virtue of
instantiating the particular neurological facts N1-N850. If you change any of those facts,
you change (if not destroy) the whole, i.e., me. (You might still have a person, of course,
but arguably not a person that is me.) Then, take any neurological fact in this series, say,
N46. Is instantiating N46 valuable in itself? It is not—despite it being part of the
supervenience base of something that (by hypothesis) is valuable in itself. In other words,
whatever value attaches to the supervenient does not trickle down to the individual parts
that make up the subvenient, in the manner that Lynch’s argument requires.

50
 […] one must be open to having true belief in general, and pursuing truth in
general, on those questions that come before you, whatever those may be. Caring
about truth for its own sake, that is, being willing to both pursue it and be open to
it, is therefore a necessary condition for intellectual integrity (p. 133).

According to Lynch, intellectual integrity is what he refers to as a constitutive good, or

something worth caring about for its own sake. We will return to this claim

momentarily, but let us first consider the argument that he sets out (and I quote):

(1) If intellectual integrity is a constitutive good, then so is caring about truth as

such.
(2) Intellectual integrity is a constitutive good.
(3) Therefore, caring about the truth as such is a constitutive good.
(4) If caring about truth as such is a constitutive good, then truth as such is worth
caring about for its own sake.
(5) Therefore, truth as such is worth caring about for its own sake (Lynch 2004,
p. 136).

This is a valid argument, but it is not sound. Consider the conditional in premise (1).

Why would it follow from intellectual integrity being a constitutive good, that caring

about truth as such is also a constitutive good? Lynch suggests that “[c]aring about the

truth as such is a constitutive good because it is essential to intellectual integrity, and

therefore, to integrity proper” (p. 136). But, as already noted above in relation to (AP*),

it does not follow from x being a necessary condition for y, that, if y is worth caring

about for its own sake, then the same goes for x. Consequently, we cannot infer (3) from

(1) and (2), nor (5) from (3) and (4).20

Two things need to be noted about what should not be inferred from the failure of

Lynch’s arguments. First, while the arguments in question failed to establish that some

20
The same goes for Lynch’s (2004, p. 157) structurally identical argument about
sincerity.

51
true beliefs are non-instrumentally valuable (pro tanto or otherwise) irrespective of your

particular conception of inquiry, nothing said so far rules out the possibility that having

certain true beliefs (e.g., about what you care about), or perhaps even caring about truth

as such, is under many circumstances conducive to leading a flourishing life. However,

that at most establishes that some instances of true belief are instrumentally valuable, not

that they are non-instrumentally valuable. As such, it provides no reason for those who

do not already value (any) true belief non-instrumentally to start doing so.

Second, in accordance with what was noticed above, the failure of Lynch’s

arguments does not go to show that those who already value true belief non-

instrumentally should stop doing so. But in want of an account of the non-instrumental

value of true belief that appeals to something beyond conceptions of inquiry that are not

universally shared, the veritist must accept the fact that she can provide no reasons for

those who do not already share the relevant notion of inquiry to start valuing true belief

non-instrumentally. The next section considers the matter from the other side, by

investigating the possibility that the evidentialist can provide the kind of argument that

Lynch could not, i.e., an argument to the effect that we should value basing our beliefs

upon sufficient evidence non-instrumentally, irrespective of what happens to be our

particular conception of inquiry.

2.5. Epistemic Value and the Ethics of Belief

In his essay “The Ethics of Belief” (1866), W. K. Clifford famously argues that it is

wrong to believe upon insufficient evidence. Clifford argues by way of examples in

52
which beliefs formed on the basis of insufficient evidence have dire consequences for

others, but suggests that believing thus would have been wrong even if the consequences

in question had not ensued. Moreover, Clifford takes it that the duty not to believe upon

insufficient evidence extends beyond those engaged in decisions involving high stakes.

Indeed, the duty is universal. As he famously sums up his position on the issue: “It is

wrong, always, and everywhere, and for everyone, to believe anything upon insufficient

evidence” (p. 346).

For Clifford, this synchronic duty not to believe upon insufficient evidence

follows from a more fundamental, diachronic duty of inquiry, i.e., “the universal duty of

questioning all that we believe” (p. 343). In questioning our beliefs, we attempt to find

out whether or not they are based on sufficient evidence. If they are, we may continue

holding them, knowing that they have been “fairly earned by investigation” (ibid.). If

they are not, “we have to begin again at the beginning, and try to learn what the thing

[under investigation] is and how it is to be dealt with—if indeed anything can be learnt

about it” (ibid.). In other words, through persistent inquiry, we either uncover sufficient

evidence one way or the other and adjust our belief accordingly, or find that there is no

evidence either way and that suspension of belief, consequently, is the most appropriate

attitude to take toward the relevant proposition.

It is as a fruit of such “scrupulous care and self-control” (p. 347) that we, on

Clifford’s picture, should understand the value of believing upon sufficient evidence. To

believe thus is to have satisfied the duty of inquiry by making sure that, for every

proposition under consideration, that proposition is only believed if it is based on

sufficient evidence. As suggested by the case of belief suspension in situations where it

53
turns out that nothing can be learnt about the object under investigation, believing upon

sufficient evidence may not be the only (albeit, possibly, the only happy) way of

satisfying the duty of inquiry.21 Still, if Clifford is right, even those who do not happen

to share Feldman’s conception of inquiry ought to value having their beliefs be based

upon sufficient evidence non-instrumentally. To see why, consider the fact that Clifford,

unlike the great majority of contemporary epistemologists, seems to take our epistemic

duties to be a mere subset of our ethical duties.22 With this in mind, consider the

following argument:

(1) We have a duty to only believe upon sufficient evidence, and the duty is an

ethical one.

(2) To believe propositions upon sufficient evidence is to satisfy the duty to

only believe upon sufficient evidence with respect to the relevant

propositions.

(3) Satisfying an ethical duty is non-instrumentally valuable.

(4) Hence, to believe upon sufficient evidence is non-instrumentally valuable.

(5) The relevant ethical duty applies universally.

(6) To believe upon sufficient evidence is non-instrumentally valuable for

everyone.

21
However, given Clifford’s emphasis on our diachronic duty to question all that we
believe and, in effect, inquire as far as possible, it is unlikely that he would consider
complete suspension of belief as a acceptable way to satisfy the duty of inquiry, unless
such suspension constitutes the inescapable end point of pain-staking investigation.
22
See Haack (2001) for a convincing set of arguments to this effect.

54
Premise (1) captures the duty Clifford infers from the duty of inquiry, and identifies it as

an ethical one. Premise (2) corresponds to aforementioned observation about how we

may satisfy the duty in question by believing upon sufficient evidence. Premise (3)

amounts to an assumption that will be granted for the sake of the argument, i.e., that

satisfying our ethical duties—whatever they are—is non-instrumentally valuable.

Moreover, if all epistemic duties are ethical duties, the same applies to the former, as per

premise (4). Assuming that the relevant ethical duties apply universally, as per premise

(5), believing upon sufficient evidence is non-instrumentally valuable not only for those

who, like Feldman, share the evidentialist conception of epistemic success, but to

everyone, as per (6).

Albeit a valid argument, premise (1) is highly implausible. To see why, consider

that there, clearly, are cases in which one is not morally reproachable for believing upon

insufficient evidence, as famously pointed out by William James (1907) in his critique of

Clifford. Clifford denied this possibility, by suggesting that no belief is (morally)

inconsequential.23 In so doing, Clifford assumed that mere potential harm, however

remote, is sufficient for moral culpability—a position that, as Susan Haack (2001) points

out, would imply that “not only drunk driving, but owning a car would be morally

culpable” (p. 27). Moreover, Clifford maintains that believing upon insufficient evidence

invariably strengthens the habit of credulity, in turn amounting to a risk for harm.24

Granted, belief upon insufficient evidence is sometimes the result of cognitive

tendencies that may lead to harm. But in maintaining that believing upon insufficient

23
See Clifford (1866, p. 342).
24
See Clifford (1866, pp. 344-346).

55
evidence invariably leads to credulity, Clifford seems to assume that such belief always

is the result of potentially harmful cognitive traits, and that is highly questionable.

Consequently, it would seem that Clifford has to grant James’ point that there are

cases in which believing upon insufficient evidence is morally permissible. In fact,

contrary to Clifford’s exaggerated sense of potential harm, a great many beliefs,

including some held and debated within science and philosophy, are of no great moral

consequence. If Clifford were to acknowledge this, and at the same time hold on to (1),

he would have to maintain that the cases in question fall outside the scope not only of

moral evaluation but (consequently) also of epistemic evaluation, the latter being a

special case of the former. From this, it would follow that none of the beliefs in question

are held in ways that are epistemically impermissible. Such epistemic laissez-faire seems

undesirable, to say the least, and the way to block the conclusion is, of course, to deny

(1), and maintain that, in so far as we have a duty to only believe upon sufficient

evidence, that duty is not an ethical one.

In light of this, it should come as no surprise that contemporary epistemologists

sympathetic to the idea of an ethic of belief, nevertheless, tend to make a distinction

between ethical and epistemic duties. For example, in spelling out his evidentialist take

on Clifford’s position, Feldman (1988, pp. 236-237) explicitly rejects the identification

of the domains of epistemic and moral obligations, and maintains that beliefs with

morally bad consequences may still be epistemically obligatory. Similarly, when

claiming that every person is required to try her best to bring it about that, for every

proposition considered, she accepts the proposition only if it is true, Roderick Chisholm

56
(1977, p. 14) specifies the requirement as a purely intellectual requirement.25 In other

words, the main modern defenders of Clifford’s project construes it not so much an

ethics of belief as simply an epistemology of belief formation.26

In so doing, however, the neo-Cliffordians also surrender the one aspect of

Clifford’s theory that would provide them with a way to convince the veritist of the non-

instrumental value of basing one’s belief upon sufficient evidence. As argument (1)-(6)

above makes clear, the reason Clifford’s account was promising on this score was

exactly because it took our duty to base our beliefs thus to be an ethical duty—a fact

that, given the wide scope and non-instrumental nature of our ethical duties, would

provide a lever for extending the normative force of the evidentialists’ emphasis on the

value of believing upon sufficient evidence beyond those who happen to share their

particular conception of inquiry. In so far as this ambition is not present in modern

defenders of an ethics of belief, they may thereby be constructing far more plausible

accounts of the relevant evidentialist norms than Clifford did, albeit at the expense of

providing little by way of arguments for convincing the veritist of the non-instrumental

value of believing upon the basis of sufficient evidence.

25
Chisholm does at times write as if epistemic duties are a mere subspecies of ethical
duties, e.g., in his (1991, p. 119). However, when spelled out, the claim turns out to be
that (a) ethical duties are duties that are not overridden by any other requirement, and (b)
non-overridden epistemic duties (like any non-overridden duties), hence, are ethical
duties (see pp. 127-128). Granting this somewhat questionable definition of an ethical
duty, it still only goes to show that some epistemic duties are ethical duties, not that all
are, nor consequently that the former are a mere subspecies of the latter.
26
Wolterstorff (2005) is a possible exception.

57
2.6. Conclusion

The failure of surveyed accounts to break the dialectical symmetry between the veritist

and the evidentialist does not prove that the symmetry cannot be broken. Still, the

absence of further, more successful accounts leaves us with no reason to think that the

requisite arguments are forthcoming. This is not to deny that we all have ample

instrumental reason to value true belief, given the ways in which believing truly may

further our pragmatic, moral or prudential endeavors. Furthermore, and as has been

stressed several times already, what has been argued here is neither to deny that many of

us do value at least some true beliefs irrespective of whether doing so is conducive to

something else of epistemic value, nor to maintain that those of us who do should stop

doing so. As we have seen, there is a perfectly good answer to the question of why true

belief is of non-instrumental epistemic valuable, namely the one provided in terms of

true belief being a goal of inquiry. It just happens to be an answer that cannot be invoked

in an argument to the effect that those who do not share the relevant notion of inquiry

should start valuing true belief non-instrumentally.

This fact, taken together with the failure of surveyed accounts to ground the

relevant claims about value in something other than our notions of inquiry, suggests that

the prospects for convincing someone who does not share one’s love for the relevant

non-instrumental, epistemic goods to start doing so seem dim indeed. However, given

that failure to convince someone to value something by no means is a sufficient reason

for ceasing to value that thing, there is no need to reject the evidentialist’s take on value

in order to protect the idea that truth is of non-instrumental value from the evidentialist’s

contention that it is not. Granted, I can’t make you love truth, if you don’t already do.

58
But by aforementioned symmetry you also can’t make me love evidence. For this reason,

subsequent chapters will assume the position taken by most epistemologists, to the effect

that true belief is, indeed, a non-instrumental, epistemic good—albeit, as we have seen,

not necessarily for everyone.

59
 CHAPTER 3

IN DEFENSE OF EPISTEMIC VALUE MONISM

3.1. Varieties of Value

As noted in the previous chapter, most epistemologists take believing truly to be of non-

instrumental epistemic value in virtue of true belief being a goal of inquiry. It is far more

controversial, however, whether the following thesis holds—a thesis that we may refer to

as epistemic value monism, or EVM for short:

(EVM) Believing truly is unique in possessing fundamental, non-instrumental

epistemic value.1

The present chapter defends EVM against a series of objections coming out of recent

discussions about the value of knowledge. There is no doubt that these discussions have

been extremely rewarding in bringing about something of an axiological renaissance in

epistemology. Still, it is the thesis of the present chapter that recent pluralistic critiques of

EVM fail to provide any convincing counterexamples, and that epistemic value monism

remains a perfectly defensible thesis.

Before proceeding to EVM’s defense, let us clarify its component notions. First,

what is epistemic value? As noted in the previous chapter, it seems reasonable to believe

that (a) goals of inquiry determine what activities, states, processes, practices, and so on,

1
Recent proponents of EVM include Goldman (1999) and David (2005).

60
are epistemically valuable, (b) true belief is one such goal, (c) true belief, hence, is of

non-instrumental epistemic value, and (d) everything that is an effective means to

believing truly is of instrumental epistemic value. The distinction between instrumental

and non-instrumental value pertains to how value may be inherited across causal

relations. Another way in which value may be inherited is from parts to wholes. As we

shall see in §3.2, that some states inherit value from one of their components is important

to keep in mind when thinking about the comparative value of complex states. With this

in mind, let us say that something is of fundamental value if and only if its value does not

derive from the value of any of its components.

By way of illustration, EVM is compatible with knowing being of non-

instrumental epistemic value in virtue of believing truly being a component of knowing.2

What is denied is simply that such a state—or, more generally, any state that involves

true belief as a mere component—is of fundamental, as opposed to derived, non-

instrumental value. The notions of fundamental versus derived value may be applied also

to instrumental values. However, it will suffice to make the distinction in relation to non-

instrumental value for the purposes of the present investigation. Consequently, any

unqualified use of the terms ‘fundamental’ or ‘derived’ shall, henceforth, be understood

in terms of fundamental or derived non-instrumental value.3

2
The relevant states do not literally have components. The components in question
correspond to the conditions included in a correct analysis of the state(s) in question. If a
state has no analysis—as Williamson (2000) has argued is the case for knowledge—it has
no components, but may still be of fundamental value.
3
The taxonomy developed here is not meant to cover all possible mechanisms for value
inheritance. For example, Goldman (Goldman and Olsson, 2009) argues that there are
instances in which the non-instrumental value of true belief may be inherited by token
processes in virtue of being instances of reliable types—a pattern of inheritance that can
be categorized neither as an instance of instrumental value nor as one of derived value.

61
 It is controversial whether all bearers of non-instrumental value are intrinsically

valuable, or valuable in themselves. For present purposes, two issues need to be

addressed. The first issue is whether to understand intrinsic value as applying to objects

or to states of affairs. If the former, it is easy to come up with examples involving objects

identical in intrinsic properties, yet (intuitively) differing in intrinsic value.4 However, if

we take intrinsic value to apply to states of affairs, as is common in the Moorean

tradition, such examples can be accounted for, and the claim that non-instrumental and

intrinsic values coincide be maintained.5 It is easy to see why one might want to opt for

an object-construal of intrinsic value when it comes to a bearer of value that has figured

prominently in the Kantian tradition, viz. the person. However, as far as epistemic value

is concerned, it is reasonable to think that the bearers of value are states of affairs—be it

states of affairs involving doxastic states, processes, practices, or what have you—rather

than objects. One strong indication of this is that it seems odd to say that a true belief (an

object) is valuable, unless this is just a different way of saying that someone believing

truly (a state of affairs) is valuable, and similarly for someone knowing, understanding,

being justified, and so on.6 Construed thus, the notion of non-instrumental value, as

understood here, may be taken to coincide with intrinsic value.

That said, a lack of taxonomical exhaustiveness would only be a problem if the taxonomy
used somehow stacked the dialectical deck against the pluralist, which it does not. If
anything, by not incorporating the kind of value that Goldman introduces and argues can
be invoked in a solution to problems typically leveraged against monists, it might even be
that the present taxonomy stacks the deck against the monist—which, of course, makes
any defense of monism in terms of this taxonomy all the more significant.
4
See, e.g., Rabinowicz and Rønnow-Rasmussen (2000).
5
See Bradley (2006).
6
Similarly, in so far as references will be made below to “true belief” or “knowledge”
being epistemically valuable, this is not to suggest that abstract objects are proper bearers

62
 Still, the discussions below will be framed in terms of non-instrumental rather

than intrinsic values. The reason for this taxonomical decision brings us to the second

issue, which is whether intrinsic value is to be identified with final value, i.e., the kind of

value that pertains to that which is valuable irrespective of any considerations about

conduciveness whatsoever. By way of illustration, consider the claim that true belief is

non-instrumentally valuable in virtue of being a goal of inquiry, which, as argued above,

implies that true belief is of non-instrumental epistemic value. Maintaining that true

belief is valuable thus is perfectly compatible with also claiming that believing truly only

is of non-instrumental epistemic value in so far as the beliefs at issue are of instrumental

non-epistemic value (e.g., practical, prudential or moral), perhaps because they pertain to

questions posed by some relevant set of inquirers.7 In other words, true belief being of

non-instrumental epistemic value does not imply that true belief is valuable thus

independently of any considerations about conduciveness whatsoever.

Differently put, unlike bearers of final value, not all bearers of non-instrumental

value are non-instrumentally valuable simpliciter or without qualification (such as

“epistemic,” “moral,” or “practical”). Consequently, if intrinsic value is to be identified

with final value, then the intrinsically valuable is not to be identified with the non-

instrumentally valuable, as understood here. As already noted, being non-instrumentally

valuable on one qualification does not imply being non-instrumentally valuable without

qualification. As will be explored further in chapter 4, this is important to note not only

because it shows that some non-instrumental values are not final values, but also because

of such value, but simply short-hand for saying that (at least some) instances of people
believing truly or knowing things are epistemically valuable.
7
See, e.g., Goldman (1999, pp. 94-96).

63
it suggest that it does not follow from believing truly being of non-instrumental epistemic

value that all instances of true belief are of such value. Again, perhaps only those that

pertain to questions posed by some relevant set of inquirers are valuable thus. But more

on this in the next chapter.

For now, note that it not only does not follow from something being valuable on

one qualification that it is valuable without qualification; it also does not follow from

something being non-instrumentally valuable without qualification that it is non-

instrumentally valuable on every qualification. In other words, something may be

valuable simpliciter without being of non-instrumental epistemic value. By way of

illustration, assume that undergoing an episode of pleasure is non-instrumentally valuable

simpliciter. Even so, it does not follow that pleasure is of non-instrumental epistemic

value, for the simple reason that pleasure is not an epistemic value in the first place. We

will return to this point in §3.5 below.

To sum up, what follows from the above discussion is the following:

(1) Some instances of true belief are of fundamental, non-instrumental epistemic

value.

Note that it does not follow from (1) that all true beliefs that are non-instrumentally

valuable are of equal non-instrumental value; some truths might very well be of greater

non-instrumental value than others. Furthermore, that some true beliefs are of

fundamental, non-instrumental epistemic value, clearly, does not imply that they are

64
unique in this respect—and this is a problem, in so far as we want to get from (1) to

EVM. To do so, we need to also make plausible the following:

(2) Nothing is of non-instrumental epistemic value, unless it is so in virtue of the

non-instrumental epistemic value of believing truly.

If some instances of true belief are of fundamental, non-instrumental epistemic value, and

nothing is of non-instrumental epistemic value unless it is so in virtue of the non-

instrumental epistemic value of believing truly, then believing truly is unique in

possessing fundamental, non-instrumental epistemic value—which, of course, is exactly

what EVM states.

While (1) was motivated in the previous chapter with reference to how true belief

is at least one goal of inquiry, defending (2) from potential objections will take further

axiological legwork. To be exact, it will take the remainder of this chapter. It should be

noted right at the outset, however, that there is a presumption in favor of conjoining (1)

with (2), rather than conjoining (1) with a claim leaving room for a multitude of

fundamental, non-instrumental, epistemic values. The presumption can be motivated with

reference to a principle of axiological parsimony with respect to the number of kinds of

bearers of fundamental value that is postulated. This principle is, in turn, an instance of

the more general thesis that we should prefer ontologies with fewer rather than more

existential commitments, ceteris paribus.8 More specifically, the presumption is in favor

8
Invoking this principle does not commit us to taking simplicity to be of fundamental,
non-instrumental epistemic value, as opposed to of instrumental epistemic value (e.g.,
pursuing simplicity in theorizing is conducive to forming true belief), or non-epistemic

65
of explaining all judgments regarding absolute as well as comparative epistemic value by

assuming one as opposed to several bearers of fundamental value. A pluralist might very

well accept that there is such a presumption. What she would argue, however, is that this

presumption is defeated by evidence to the effect that we cannot account fully for the

domain of epistemic value in monistic terms, which is exactly the kind of charge that we

will be concerned with below. Still, the presumption provides us with reason to prefer

conjoining (1) with (2) in the absence of any convincing evidence for such an inability,

and it is the burden of the remaining sections to show that there is no such evidence.

What constitutes relevant evidence here? (2) is equivalent to the universal claim

that, for every x, if x is of non-instrumental epistemic value, then it is so in virtue of the

value of believing truly. Given the obvious difficulty of defending a universal claim, I

will restrict my attention to what, judging by recent work in epistemological axiology, are

the most prominent counterexamples to (2). The counterexamples in question are

formulated in terms of three candidates for fundamental, non-instrumental epistemic

values, i.e., knowing, being justified, and understanding, to be discussed in §§3.2-3, 3.4,

and 3.5, respectively. As we shall see, these candidates are typically taken to be backed

up by strong and widespread axiological intuitions. This suggests that, if any candidate

constitutes a challenge to (2), it will be found among these three. Consequently, if it can

be shown that none of them poses such a challenge, we may conclude that there is no

convincing evidence against (2), and that the presumption in favor of EVM still stands.

value (e.g., simplicity brings tractability, which is pragmatically valuable). However, see
Sober (2001) for a skeptical take on the possibility of understanding the value of
simplicity in terms of other values.

66
3.2. On the (Alleged) Surplus Value of Knowledge

What reason do we have for thinking that knowing is non-instrumentally valuable?9 One

reason is that knowledge is factive.10 Given that some instance of true belief are non-

instrumentally valuable, and knowledge implies truth, some instances of knowing may,

thereby, be of derived non-instrumental value in virtue of being factive. The problem,

however, is that many philosophers have wanted to make a claim not merely about

knowing being valuable but about it being more valuable than believing truly. For

example, Ward Jones (1997) writes that “I want all of my beliefs to be true—otherwise I

would not believe them—but I would also rather they be known beliefs than mere true

beliefs” (p. 424). And Jones is by no means alone. Duncan Pritchard starts out an

overview of the literature on recent discussions regarding the value of knowledge by

stating what he, correctly, takes to be an assumption shared by virtually everyone

involved in the relevant discussions, namely that “we clearly do value knowledge more

than mere true belief.” (p. 86). Moreover, many have argued that this idea can be found

already in Plato’s Meno. Jonathan Kvanvig writes:

[…] part of the challenge of explaining the value of knowledge is explaining how
it has more value than other things, one of these other things being true opinion—
as Meno claims after acquiescing to Socrates’ point that true belief is every bit as
useful as knowledge. “In that case, I wonder why knowledge should be so much
more prized than right opinion” (97c-d). Meno expresses here a common
presupposition about knowledge, one that is widely, if not universally shared.
Given this presupposition, an account of the value of knowledge must explain
more than how knowledge is valuable. It must also explain why the value of

9
Here and henceforth, any unqualified statement about value concerns epistemic value.
10
On one understanding of what knowledge is, knowledge just is true belief. We will not
be concerned with this weak notion of knowledge here, but see Goldman and Olsson
(2009) as well as Sartwell (1992) for two relevant discussions.

67
 knowledge is superior to the value of true opinion (Kvanvig 2003, pp. 3-4;
emphasis added).

We will return to the issue of what exactly Plato seems to be saying about the value of

knowledge in the Meno in the next section. As far as contemporary discussions are

concerned, the particular interpretation of aforementioned claims about the surplus value

of knowledge that is relevant for present purposes is the following:

(K) Knowing is more epistemically valuable than merely believing truly.

After all, there might be many respects in which knowing is superior to mere true belief,

including moral or practical. If it can be shown that knowing is valuable thus, that would

be interesting and important.11 But unless the surplus value in question is of a specifically

epistemic kind, it would not speak to EVM, which is a thesis exclusively about epistemic

value.

Could someone committed to EVM account for (K)? For one thing, (K) cannot be

accounted for in terms of the non-instrumental value that derives from the factivity of

knowing. Moreover, postulating an additional fundamental epistemic value, distinct from

that of true belief, would clearly not be an option for a defender of EVM. Another option

is, therefore, to argue that the non-factive component of knowledge contributes an

additional instrumental value, since instantiating that component is conducive to

believing truly. However, this strategy raises a problem that is typically directed against a

particular account of what constitutes knowledge, namely reliabilism, but applies to any

11
See, e.g., Olsson (2007) on the practical surplus value of knowledge.

68
axiological account that defines the non-factive component of knowledge in terms of

truth-conducivity. Richard Swinburne sums up the central worry as follows:

Now clearly it is a good thing that our beliefs satisfy the reliabilist requirement,
for the fact that they do means that […] they will probably be true. But, if a given
belief of mine is true, I cannot see that it is any more worth having for satisfying
the reliabilist requirement. So long as the belief is true, the fact that the process
which produced it usually produces true belief does not seem to make that belief
any more worth having (Swinburne 1999, p. 58).

In other words, the problem is that the presence of truth seems to “swamp” the epistemic

value of truth-conducivity, to use Kvanvig’s (2003) term. This does not mean that a

defender of EVM cannot account for knowledge being epistemically valuable. It does,

however, throw doubt upon her ability to account for (K).

Reliabilists have resisted this conclusion. For example, Alvin Goldman (Goldman

and Olsson 2009) argues that (K) can be accounted for in terms of a pattern of inheritance

by which token processes inherit value from states of true belief in virtue of being

instances of process types that reliably produce such states. Moreover, he argues, the

value inherited is autonomous in that it “isn’t dependent, on a case-by-case basis, on the

value of resultant true beliefs” (p. 31) and may, as such, be added to the value that the

relevant state of knowing has in virtue of its factivity. Presently, we will not be concerned

with whether this is a successful solution to the swamping problem. Instead, it will be

argued that any monistic attempt to account for (K) grants the defender of the swamping

argument too much, since we lack any strong, pre-theoretic reasons for accepting (K) in

the first place.12 If we lack such reasons, there is no longer any pressure to surrender

12
(K) has received some scrutiny in recent contributions to the debate, but not in a
manner amounting to a rejection. Baehr (2009) raises doubt about the extent to which the

69
EVM for the purpose of accounting for (K) with reference to some non-instrumental

value distinct from truth—or so it will be argued. Let us first spell out the case for the

conditional, and then turn to the case for the antecedent.

Over the last couple of years, several analyses of knowledge have been motivated

on the ground that they imply and, thereby, can account for (K).13 Focusing on pre-

theoretic reasons amounts to excluding reasons flowing from such implications. In fact,

disqualifying theoretical reasons is in the interest of both parties of the debate, since the

dialectical deck can be stacked either way. In particular, allowing for theoretical reasons

to motivate or discredit (K) opens up for a very simple reply by the monist: Taken

together, reliabilism and aforementioned swamping considerations imply that (K) does

not hold; consequently, we have reason to believe that it does not. No one has pursued

this strategy and the reason is simple: Those involved in the debate assume that (K)

amounts to a constraint on any theory of knowledge, and acknowledge that reasons

flowing from particular theories of knowledge carry no cross-theoretical force. The

present focus on pre-theoretic reasons stays true to this dialectic.

That said, making the case that we lack any pre-theoretic reasons for accepting

(K) is made complicated by the fact that it is hard to find any arguments for (as opposed

to statements of) (K) in the literature. One relevant discussion can be found in one of the

intuition typically reported in the form of (K) really is fully general, in the sense of
concerning all instances of knowledge, but seems to accept (K) as a rough generalization.
Greco (2009) questions the idea that there is widespread intuitive support for the further
idea that knowledge is more valuable than any subset of its constituents, while accepting
(K). As will be discussed below, Kvanvig (2003) denies that knowledge is more valuable
than any subset of its constituents, but accepts that (K) holds in virtue of the surplus value
of justified true belief over mere true belief. Pritchard (forthcoming) initially rejects (K),
but re-establishes it (at least in part) with reference to the value of understanding, arguing
that knowledge typically goes hand-in-hand with understanding.
13
See Pritchard (2007) for an overview.

70
earliest contributions to the contemporary debate regarding the value of knowledge, due

to Jones (1997). According to Jones, one possible motivation for (K) is that it constitutes

the best explanation of why so many philosophers have concerned themselves with

analyzing knowledge. He writes:

Surely it is the high value we place on knowledge which has motivated the
extraordinary volume of philosophical work which has gone into setting out the
conditions for knowledge as opposed to true belief (Jones 1997, pp. 423-424).

The problem with this line of reasoning can be brought out by considering the fact that

what any particular philosopher shows great interest in is a function of a whole host of

factors, such as (a) what her colleagues happen to be showing great interest in, which

influences the extent to which she will be successful in publishing her work or landing an

academic position that will enable her to remain in the field; (b) what her supervisors

happen to show great interest in, which influences the extent to which she will receive

sufficient professional attention to develop necessary academic skills; and (c) what her

philosophical predecessors happened to take great interest in, which influences the extent

to which she will consider becoming a philosopher in the first place. This is not to

suggest that philosophical inquiry is in no way sensitive to axiological facts, but merely

to point out that establishing facts about comparative interests is not sufficient for

establishing facts about comparative value, since what philosophers take great interest in

is partly a function of a great many sociological facts that cannot be assumed to track

relevant axiological facts about what degrees of value happen to apply to what states.

A second motivation for (K) may be gleaned from the fact that knowledge tends

to be highly regarded in common discourse. Jones again:

71
 Even more telling than philosophical infatuation is our unwillingness to endorse
or applaud those who succeed in guessing. What is more, those who repeatedly
guess tend to gain our disapproval, even though they may succeed at guessing. At
the very least, we lose trust in their reliability as informants. Even more than
valuing others’ knowledge, we value ourselves as knowers. For most (if not all) of
my factual beliefs, I take it to be important that they are knowledge and not lucky
guesses. I want all of my beliefs to be true—otherwise I would not believe
them—but I would also rather they be known beliefs than mere true beliefs (Jones
1997, p. 424).

To see the problem with this line of reasoning, we need to distinguish between two

targets of evaluation, i.e., doxastic states and belief-forming processes. With this

distinction in mind, consider, first, the fact that we do not applaud guessing. Construed as

a process, this could be explained with reference to the unreliability of guessing, and how

the process, hence, is of instrumental disvalue. But what about the doxastic states ensuing

from guessing? A lucky guess amounts to something of non-instrumental value, i.e., a

true belief, which, furthermore, explains why we value lucky over unlucky guesses,

ceteris paribus. Notice, however, that it makes little sense to talk about a process of lucky

guessing, as opposed to one of guessing simpliciter. After all, what would such a process

involve? Surely, it would not involve consistently getting things right; in that case, it

would no longer be plausible to refer to the process as involving guessing, as opposed to

some sub-personal sensitivity that (save for its reliability) might resemble guessing in not

being preceded by any deliberation. Indeed, as with any form of guessing, the process

would have to be one that is generally unreliable and, as such, one that we have little

reason to value.

Next, turn to the claim that we value ourselves as knowers. Construed as a claim

about how we value knowing things, this can be accounted for with reference to the

72
factivity of knowing. The same holds if we construe the claim as being about how we

value engaging in processes that involve us coming to know things. Consider, next, the

comparative claim that we value knowledge more than we value lucky guesses. On a

state reading, it is not clear that this claim is true. Imagine two scenarios. In one scenario,

you make a guess and form a true belief as a result. In the second scenario, you make a

reliable inference and, as a result, form a true belief that, moreover, amounts to

knowledge. Now, consider the comparative value not of your means of forming belief but

merely of the two resulting states. Is one really more valuable from an epistemic point of

view than the other? It is not clear that it is, given that they both involve true belief.14 In

so far as we intuit a difference here, it is likely that the intuition involved tracks what

does seem clear, namely that the processes involved in arriving at the respective states

are not equally valuable. The procedure utilized in the first scenario is of instrumental

disvalue in virtue of being unreliable, while that of the former is of instrumental value in

virtue of being reliable.

This brings us to the final claim of Jones’, i.e., that “I would also rather [my

beliefs] be known beliefs than mere true belief.” This is simply a restatement of (K), i.e.,

the very claim for which we are seeking a motivation. If it were the case that the

preceding claims could only be accounted for by assuming (K), then we could have

considered the statement a hypothesis, needed to account for the truth of the relevant

axiological claims. But what we have just seen is that it is perfectly possible to account

for those claims within a monistic framework in the manner just outlined. Consequently,

Jones’ assertion of (K) amounts not to an explanatory hypothesis necessary for explaining

14
Here, I’m siding with Kvanvig (2003, p. 148).

73
the observations preceding it, but simply to the statement of a preference. This is

certainly not to call into question the sincerity of Jones’ remark; it is simply to ask

whether a single, reported preference is sufficient to warrant acceptance of (K).

This brings us to the third potential motivation for (K), gleaned from the

observation that Jones’ preference is fairly widespread, as noted already above. However,

it is questionable whether this provides any pre-theoretic reason for accepting (K) which,

as we saw above, is the kind of reason that needs to be provided. For one thing, we would

need evidence to the effect that the intuition in question is not merely a function of

previous theoretical commitments. This is not to defend some general form of skepticism

regarding the evidentiary force of intuitions in philosophical inquiry. It is merely to

suggest that intuitions have to be bolstered by supporting considerations (as opposed to

simply being taken at face value) exactly to the extent that they (a) pertain to subtle

philosophical matters and (b) are leveraged against reputable philosophical theories.

Since matters of comparative epistemic value are subtle, and EVM is a reputable theory

(which is something that has to be granted even by those who take it to be false), a

question arises: Is it the intuition that knowledge is more valuable than mere true belief

that provides the impetus for coming up with a theory that satisfies this intuition, or is it

the having of a theory on which knowledge is more valuable than mere true belief that

provides the impetus for the intuition? The absence of any arguments ruling out the latter

should give us pause before accepting (K) as a pre-theoretically motivated constraint on

investigations into the value of knowledge.

74
3.3. Meno and the Monist

There is one final place where the defender of (K) might seek pre-theoretic support for

the idea that knowing is more epistemically valuable than merely believing truly, even in

the absence of contemporary pre-theoretic support. That place is Plato’s Meno. In fact,

Socrates’ observations in the Meno about the surplus value of knowledge over mere true

belief has become such a common point of reference for defenders of (K) that the

problem of explaining why (K) holds often is referred to simply as the Meno problem.15

That said, this taxonomy is misleading on two points. First, as pointed out by

Dennis Whitcomb (forthcoming), it is misleading to talk about the Meno problem. In fact,

Whitcomb argues that, depending on exactly how the relevant surplus claim is spelled

out, there are at least 540 different problems that one can mean to be addressing. This, of

course, is compatible with it being the case that Plato and those that turn to him in

support of (K) still have the very same problem in mind. But this brings us to the second

point: As far as we are concerned with (K), it is even misleading to talk about the Meno

problem. More specifically, it will be argued that, to the extent that the there is a surplus

claim in the Meno, it is a claim that is differs from (K) with respect to both the scope of

the relevant claim and the kind of value that is at issue. Consequently, the defender of (K)

has little to gain from turning to the Meno for support.16

15
Here and henceforth, any reference to Socrates refers to the Socrates of Plato’s
dialogues, not to the historical Socrates.
16
It will be assumed presently that the term episteme is best translated as “knowledge.”
This translation has supporters in the literature (e.g., Fine 2004), but is not altogether
uncontroversial (see, e.g., Burnyeat 1980). Notice, however, that if episteme should not
be translated as “knowledge,” that is just yet another reason not to invoke the Meno in
support of (K).

75
 Consider, first, the scope of the surplus claim that figures in the Meno. Towards

the end of the dialogue, Socrates suggests that knowledge is no more useful than mere

true belief (96c), and illustrates his point with reference to how someone knowing the

way to Larissa will be no worse of a guide to others with respect to getting them there

than someone who merely has a true belief about the way to Larissa (97a). This,

famously, makes Meno wonder “why knowledge is prized far more highly than right

opinion, and why they are different” (97c-d).17 In response, Socrates invokes the idea that

knowledge is true belief tied down by reason:

[…] true opinions, as long as they remain, are a fine thing and all they do is good,
but they are not willing to remain long, and they escape from a man’s mind, so
that they are not worth much until one ties them down by (giving) an account of
the reason why. And that, Meno, my friend, is recollection, as we previously
agreed. After they are tied down, in the first place they become knowledge, and
then they remain in place. That is why knowledge is prized higher than correct
opinion, and knowledge differs from correct opinion in being tied down (97e-
98a).

This, of course, is the very passage that has lead many contemporary epistemologists to

suggest that Plato subscribed to something along the lines of (K) in the Meno. However,

interpreting the passage thus is problematic. Notice that Plato identifies the reasoning18

relevant to knowing with recollection. Earlier in the dialogue, the process of recollection

is invoked in a solution to the so-called paradox of inquiry, introduced by Meno and

reformulated by Socrates in terms of a dilemma. According to the first horn, the inquirer

“cannot search for what he knows—since he knows it, there is no need to search” (80e).

17
All translation of Plato are from Cooper (1997).
18
As indicated by the parenthetical remark in the translation “(giving) an account of the
reason why,” the relevant Greek phrase “aitias logismos” is generally thought to refer not
to the product but to the process of reasoning (logismos) to the cause or explanation
(aitia). See Fine (2004, p 58) for a discussion.

76
On the second horn, the inquirer also cannot search “for what he does not know, for he

does not know what to look for” (ibid.).

Socrates’ solution to the paradox, involving the famous interrogation of the slave

boy about how to double the area of a square (82b-85c), is that “the man who does not

know has within himself true opinions about the things that he does not know” (85c) and

that, “if he were repeatedly asked these same questions in various ways, you know that in

the end his knowledge about these things would be as accurate as anyone’s” (85c-d). In

other words, the reason we can engage in philosophical inquiry is that we have tacit yet

true belief.19 All it takes is an elenchus. Through a persistent, dialectical inquiry in the

form of question and answer adversary argument, we may recollect what we used to

know when disembodied, but that has since been demoted from explicit knowledge to

tacit but nevertheless true belief.

That said, the claim that the theory of recollection is invoked by Plato to solve the

problem of inquiry is not altogether uncontroversial. For example, Dominic Scott (2006)

argues that the theory is instead invoked in a psychological strategy of carrot and stick,

attempting to both exploit Meno’s weakness for the exotic through allusions to ancients

myths in a context where he is getting tired of Socrates’ constant refutations, and arouse a

sense of shame in Meno by suggesting that he is weak for wanting to leave the

19
Some commentators (e.g., Taylor 2008 and Scott 2006) would prefer to talk about tacit
knowledge, and point to how Socrates (at 85d) talks about “the knowledge he [i.e., the
slave boy] now possesses”. Other commentators (e.g., Fine 2003) suggest that Socrates
here is referring to an envisaged future time, at which the slave boy has undergone a
successful elenchus, and point to how Socrates denies (at 85c) that the slave boy knows
anything at the end of the interrogation. Presently, I’m working with the latter
interpretation. Note, however, that the arguments to be made can be made just as easily
on the former interpretation as on the latter.

77
discussion.20 To the extent that the theory of recollection is introduced to solve any

philosophical problem, Scott argues, it is introduced to address a challenge that (Scott

maintains) is ignored in Socrates’ reconstruction of Meno’s challenge, when the latter

asks, “if you really stumble upon it [i.e., the definiens], how will you know that this is the

thing you didn’t know before?” (80d). Scott refers to this question as posing a problem of

discovery.

There is no reason to doubt that Plato introduced the theory of recollection to play

a subtle rhetorical role in relation to Meno. It is less clear, however, whether it was not

also meant to solve the problem of inquiry. After all, what Scott refers to as the problem

of discovery seems a mere special case of the problem posed by Socrates’ second horn,

i.e., that the inquirer cannot search “for what he does not know, for he does not know

what to look for” (80e).21 On this point, notice that Meno does not resist Socrates’

reconstruction (81a). And why should he? Someone who does not know what he is

looking for has several problems as far as inquiry is concerned, one of which is that he

will not be able to tell if he has stumbled upon the target of investigation—unless his

inquiry is guided by tacit true belief, as postulated by the theory of recollection. That is

why inquiry is possible and the inquirer avoids the second horn. Moreover, given that the

beliefs guiding him do not constitute knowledge until tied down through the relevant

process of reasoning (i.e., through recollection), inquiry is still worthwhile and the

inquirer avoids the first horn.

20
Hence, Socrates suggests that “we would be better, more manly and less lazy if we
believed that we ought to inquire into what we do not know, than if we believed that we
cannot discover what we do not know and so have no duty to inquire” (86b-c).
21
On this point, notice that Meno does not resist Socrates’ reconstruction (81a).

78
 All of this is relevant presently for the following reason: If (a) the kind of

knowledge that Plato has in mind when having Meno suggest that knowledge is prized far

more highly than mere true opinion is the kind of knowledge that results from true belief

being tied through a process of reasoning; (b) to undergo the relevant process of

reasoning is, as Socrates suggests, to recollect; and (c) recollection is introduced for the

purpose of explaining why philosophical inquiry is possible, then the kind of knowledge

that Plato is concerned with in the Meno is not just any kind of knowledge; it is the

particular kind of knowledge had by those who have gained successful insight into the

nature of things. In short, it is philosophical knowledge, not knowledge in general.

This is not to suggest that Plato was committed to a theory of Forms already in the

Meno. It suffices to note that the one theme that is recurrent throughout the dialogues is

an inquiry into essences, or that by which the phenomena under investigation are what

they are.22 Given the epistemology of recollection introduced in the Meno, interpreting

that by which virtues “have one and the same form which makes them virtues” (72c) in

terms of proto-Forms is tempting but not necessary for present purposes. The only thing

maintained here is that, as far as the Meno is concerned, claims about the surplus value of

knowledge apply to the kind of knowledge relevant to recollection, and the kind of

knowledge relevant thus is philosophical knowledge, as in knowledge of essences,

22
Hence, when Socrates asks in the Meno “If I do not know what something is, how
could I know what qualities it possesses?” (71b), he not only seems to appeal to an
implicit distinction between essential and non-essential properties, but also follows up the
request for an account of the latter (at 72b) by asking for the nature of something, using
the same word (ousia) as he uses in the Euthyphro (at 11a-b) when asking for the essence
of piety. The same distinction between the essential and the non-essential appears in the
Laches (189e-190c), Protagoras (312c and 360e-361d), and Gorgias (463c).

79
whether or not such essences are to be spelled out in terms of Plato’s later ontology of

Forms, or in terms of his earlier ontology in terms of essences more generally.

Moreover, maintaining that the Meno is exclusively concerned with philosophical

knowledge is also not to suggest that Plato necessarily took philosophical knowledge to

be the only kind of knowledge there is.23 If he did, then no qualification would be needed,

since there would no longer be a distinction between philosophical knowledge and

knowledge in general. Be that as it may; the only things necessary to make the present

point is that (a) contemporary epistemologists looking to the Meno in support of a claim

about the surplus value of knowledge do not restrict their claims to philosophical

knowledge, despite the fact that (b) the kind of knowledge at issue in the Meno’s

discussions about the value of knowledge is philosophical and only philosophical

knowledge, whether or not Plato at any point took that to be the only kind of knowledge

there is.

One potential counterexample to (b) is the fact that the distinction between true

belief and knowledge is introduced in the Meno with reference to knowing the way and

guiding others to Larissa (97a). Taken literally, this example suggests that Plato’s

concerns in the Meno were not exclusively with philosophical knowledge. However,

consider that the topic of the dialogue, i.e., arete, which is typically translated as ‘virtue,’

is an inherently political concept as far as the Meno is concerned.24 In light of this, it is

not clear that we should understand the Larissa example as anything but a mere metaphor

for the kind of leading or guidance that Socrates ultimately seems to be concerned with,

23
There is some indication to the effect that Plato took knowledge to be restricted to
knowledge of forms in the Republic (see particularly the end of book V and books VI-
VII). See Taylor (2008, pp. 178-9) for a discussion.
24
See Scott (2006, p. 14) for a discussion.

80
i.e., the kind that Themistocles et al. are engaging in when leading their cities (99b).

After all, Larissa only figures twice in the dialogue; first, as a place inhabited by people

admired for their wisdom (70b), and, then, in aforementioned example, illustrating the

absence of a distinction between knowledge and mere true belief as far as usefulness is

concerned. At no point in the dialogue does Socrates concern himself with the kind of

knowledge that literal, geographical location and guidance would have to pertain to, i.e.,

empirical knowledge. Consequently, it is far from clear that Socrates would say that we

may literally know the way to Larissa, in the way that we (if genuinely virtuous) may

know how to lead a city, which is the kind of knowledge occupying the discussion at

large.25

As such, we may conclude that, to the extent that Plato subscribes to anything like

(K) in the Meno, he is, at most, committed to the following:

(K*) Having philosophical knowledge is more epistemically valuable than is

merely believing truly.

Understood thus, the problem as far as invoking the Meno in contemporary debates about

the value of knowledge is concerned, is that those invoking (K) in an argument against

EVM, as already noted, are concerned with knowledge in general. As such, there is a

clear difference in scope between the surplus claim found in the Meno and that defended

25
A similar point can be made about the passage at the very beginning of the dialogue
where Socrates talks about knowing who Meno is (71b), which, as Scott (2006, p. 21)
points out, “is best treated as a pedagogical device to give Meno an intuitive hold on the
idea of one question (‘what is x?’) having priority over another (‘what is x like?’).”

81
by recent epistemologists. Consequently, it is not clear that the passages typically quoted

from the Meno provide great support—pre-theoretic or otherwise—for (K).

In fact, further consideration of what Plato has Socrates say about the value of

knowledge suggests that the former’s take on the value of knowledge not only differs

from (K) with respect to scope, but also with respect to the kind of value involved.

According to Socrates, (philosophical) knowledge is more valuable than mere true belief

for being tied down through a process of reasoning about the relevant essences in virtue

of which the knowledge at issue, unlike the untied work of Daedalus, will “remain in

place” (98a). In other words, philosophical knowledge is more valuable than mere true

belief (philosophical or otherwise) on account of the former’s stability. This begs a

further question: Why is stability valuable? Surely, it cannot be because it makes

philosophical knowledge more fit for use in action, since Socrates explicitly denies that

there is any difference between knowledge and mere true belief on this score (98c).

A more plausible interpretation stays true to the central theme of the Meno, i.e.,

the relation between possessing philosophical knowledge and being able to teach virtue.

The connection between teachability and the stability of the former is not immediately

obvious from Socrates’ brief remarks in the Meno, but we may begin to see what is going

on if we follow Gail Fine (2004) in looking beyond the Meno toward the Republic.

According to Fine, the kind of stability at issue in the Meno should be understood in

terms of what is said in the Republic (540b-c) about how someone who knows (at least in

the limiting case of someone grasping the Good itself), unlike someone who merely

82
believes truly, is invulnerable to refutation.26 The same goes, of course, for someone who

just happens to be extremely dogmatic, which suggests that what is valuable about the

relevant kind of stability is not invulnerability per se, but rather the particular kind of

invulnerability that arises from judging things “in accordance with being” (543c).

Differently put, philosophical knowledge is not valuable because it is tied down, but

because of what it is tied down to, i.e., an account of the essence of things.

Why is it valuable for knowledge to be tied down to such accounts? As Socrates

proposes in the Meno, having one’s beliefs be tied down thus is to have philosophical

knowledge, and to have philosophical knowledge is, as Socrates goes on to suggest, to be

able to teach virtue and, in effect, “make another into a statesman” (100a). In other

words, philosophical knowledge is (at the very least) valuable in that it enables one to

multiply virtue, which raises a further question: Why is it valuable to be virtuous?

Throughout the Meno’s discussion about the value of virtue (87d-89a), the only value it is

explicitly said to have is that which it has on account of enabling one to use one’s assets

correctly and, in effect, produce happiness (eudaimonia, at 88c). In modern parlance,

virtue is, thereby, instrumentally valuable, in virtue of its conduciveness to happiness.

In other words, as far as the Meno is concerned, both philosophical knowledge

and philosophical true belief is instrumentally valuable, in that it enables one to use one’s

assets correctly and, thereby, produce happiness. However, having philosophical

knowledge is more valuable than having true belief (philosophical or otherwise) because

someone endowed with the former not only will be able to use her assets correctly, but

26
See also Euthyphro, where Euthyphro complains that “whatever proposition we put
forward [for cross-examination] goes around and refuses to stay put where we establish
it” (11b), which prompts Socrates to make an allusion to Daedalus (11b-e).

83
also be capable of instilling the same capacities in others and, thereby, bring about

happiness on a much greater scale. More specifically, in so far as one has philosophical

knowledge, one will in general be able to bring about a much greater amount of

happiness, since the beneficial consequences of one’s endowment extends beyond the

(correct) use of one’s own assets, to include also the aggregate benefits of the actions of

one’s students of virtue. Consequently, philosophical knowledge is prized for more

highly (as Meno puts it) than mere philosophical true belief simply in virtue of how

bringing about more good is better than bringing about less good.

Aforementioned line of reasoning makes sense of Socrates’ approval of Meno’s

suggestion that philosophical knowledge tends to be prized far more highly than mere

philosophical true belief, as well as his explanation in terms of stability and recollection.

It does not go to show, however, that every instance of philosophical knowledge is more

valuable than every instance of mere philosophical true belief. After all, we can surely

imagine a statesman of mere true (philosophical) belief bringing about an immense

amount of good on account of ruling a massive city correctly, while another statesman,

albeit endowed with knowledge, is unable to find neither subjects nor students on account

of geographical isolation or other external circumstances. However, as already noted,

what Socrates sets out to show is merely that the former is prized far more highly than the

latter. To make that claim plausible, it is sufficient to show that philosophical knowledge

in general gives rise to more happiness than mere true belief, philosophical or otherwise.

Notice, however, that the above interpretation of Plato’s take on the value of

knowledge in the Meno, interestingly, gives rise to a kind of swamping problem,

pertaining not to conduciveness to truth but to happiness. More specifically, if the value

84
of virtue is exhausted by the instrumental value it has because it is conducive to

happiness, does the presence of happiness swamp the value of virtue? It is not clear what

Plato would say here at the time of the Meno. It is interesting to note, however, that Plato

in the Republic sets out to defend the claim that justice is not only good as an instrument

but also as an end, in being both an instrument to happiness and an integral part of the

good life. However, no analogous claim is made about the value of virtue in the Meno,

which merely talks about the instrumental value of being virtuous. As such, I will leave it

open whether such a swamping problem would worry Plato at the time of the Meno.

This leaves us with one final question about Plato’s take on the value of

knowledge in the Meno: What kind of value applies to philosophical knowledge? If

philosophical knowledge is valuable because it enables the knower to multiply virtue, and

virtue is valuable because it enables one to use one’s assets correctly and, in effect,

produce happiness, it seems fair to say that, whatever value attaches to knowledge on

account of being generally conducive to happiness thus, it is not an epistemic kind of

value. In other words, in so far as Plato subscribed to anything like (K) at the time of the

Meno, the most plausible interpretation is the following:

(K**) Having philosophical knowledge is more non-epistemically valuable than

merely believing truly.

Setting aside the issue that (K**), like (K*), only pertains to philosophical knowledge, as

opposed to knowledge in general, notice that (K**) is compatible with EVM, which is a

thesis exclusively about epistemic value. More generally, given that it does not follow

85
from something being valuable on one qualification (e.g., morally valuable, prudentially

valuable, or what have you) that it is valuable on every other qualification, (K**) is

compatible with EVM no matter how the relevant non-epistemic value is spelled out. In

fact, given that it does not follow from something being valuable without qualification

that it is valuable on every qualification, even a construal of (K**) in terms of knowledge

simply being more valuable simpliciter than mere true belief is compatible with EVM. In

conclusions, it is far from clear that contemporary critics of EVM have much to gain in

terms of support from Plato’s Meno.

3.4. Justification and Non-instrumental Value

The previous sections suggested that there is scant pre-theoretic support for (K) in the

contemporary literature, and far less to gain for the defender of (K) in turning to the

Meno than what seems to be assumed in the literature. As such, there are no strong

reasons to account for (K), let alone to account for it in terms of knowledge being of

fundamental value. However, consider a point made recently by Jason Baehr (2009), who

argues that it is implausible to believe that, “when we think of knowledge as more

valuable than true belief at the relevant intuitive level, we are thinking of it purely in the

abstract, without any (even implicit) reference to any of the features in virtue of which it

is apparently more valuable” (p. 51). This highlights a possibility ignored in the previous

sections, namely that the intuition behind (K) does not concern the surplus value of

knowledge directly, but rather the fundamental value of some component of knowledge,

86
that, when combined with the value of the factivity of knowledge, amounts to knowledge

being more valuable than mere true belief.

What are the candidates for such a component? Clearly, truth is not one of them,

nor is mere belief. This leaves us with two partially overlapping candidates, namely

justification and warrant. Here, warrant is understood as whatever factor turns true belief

into knowledge (Plantinga 1993), and justification as warrant minus whatever condition

needs to be added to a JTB account in order for it not to have Gettier cases come out as

cases of knowledge. Presently, the focus will be on justification, the reason being that it is

unclear whether an account of warrant will tell us anything interesting about value. In a

passage commenting on Peter Klein’s (1981) attempt to avoid the Gettier problem by

way of a distinction between different kinds of defeaters, Kvanvig writes:

It is hard to see this distinction as anything more than gerrymandering needed to

prevent counterexamples to one’s account of knowledge, and it is easy to side
with Williamson in remarking ‘Why should we care about that?’ The distinction
between these kinds of defeaters tracks no intuitive difference in value, leaving us
with an account of the nature of knowledge incapable of helping to explain the
value of knowledge (Kvanvig 2003, pp. 129-130).

According to Kvanvig, Klein’s analysis is in no way unique in this respect. In fact,

[…] the better an approach to the Gettier problem is at carving cases of

knowledge off from cases of non-knowledge, the more ad hoc and gerrymandered
the proposal. The result is a condition which has no hope whatsoever of giving a
decent answer to the question of what makes un-Gettiered justified true belief
more valuable than justified true belief. Hence, the hope of defending the view
that knowledge is more valuable than any proper subset of its parts is dim indeed
(Kvanvig 2009, p. 103).

In other words, any account that includes a Gettier clause (which any account of warrant

will) is bound to be ad hoc and gerrymandered, and any ad hoc and gerrymandered

87
analysis will fail to explain the surplus value of knowledge over justification. This

suggests that the most promising place to look for an account of the surplus value of

knowledge over true belief is not in the Gettier clause. Furthermore, since the Gettier

clause is what differentiates warrant from justification, focusing on the latter rather than

the former seems a more fruitful strategy.27

In fact, Kvanvig argues that there are at least two kinds of accounts of

justification that may escape the swamping problem and, thereby, account for (K),

namely certain (access) internalist and virtue epistemological theories. Further examples

can be found. As we saw in the previous chapter, Richard Feldman (2002) denies that

believing truly as opposed to being justified is of non-instrumental epistemic value. If

Feldman is right, (K) holds simply in virtue of the fact that (merely) believing truly is of

no non-instrumental epistemic value, while knowing is of such value in virtue of the non-

instrumental value of being justified. Since I have already discussed Feldman’s position

at length, I will not discuss it further here. Moreover, it is important to remember that

what we are after here are pre-theoretic reasons for at all seeking to account for (K). As

stressed above, simply providing a theory that implies that (K) holds, without providing

any reasons as for why theories should imply (K), would be in the interest of neither

party of the debate.

With this in mind, return to the possibility called attention to above about the

intuition captured by (K) being an intuition not about knowledge per se, but rather about

27
This is not to suggest that an analysis being “ugly” yet correct implies that the
analysandum cannot be valuable. As pointed out by DePaul (2009), we cannot assume
that accuracy of analysis implies transparency with respect to value. The point relevant
here is simply that consulting analyses that are bound to be “ugly” does not constitute a
promising strategy, in so far as we are interested in questions about value, exactly
because we cannot assume such transparency.

88
the fundamental value of some component of knowledge, that, when combined with the

value of the factivity of knowledge, amounts to knowledge being more valuable than

mere true belief. Then assume, if only for the sake of the argument, that there is, in fact, a

widespread, pre-theoretic intuition to the following effect (for some sufficiently pre-

theoretic notions of ‘justification’ and ‘non-instrumental’):

(J) Being justified is non-instrumentally valuable.

A widespread intuition to this effect would constitute prima facie evidence for the claim

that being justified is non-instrumentally valuable. Moreover, given that that justification

is non-factive, the non-instrumental value in question cannot be derived from that of true

belief, as in the case of knowledge and other factive states. Consequently, the most

plausible response to the evidence would be to consider the relevant value fundamental.

As such, the evidence in question would not only provide indirect support for (K) but also

constitute a direct challenge to (2), i.e., the idea that nothing is of non-instrumental

epistemic value unless it is so in virtue of the non-instrumental epistemic value of

believing truly.

That said, the prima facie evidence in question might, like any prima facie

evidence, be defeated. In fact, it will be argued momentarily that the evidence in question

is defeated. More specifically, it will be argued that it cannot be ruled out that we would

intuitively take being justified to be non-instrumentally valuable even if it was not, given

what can reasonably be assumed about the central role that the inculcation and

reinforcement of an appreciation for instrumentally valuable states and processes plays in

89
inquiry. As such, the argument to be provided discredits the reliability of the particular

intuition in question, as far as the tracking of relevant axiological facts about the value of

justification is concerned, and suggests that even a widespread intuition to the effect that

(J) holds does not provide a reason for taking its content at face value.

First, however, it needs to be established that the relevant kind of axiological

confusion is at all conceptually possible. After all, if what it is for something to be

valuable simply is for it to be valued (or preferred) in the relevant way, the confusion in

question is not possible. However, there are two problems with such a preferentialist

account of epistemic value. First, it does not do a good job of accounting for instrumental

value. Consider, for example, a particular research method. In so far as using that method

is conducive to something that is valuable for the inquirer, such as true belief, using the

method will be instrumentally valuable for the inquirer. However, it does not follow that

such usage, thereby, is instrumentally valued by the subject—after all, she may not even

be aware of the method, let alone of its virtues. Consequently, preferentialism is not a

plausible theory for instrumental epistemic value.

In fact, the thesis of preferentialism remains questionable, even if restricted to

non-instrumental value. The main problem is that such a thesis would either beg the

question against a defender of (J), or run the risk of rendering the domain of epistemic

value unrecognizable. To see why, consider that the thesis would be that all and only

targets of non-instrumental preferences constitute non-instrumental values. Specified

thus, there is no constraint whatsoever on what kinds of preferences would be relevant to

epistemic as opposed to other kinds of value. In particular, there would be no way to

disqualify completely arbitrary preferences, such as a non-instrumental preference for

90
beliefs formed on Wednesdays. Or rather, there is a possible constraint, to the effect that

the preferences in question must be connected to considerations about truth. However,

that would simply beg the question against those who want to take the intuition to the

effect that (J) holds at face value and maintain that justification is valuable independently

of considerations about truth.

Consequently, it should come as no surprise that preferentialism typically is

identified with the significantly weaker thesis that non-instrumental value accrues only to

(but not to all) targets of non-instrumental preferences.28 Notice, however, that since such

a preferentialism does not take non-instrumental preferences to be sufficient for non-

instrumental value, it does not rule out the possibility that an inquirer values something

non-instrumentally, even if that something is only instrumentally valuable.

But conceptual possibility is one thing; in order to speak to the question of

whether we should take an intuition to the effect that (J) holds at face value, we need to

say something about whether it is at all likely that the relevant kind of confusion—i.e., a

confusion to the effect that we mistakenly have come to consider justification non-

instrumentally valuable, despite it only being instrumentally valuable—might have taken

place and is driving the intuition in question. Above, it was suggested that epistemic

value is to be understood in terms of the goals of inquiry. As such, it seems reasonable to

suppose that our intuitions about epistemic value will be sensitive to considerations about

the manners in which such goals are pursued in the practice of inquiry. With this in mind,

consider that we neither have voluntary influence over what we believe, nor an ability to

determine directly whether what we believe is true. Indeed, even such an optimist about

28
See, e.g., Rønnow-Rasmussen (2002).

91
introspection as Descartes took to developing rules for the direction of mind. Today, we

may have our doubts about the particular rules that he settled on, but the general insight

remains: we want truth, but in a “nonmagical world” like ours, to borrow a phrase from

Bernard Williams (1978), we have no choice but to resort to pursuing means for

increasing the chances that we will come to believe truly.

This predicament has implications for what kind of norms of inquiry can be

fruitfully used and communicated in the process of epistemic education. The first

implication is a negative one: Norms formulated exclusively in terms of true belief, e.g.,

“Believe what is true,” will not only have an air of complete triviality about them—“Of

course we should believe what’s true,” the inquirer in training might say, and rightly so

in light of above reflections on the goal of inquiry—but also be completely unhelpful

from the point of view of teaching a budding inquirer how to conduct inquiry

successfully. The second implication is a positive one: The kind of norms that may be

fruitfully communicated in the process of epistemic education is exactly the kind that

pertains to means to truth, rather than to truth itself. This is mirrored by the way in which

inquiry (in its many manifestations) is typically taught. What is inculcated is not only a

love of truth, but also—and perhaps even more persistently—a love for a variety of

processes and activities of epistemically informed conduct, be it in the form of the

sometimes quite abstract principles about evidence, reflection, and coherence often found

in epistemological analyses, the variety of logical tools taught in philosophy classes, or

the rigorous methods of scientific inquiry used in the natural sciences.

In fact, it seems reasonable to believe that the majority not only of the inquirer’s

education but also of her subsequent inquiry will be concerned not with questions of truth

92
per se, but rather with the pursuit of means to truth. After all, the reason norms

formulated exclusively in terms of true belief are of so little use to inquiry is not merely

that they play a very limited role in the process of education, but also because they are

largely unhelpful for the practice of inquiry in general. More specifically, for any

question pondered by the inquirer, she either knows the answer, or she does not. If she

knows the answer, any injunction to the effect that she should believe what is true will be

unnecessary. If she does not know the answer, the same injunction will be useless. For

this reason, the love for the means to truth that is inculcated throughout an inquirer’s

education is likely to be further reinforced throughout her practice, and it cannot be ruled

out that she over time comes to value those states in themselves.

This hypothesized process of inculcation and reinforcement has a lot in common

with what Goldman (Goldman and Olsson, 2009) recently has referred to as value

autonomization, whereby processes that originally are regarded as merely instrumentally

valuable later are “upgraded to the status of independent value, thereby accommodating

the legitimacy of adding their value to that of true-belief outcomes” (p. 35). As we saw

above, Goldman invokes this process in order to account for why we ascribe value in

accordance with (K). However, in the preceding sections, doubts were raised as to

whether we really have any pre-theoretic reasons to account for (K) in the first place. The

purpose of invoking an analogous process here is, therefore, not to account for (K), but

rather to undercut the claim that an intuition to the effect that (J) holds should be taken to

provide indirect, pre-theoretic support for (K), by virtue of justification being a

component of knowledge.

93
 As such, the present argument is more directly related to John Stuart Mill’s (1861)

argument against virtue theoretic counterexamples to the principle of utility. Mill

famously maintained about virtue that “[t]here was no original desire of it, or motive to it,

save for its conduciveness to pleasure, and especially to protection from pain” (p. 38).

However, due to that very conduciveness, we have come to not only associate the virtues

and vices with the moral and the immoral, respectively, but over time also forgotten the

yardstick against which the virtues originally earned their titles. Mill makes an analogy

with the love of money, noting that, even if the worth of money is solely that of the things

which it will buy, “money is in many cases desired in and for itself; the desire to possess

it is often stronger than the desire to use it, and goes on increasing when all the desires

which point to ends beyond it, to be compassed by it, are falling off” (p. 37). The

possibility pondered presently is that something similar might have happened with our

axiological conceptions of justification. As aforementioned reflections on preferentialism

revealed, however, such a re-conceptualization does not make justification non-

instrumentally valuable. Albeit valued in itself—and probably for good reason, given

how inquiry is taught and practiced—the means associated with justification remain of

mere instrumental value, in so far (and only in so far) as they enable the inquirer to form

true belief.

Naturally, none of this goes to establish beyond all possible doubt that such

axiological confusion has actually taken place. Still, the purpose of the present discussion

is merely to suggest that such confusion cannot be ruled out, given what we can

reasonably assume about the way in which inquiry is typically taught and practiced. This

ties in with what was suggested above about (K), namely that intuitions that pertain to

94
subtle philosophical matters and are leveraged against reputable philosophical theories

have to be bolstered by supporting considerations, as opposed to simply being taken at

face value. It is in this context that the above line of reasoning should be understood. By

providing a plausible story suggesting that any intuition to the effect that (J) might not

track the relevant axiological facts, the argument puts a significant burden of proof on

those who want to take (J) at face value. In the absence of a rival account of the source of

the intuition in question that explains why it would track the relevant facts, we have no

reason to endow intuitions to the effect that (J) holds with any great evidential weight, or

take it to provide a convincing counterexample to (2).

As above, it bears stressing that this is not a case for general skepticism regarding

the evidentiary force of intuitions, but merely a suggestion to the effect that a particular

intuition cannot be taken at face value, given what we can reasonably assume about its

potential insensitivity to the relevant facts. Granted, a structurally similar argument could,

in principle, be given for other candidates for non-instrumental value as well. However,

two things should be noted. First, each such additional argument would have to be

considered on its own merit, with respect to how plausible a story it can provide about

how the intuitions in question fail to track the relevant axiological facts. In other words,

while the above argument may provide a structure for such arguments, it certainly does

not provide their conclusions. Second, interestingly, it is not clear that such an argument

could plausibly be given with respect the one value on which the present investigation

rests, i.e., the non-instrumental value of true belief. After all, such an argument would

have to conclude that true belief is not even a goal of inquiry—a claim that could be

interpreted in either of two ways. One the one hand, it could be understood as a claim to

95
the effect that there is a multitude of incompatible yet equally legitimate notions of

inquiry, only one of which is defined in terms of true belief. However, as we saw in the

previous chapter, this would, at most, imply that true belief is not non-instrumentally

valuable for everyone, not that true belief is not non-instrumentally valuable for anyone,

including those who engage in inquiry, as the latter is construed by the great majority of

epistemologists. On the other hand, it could be interpreted as a claim to the effect that

there is no such thing as inquiry at all. It is terribly hard to make sense of such a claim,

and an argument terminating in such a conclusion would, most plausibly, not be properly

described as a modus ponens but as a modus tollens.

3.5. The Trials and Achievements of Understanding

To recapitulate, it has been argued that we lack convincing pre-theoretic reason to take

knowing to be of greater epistemic value than merely believing truly, or to consider being

justified non-instrumentally valuable. The present section considers the third and final

candidate for a bearer of such value: understanding. Two recent accounts of

understanding will be discussed, the first one due to Kvanvig, and the second one due to

Pritchard. Let us consider the former first.

3.5.1. Objectual Understanding

Kvanvig (2003) is concerned with what he refers to as objectual understanding, i.e., the

kind of understanding ascribed with respect to an object (as opposed to a proposition), as

when we talk about someone understanding politics or the presidency (p. 191). According

96
to Kvanvig, such understanding is factive (p. 190), and requires the satisfaction of a

coherentist condition, consisting in “an internal grasping or appreciation of how the

various elements in a body of information are related to each other in terms of

explanatory, logical, probabilistic, and other kinds of relations that coherentists have

thought constitutive of justification” (pp. 192-193). Crucially, however, understanding is,

on Kvanvig’s analysis, compatible with luck. As such, an analysis of understanding does

not need to satisfy a Gettier condition and is not susceptible to the worries discussed at

the beginning of §3.4.

This opens up the possibility that it can be shown for understanding what Kvanvig

argues cannot be shown for knowing, i.e., that it is more valuable than any proper subset

of its components. However, as stressed above, it is one thing to construct a theory

implying that p, and quite another to provide pre-theoretic reasons as for why we should

want our theory to imply that p in the first place. According to Kvanvig, “there is at least

as much intuitive support for the idea that understanding has value beyond that of its

subparts as there is for the idea that knowledge has such value” (p. 188). In light of what

has been argued above, this statement might be less comforting than Kvanvig probably

intended it—a problem that is exacerbated by the fact that Kvanvig provides no

additional pre-theoretic support for such value in the case of understanding, beyond that

which he assumes already exists for knowing.

Instead, Kvanvig proceeds to argue that this alleged surplus value of

understanding can be accounted for. Since the account provided is a highly intriguing

one, let us make an exception and waive the above requirement to provide pre-theoretic

97
support, if only for the sake of the argument. More specifically, let us grant Kvanvig the

following dialectical advantage: If the account he provides makes it plausible that

(U) understanding is more epistemically valuable than merely believing truly,

then we have good reason to believe that we have found a candidate counterexample to

(2). After all, in analogy with what was said about (K) above, (U) could neither be

accounted for with reference to the factivity of understanding, nor in terms of

instrumental connections to true belief on part of any components of understanding (on

pain of falling prey to the swamping problem). Consequently, the most plausible

inference to draw from (U) would be that the surplus value of understanding over mere

true belief is due to a fundamental, non-instrumental value, distinct from that which

applies to true belief, which, clearly, would contradict (2). On the other hand, if the

account provided by Kvanvig does not make it plausible that (U) holds—and below it

will be argued that it does not—then (a) the same would have to be said for the stronger

claim about understanding being more epistemically valuable than any proper subset of

its components, and (b) (2) remains unchallenged.

According to Kvanvig, the value relevant to accounting for (U) is that associated

with the satisfaction of the coherentist condition. In order to avoid the swamping

problem, Kvanvig takes the relevant kind of value to be non-instrumental (p. 200). In

explicating the relevant kind of value, Kvanvig makes a distinction between intentional

and effective means to goals, such as the goal of believing truly (pp. 63-64). Effective

means are instrumentally valuable in so far as they raise the likelihood of securing the

98
relevant ends. Intentional means are also valuable, according to Kvanvig, but not because

they raise the likelihood of securing the goals in question. Such means are involved in

trying to attain goals, and account for the value of such attempts even if they fail to raise

the likelihood of attaining the goals in question as a result of trying. It is in this kind of

value that we find the surplus value of understanding over mere true belief. More

specifically, satisfying the coherentist condition is “valuable because it is constituted by

adopting intentional means to the goal of truth” (p. 200).

Why is merely trying to attain true belief valuable, even if so doing goes no

lengths whatsoever toward actually attaining true belief? Kvanvig’s answer invokes a

story not too different from the one alluded to in §3.4, albeit with an introspectionist

twist:

Truth is not a property that is always reflectively accessible […] and so we must
adopt some means for identifying beliefs to hold in order to achieve the indirect
goal of believing a claim if and only if it is true. So we should try to have, or
value, beliefs with some other property, one that we can tell directly and
immediately whether a belief has (Kvanvig 2003, pp. 65-66).

Internalist justification is one such property—a “mark of truth” (p. 64), as Kvanvig calls

it—albeit one that is unable to account for the surplus value of knowledge over any

subset of its constituents due to aforementioned Gettier worries. Another such property is

the internal grasping of how various elements in a body of information are related.

Moreover, according to Kvanvig, the value associated with adopting such a grasp as an

intentional means toward believing truly is exactly what accounts for (U).

Or does it? Consider two possible interpretations of Kvanvig’s claim about the

value of adopting intentional means. First, let us assume that the idea is that simply trying

99
to attain true belief—no matter the instrumental merits of the means utilized—is

epistemically valuable. This would constitute a highly implausible commitment. Imagine

a subject that sincerely tries to attain true belief, yet opts for a reflectively accessible

means toward truth that, due to a particularly deviant conception on the subject’s part

about what will actually get her the truth, actually is highly detrimental to her attainment

of true belief. Claiming that the mere fact that she has adopted certain intentional goals

by virtue of sincerely trying to attain true belief makes her conduct epistemically valuable

seems highly implausible.29 This is not to deny that opting for reflectively accessible

surrogates for truth might often be a good thing for reasons not too different from those

discussed in the previous section, but merely to point out that opting for just any

reflectively acceptable surrogate does not yield any epistemic value as such.

Second, it might be argued that this interpretation ignores the nature of the

coherentist condition that Kvanvig takes to be central to understanding. That is, let us

assume that Kvanvig has in mind the significantly weaker claim that aiming for truth by

way of some particular intentional means—including but not necessarily limited to the

kind of grasp involved in understanding—is non-instrumentally valuable. In other words,

perhaps what explains the surplus value of understanding is not that something is

intentionally pursued as a mark of truth, but that what is intentionally pursued thus is

something with epistemically virtuous properties, where the virtue in question (on pain of

rendering the account susceptible to swamping problems) is not parasitic upon

29
It might be objected that there is still something admirable or valuable about someone
sincerely pursuing truth, even if she does a horrible job at it. That may be so—after all,
we admire people and their conduct for all kinds of reason, most of which are not
importantly related to the formation of true belief. That said, to be relevant presently, the
kind of admiration or value at issue would need to be epistemic, and that is what seems
highly doubtful.

100
instrumental connections. This interpretation seems more faithful to what Kvanvig

eventually concludes about the value of understanding. At the very end of his book, he

writes about the grasping of explanatory relationships that is involved in understanding

that

[…] to have mastered such explanatory relationships is valuable not only because
it involves the finding of new truths but also because finding such relationships
organizes and systematizes our thinking on a subject matter in a way beyond the
mere addition of more true beliefs or even justified true beliefs. Such organization
is pragmatically useful because it allows us to reason from one bit of information
to other related information that is useful as a basis for action, where unorganized
thinking provides no such basis for inference. Moreover, such organized elements
of thought provide intrinsically satisfying closure to the process of inquiry,
yielding a sense or feeling of completeness to our grasp of a particular subject
matter. In sum, understanding is valuable because it is constituted by subjectively
justified true belief across an appropriately individuated body of information that
is systematized and organized in the process of achieving understanding, and
subjectively justified true belief that is systematized in this way is valuable
(Kvanvig 2003, p. 202).

In other words, there are three potential sources for the value of understanding. The first

one is that the relevant grasp may lead to new truths being found. However, considering

that the value in question would be instrumental, it could not account for (U), due to

aforementioned swamping problem. The second source is that understanding brings

systematicity to our thinking. That may be a good thing, but Kvanvig explicitly states that

the kind of value relevant to such systematicity is pragmatic, not epistemic. As such, it

cannot account for the surplus value relevant to (U). The third source is that

understanding brings a feeling of closure. Such a feeling may be pleasurable and, hence,

prudentially valuable. However, as such, it does not seem relevant to epistemic value,

nor, consequently, to (U). Hence, even waiving the requirement to provide pre-theoretic

support for (U), we still lack reason to believe that the value in question is not exhausted

101
by (a) the non-instrumental value derived from its factivity, and (b) the instrumental

value it may have as an effective means to truth. As such, Kvanvig’s notion of

understanding poses no challenge to EVM.

3.5.2. Understanding-Why

There are, of course, other types of understanding besides objectual understanding. More

recently, Pritchard (forthcoming) has defended a view about understanding-why, i.e., the

kind of understanding that is involved in understanding why such-and-such is the case.

Pritchard’s account is meant to account for what he refers to as the distinctive value of

such understanding. To be of distinctive value, Pritchard explains, is both a matter of

degree and kind (p. 13). More specifically, while Pritchard takes understanding-why

(henceforth, simply “understanding”) to be factive (p. 113), he does not take the value of

understanding to be exhausted by the value that it has in virtue of its factivity. For

understanding to be of distinctive value is for understanding to be both more valuable

than anything that falls short of understanding, including true belief, and for that surplus

value to be due to understanding being of fundamental epistemic value.

Presently, we will focus on the latter claim, since it is the more basic one of the

two. After all, the surplus value in question is supposed to be explained with reference to

the non-instrumental value of understanding. But before delving into Pritchard’s theory,

let us consider what pre-theoretic reasons we have for at all taking understanding to be

distinctively valuable. According to Pritchard, “[t]he intuition that understanding is

distinctively valuable is surely even stronger than the intuition that knowledge is

distinctively valuable” (p. 101). If what has been argued above is correct, it better be. But

102
since Pritchard does not elaborate on the claim, let us simply grant him an advantage

analogous to that granted to Kvanvig above, and waive the requirement on pre-theoretic

support; all that is required for Pritchard’s theory of understanding to amount to a

challenge to (2) is that it makes plausible that understanding is of a non-instrumental

value that does not derive from its factivity.

Pritchard’s case for understanding being valuable thus rests on a distinction

between two kinds of value that we familiarized ourselves with already in the above, i.e.,

final value and fundamental value. According to Pritchard, a fundamental epistemic good

is an “epistemic good whose epistemic value is at least sometimes not simply [an]

instrumental value relative to a further epistemic good” (p. 19). More specifically,

Pritchard takes the defining feature of fundamental epistemic value to be that it can “act

as the terminus for the instrumental regress of epistemic value” (p. 19). Consequently,

Pritchard’s fundamental goods correspond to what we have been referring to as goods of

fundamental (non-instrumental) epistemic value above. After all, any good whose non-

instrumental epistemic value derives from the value of one of its components can not act

as the relevant kind of terminus, since any question about why that good is of non-

instrumental epistemic value would have to be answered with reference to a further

epistemic good, i.e., the epistemic good of the component in question.

Next, consider final value. According to Pritchard, a good is of final value if it is

of non-instrumental value simpliciter, i.e., if its value holds independently of any

considerations about conduciveness whatsoever. As noted in §3.1, assuming that at least

some instances of true belief are of fundamental epistemic value is compatible with

maintaining that (those) instances are only valuable thus in so far as they pertain to

103
questions posed by some relevant set of inquirers. In other words, maintaining that some

instances of true belief are of fundamental non-instrumental epistemic value is

compatible with maintaining that those instances are valuable thus only in so far as they

are of instrumental non-epistemic value, and, hence, not of non-instrumental value

simpliciter. So, again, fundamental non-instrumental epistemic value does not imply final

value.

However, according to Prichard, “final value simpliciter does entail fundamental

epistemic value.” “After all,” he writes, “if an epistemic good has a value which is not an

instrumental value then, a fortiori, it has a value which is not an instrumental epistemic

value either” (p. 20). That seems right. But surely it does not go to show that final value

implies non-instrumental epistemic value. There are more ways of not being of

instrumental epistemic value than to be of non-instrumental epistemic value, including to

not be of any epistemic value at all. As we noted already in §3.1, there are plausible

candidates for finally valuable goods, such as the value that pertains to undergoing an

episode of pleasure, that, clearly, are not thereby of epistemic value, let alone of non-

instrumental epistemic value.

This presents a problem for Pritchard, since he rests his entire case for the value

of understanding on this alleged implication. More specifically, he argues that the reason

that such understanding is of non-instrumental, epistemic value is that it involves a

cognitive achievement, where an achievement is a success that is due to the achiever’s

skills, and thus creditable to the exercise of her abilities. Moreover, according to

Pritchard, such achievements are finally valuable. How so? Pritchard asks us to imagine

the following scenario:

104
 Imagine […] that you are about to undertake a course of action designed to attain
a certain outcome and that you are given the choice between merely being
successful in what you set out to do, and being successful in such a way that you
exhibit an achievement. Suppose further that it is stipulated in advance that there
are no practical costs or benefits to choosing either way. Even so, wouldn’t you
prefer to exhibit an achievement? And wouldn’t you be right to do so? If that’s
correct, then this is strong evidence for the final value of achievements (Pritchard
forthcoming, p. 44).

Combining Pritchard’s points about understanding and cognitive achievements, his case

for the non-instrumental epistemic value of understanding can be reconstructed thus:

(1) Cognitive achievements are finally valuable.

(2) All finally valuable goods are of non-instrumental epistemic value.

(3) Cognitive achievements are of non-instrumental epistemic value.

(4) Understanding is a cognitive achievement.

(5) Understanding is of non-instrumental epistemic value.

This is a valid argument. But it is not sound, and for two reasons. First, as called attention

to above, all finally valuable goods are not of non-instrumental epistemic value, since

some finally valuable goods are of no epistemic value at all. This undermines (2).

Second, even setting this concern aside, (1) is independently questionable.

Remember, something is finally valuable if and only if it is non-instrumentally valuable

without qualification. If we grant Pritchard’s claim that we would prefer to exhibit a

cognitive achievement in the scenario imagined (as opposed to, perhaps, having no

preference either way), we have some reason to rule out exhibiting an achievement being

of instrumental practical value. However, this establishes neither that (a) exhibiting an

105
achievement is of non-instrumental practical value (since there are more ways of not

being of instrumental practical value than to be of non-instrumental practical value,

including to not be of any practical value at all), nor that (b) exhibiting an achievement is

not of some non-instrumental qualified value. As for the latter, it might, for example, be

that there is something intrinsically pleasing about exhibiting an achievement, and that

such achievements, hence, may be of non-instrumental prudential value. However, none

of this serves to establish the far stronger claim that cognitive achievements are of final

value, i.e., of non-instrumental value without qualification.

By way of conclusion, Pritchard’s account of understanding does not successfully

support the claim that such understanding is of non-instrumental epistemic value, even if

we waive the requirement to provide pre-theoretic support. For all we are being told, the

value of understanding is exhausted by the non-instrumental value that derives from its

factivity, and whatever value is associated with understanding being a cognitive

achievement. However, contrary to what Pritchard argues, it is not clear that the latter

kind of value is final. Moreover, even if it is granted that cognitive achievements are

finally valuable, and that understanding is a form of cognitive achievement, this does not

imply that understanding is of non-instrumental epistemic value. As such, we may

conclude that Pritchard’s account of understanding poses no problem for (2), nor

consequently for EVM.

106
3.6. Conclusion

Recently, epistemic value monism has come under attack by philosophers arguing that

we cannot account fully for the domain of epistemic value in monistic terms. But as we

have seen, the relevant arguments fail to establish any such thing. For one thing, there is a

presumption of monism due to considerations about axiological parsimony. Granted, such

a presumption would be defeated by positive evidence to the effect that the relevant kind

of monism makes us unable to fully account for the domain of epistemic value. However,

a proper examination and subsequent rebuttal of the most promising counterexamples to

monism, in the form of three candidates for states of fundamental, non-instrumental

epistemic value, casts serious doubt upon the claim that there is any such evidence.

Consequently, epistemic value monism still stands.

107
 CHAPTER 4

A FRAMEWORK FOR EVALUATION

The previous two chapters established two things. First, believing truly is of non-

instrumental epistemic value (albeit not necessarily for everyone), in virtue of true belief

being a goal of inquiry. Second, a careful consideration of the most promising

counterexamples to epistemic value monism suggests that believing truly is unique in

being valuable thus. As noted earlier, this is not to say that all instances of true belief are

non-instrumentally valuable. As will be argued below, only the subset consisting of true

beliefs that are significant are valuable thus. Much of §4.1 will be devoted to spelling out

the relevant notion of significance, by explaining how believing some true propositions is

in some relevant epistemic sense more important to inquirers than believing other true

propositions, specifically in so far as the truths in question serve to answer questions that

the inquirers at issue are posing.

As such, the account to be developed tells us something about what inquirers

should do, in so far as they want answers to their questions: they should employ the

particular practices that will enable them to answer those questions. §4.2 will be

concerned with tracing the limits of this ‘should.’ In particular, it will be suggested that it

follows from the account to be defended that there are very real limits to the extent to

which we can make epistemic judgments about what questions inquirers should pose (as

opposed to about what methods or practices to opt for, given a set of questions).

However, it is also argued that, rather than indicating a flaw in the theory, this simply

108
suggests that it is able to maintain a clear distinction between epistemic and other kinds

of betterness (e.g., moral, practical, etc.), and, thereby, account for the specialized nature

of epistemic evaluation.

4.1. Defining Epistemic Betterness

As was noted already in chapter 1, one of epistemology’s main missions is to evaluate

and improve our epistemic ways through ameliorative recommendations. Such

recommendations can be implemented either on the individual or on the social level. For

reasons that will be discussed at length in the next chapter—in essence, that ameliorative

advice provided directly to the individual rather than implemented by way of a social

paternalistic mandate are likely to face problems of defection—the focus will presently

be on the evaluation and amelioration of social practices. By a ‘social practice,’ I mean a

coordinated implementation or utilization of a method or a set of methods by a group of

inquirers, where a ‘method’ is to be understood as an acquired (as opposed to native)

belief-forming process.1

However, before we can talk of an improvement of such practices, we need to talk

about what it would mean for one practice to perform better from an epistemic point of

view than another. More specifically, any evaluation (and attempt at amelioration) needs

to be made against the background of (a) an account of what practices need to produce in

order to be positively evaluated, as well as (b) a framework for comparing practices with

respect to the production of the relevant goods. In short, we need a framework of

1
See Goldman (1986, p. 93).

109
evaluation that involves an account of epistemic betterness, as applied to practices. The

present chapter develops such a framework in terms of the monistic account of epistemic

value defended in the previous two chapters.

How are we to spell out the notion of epistemic betterness, given such an

axiology? To a first approximation, we may try to do so with reference to the reliability

of a practice, i.e., the ratio of true to false belief that the practice produces (or would

produce) under a given set of circumstances. However, a practice may generate a high

ratio but a very low number of true beliefs. For this reason, we may try to invoke a

combination of reliability and power, where the latter corresponds to the number of true

belief produced, given a set of circumstances. While an improvement, a further problem

arises, to the effect that a practice may be extremely reliable and powerful without

answering any questions whatsoever, as in a case where it only generates high ratios and

numbers of true belief that are of no relevance whatsoever to the concerns of the agent.2

A more promising strategy is, therefore, to specify the relevant notion of power

further in terms of question-answering power.3 In other words, what is epistemically

relevant as far as power is concerned, is not that the agent arrives at a large number of

true beliefs per se, but that she arrives at a large number of true beliefs that answer

questions she wants answered.4 In light of this, consider the following triad of definitions,

providing a first stab at what constitutes epistemic betterness, as it applies to practices:

2
Cf. Alston (2005) and Goldman (1986).
3
See Goldman (1986, p. 123).
4
This is a slight departure from Goldman’s formulation in terms of “the proportion of
questions [a cognitive system] wants to answer that it can answer (correctly)” (1986, p.
123; emphasis added). (See Goldman 1992a, p. 195 for a similar formulation in terms of
practices.) However, note that a formulation in terms of proportions of questions that the
subject wants answered would imply that a practice can be rendered less powerful (for

110
 (B) A practice P1 is epistemically better than another practice P2 for a subject S

iff P1 outperforms P2 for S, where

(O) a practice P1 outperforms another practice P2 for S iff P1 yields (i) a higher

ratio and number, (ii) an equal ratio but a higher number, or (iii) a higher

ratio but an equal number of beliefs that are significant to S, and

(S1) a belief is significant to S iff it answers questions on S’s research agenda.

For ease of reference, I will henceforth refer to any combination of definitions by way of

their abbreviations, together with a ‘+.’ For example, I will refer to the combination of

above definitions as ‘(B)+(O)+(S1).’ (Definitions clearly assumed but not at issue will be

suppressed for the sake of brevity.) The above triad of definitions raises a number of

questions that will be addressed in the remainder of the section, where they will also be

refined and qualified in a number of ways. First off: What is a research agenda?

4.1.1. Questions and Informative Answers

A research agenda is, to a first approximation, a non-ordered set containing those

questions the pursuit of which the relevant inquirers consider most worthy of their limited

epistemic resources. A couple of comments are at place. First, the reasons for specifying

the set in terms of the questions deemed most worthy of pursuit will be discussed in due

course (see §4.2.1), and may be put to the side presently. Second, the questions featured

that subject) simply by increasing the number of questions asked, while a formulation in
terms of numbers would be sensitive to the factor that seems more relevant, i.e., the
numbers of questions answered.

111
on the agenda do not need to be explicitly formulated by the inquirer. A research agenda

may simply be the most plausible reconstruction of the questions that the inquirer deems

most pressing, as revealed by the choices that she makes in allocating her resources.

Third, while there is, clearly, legitimate room for a notion of an agenda as applicable to

an inquirer because an institution or other pertinent body considers the relevant set of

questions worthy of her limited resources, I will reserve the discussion of such an

extension of the present notion for another occasion.5

For present purposes, we may assume that there are two broad classes of reasons

for pursuing some questions rather than others. On the one hand, there is the class of

reasons that we may ascribe to curiosity, or a desire for an answer to a question for the

mere purpose of attaining that answer. I will not delve into the nature or source of this

desire, but merely take it as a psychological fact that there is such a thing as a “sheer

intellectual curiosity” (Hempel 1965, p. 333) that, to paraphrase Larry Laudan (1977, p.

225), seems to be every bit as compelling to many of us as our need for clothing and

food. On the other hand, there is also the class of reasons that we may ascribe to a desire

to attain answers for some other sake than the mere attainment of those answers, be it for

moral, practical, or some other kind of reason.

However, what is important to note is that, regardless of whether an inquirer

pursues answers to questions because of curiosity or for some other reason, she will not

necessarily consider every true answer equally worthy of pursuit. To illustrate this point,

consider two social practices, P1 and P2. P1 generates x answers to questions on the

relevant inquirer’s agenda at a truth ratio of y within the relevant circumstances, while the

5
See Goldman (1999, p. 94) for a relevant discussion.

112
corresponding values for P2 are x + n and y, where x, y, and n > 0. Moreover, while P1

generates at least some informative answers, P2 exclusively generates answers that

constitute utterly uninformative answers by all plausible standards. Still, since

(B)+(O)+(S1) makes no distinction between informative and uninformative answers, we

would have to say that P2 is epistemically better than P1 (for the inquirer in question).

Needless to say, that would not make for particularly good advice from the point of view

of amelioration.

I do not have a theory about what constitutes sufficiently informative answers, but

I am also not so sure that we need one. To make this point, I will, first, consider an

illustration of the relevant notion and, then, suggest that we have a sufficiently good

intuitive sense of paradigmatic cases of uninformative and sufficiently informative

answers, respectively, for us to not require a theory. By way of illustration, consider the

following question:

(Q) “What caused x?”

We can imagine one answer that lists the actual cause(s) of x. Then we can imagine

another answer: “Something.” Both answers are true and on topic, but there is a sense in

which the latter is genuinely unhelpful. Returning to the two classes of reasons for

pursuing answers to questions introduced above, this might be so for either of two

reasons. Assuming that we are interested in finding an answer to (Q) for some other

reason than mere curiosity, e.g., if x constitutes a malfunction that we want to correct and

prevent in the future, an answer in terms of “something” is unhelpful because it tells us

113
nothing whatsoever about how to go about doing this. Similarly, if we are interested in

(Q) as a result of curiosity, the mere fact that we are simply curious about an answer to

(Q) does not imply that we would accept just any true answer as a sufficiently

informative answer. For example, someone pondering what gave rise to some ancient

event may do so without having any practical (or moral, or prudential, or what have you)

stake in the answer—maybe she just finds thinking about these kinds of things

intellectually stimulating. Still, her curiosity might, nevertheless, only be satiated given

that the answer is sufficiently informative. Indeed, had she not been curious about the

question, she might have settled for a completely uninformative answer, like

“something.”

In other words, the intuitive notion of an answer being sufficiently informative is

roughly that of the answer actually answering the question, in a sense that goes beyond

saying something that is both true and on topic. Interesting formal work has been done on

specifying the exact features that answers are typically required to have (beyond that of

being true and on topic) for a variety of question types.6 However, more relevant than the

details of the relevant literature—details that, for present purposes, would add formal

detail but not necessarily explanatory value—is the kind of data invoked in the literature,

i.e., intuitive judgments of ordinary speakers regarding what answers constitute

sufficiently informative answers, given a particular question. In line with this

methodology, I will characterize what it is for an answer to be sufficiently informative in

operational terms as follows:

6
See, e.g., Belnap and Steel (1976), as well as Jaworski (2009) for a recent application of
Belnap and Steel’s framework to certain kinds of how-questions.

114
 (A) an answer a is a sufficiently informative answer to a question q iff (i) a is

true, and (ii) any normally functioning person that grasps the contents of

both a and q would accept a as an answer to q, provided that she takes a to

be true.

This might be more of a clarification than a definition. For one thing, a completely

satisfactory definition would need to spell out the relevant notions of normal functioning,

content and what it is to grasp content. That being said, I believe that (A), nevertheless, is

sufficient for present purposes, given our intuitive ability to provide reasonable verdicts

at least about paradigmatic instances of sufficiently and insufficiently informative

answers, respectively. Moreover, against the background of (A), we may reformulate (S1)

as follows:

(S2) A belief is significant to S iff it constitutes a sufficiently informative answer

to questions on S’s research agenda.

(B)+(O)+(S2)+(A) yields the verdict that P1 is epistemically better than P2 in the above

example, and, thereby, constitutes an improvement on (B)+(O)+(S1).

4.1.2. Problems of Individuation

(O) invokes the notions of ratios and numbers of belief. This seems to be presupposing

that beliefs can be individuated in systematic ways. However, any given belief can be

described at a number of degrees of generality. For example, should my belief “The sun

115
is shining right now” be distinguished from my belief that “The sun is shining on

September 14th, 2009”? Or from my belief that “The sun is shining in New Brunswick on

September 14th, 2009”? It is hard to say. More than that, in want of a principled account

of what degree (or degrees) of generality should be privileged, there is a potential

problem about double counting and about not counting what should be counted. For this

reason, one may have one’s doubt about the usefulness of talking about ratios and

numbers of belief. These doubts may be intensified by the fact that we, on the present

account, need to individuate ratios and numbers of significant (true) beliefs, i.e., beliefs

that constitute acceptable answers to questions on the relevant agenda(s). Consequently,

we not only need to individuate beliefs, but also answers and agenda items.

In response to these worries, it should be noted that talk about ratios and numbers,

albeit convenient for the purposes of thinking clearly about the relevant issues, has as

Andrew Latus (2000) puts it a “misleading air of precision,” as far as the evaluation of

actual practices is concerned. More specifically, what I would like to suggest is that

epistemic evaluation—at least as it concerns us here—does not require a fully systematic

account of belief, answer, and agenda item individuation. It only requires that we can

approximate the relevant magnitudes to an extent sufficient for making informative and

roughly accurate comparative judgments.

Let us start with the individuation of belief. As just noted, the kinds of claims we

have to be able to make on the present account are comparative claims. More than that,

the kind of comparative claims that concern us here are explicitly ordinal claims.

Consequently, when comparing two practices, the only thing required is that we are able

to make comparative claims about one practice yielding more significant true beliefs than

116
another, or one practice yielding a greater preponderance of significant true belief than

another. What I would like to suggest is that the kinds of problems coming out of dangers

of double-counting and not counting what should be counted are not on a scale that

undercuts the possibility of making such ordinal claims. And given that we do not need

anything beyond such claims to make sense of the comparisons involved in (O), the

present account does not fall prey to a problem of individuation, as far as the

individuation of beliefs is concerned.

A similar point can be made in relation to the individuation of answers. It was

suggested above that we have a fairly robust intuitive sense of what constitutes

sufficiently informative answers. Moreover, as already stressed, the complications

associated with individuation would have to be severe enough to make impossible even

informed approximations in the form of comparative, ordinal claims about the relative

merits of two practices. In light of this, an answer similar to the one given above in

relation to belief individuation may be given also in the case of answer individuation.

More specifically, I submit that we can approximate the relevant magnitudes and

preponderance of sufficiently informative answers to an extent sufficient for making

informative and roughly accurate comparative judgments, even in the absence of a fully

systematic account of how to individuate acceptable answers.

Consider, finally, the individuation of agenda items. Any given inquirer may want

a whole host of different questions answered—questions that, moreover, may be more or

less interconnected and of different degrees of generality. For example, consider a cancer

researcher. At the most general level, a cancer researcher may be interested in the causes,

treatment and/or prevention of cancer. With respect to each general area, the researcher

117
may, in turn, be interested in a series of more specific questions. For example, with

respect to the causes of cancer, the researcher may be interested in identifying common

types of mutated genes that tend to be causally implicated in turning normal cells into

cancer cells. Beyond that, the researcher may be particularly interested in a specific sub-

set of such genes, perhaps because understanding the nature of that sub-set has proven to

be particularly important in understanding common forms of tumor progression, which,

in turn, may have implications for the success of different forms of treatment and

strategies for prevention. So, how many items figure on her agenda?

It goes without saying that any numerical judgment made in this context is going

to be largely arbitrary. Still, it is likely that different disciplines have developed certain

rough standards for how to group different questions into hierarchies—hierarchies that

show up in everything from grant applications to conference themes, anthologies, and

calls for papers. This is not to suggest that these hierarchies are, in any way, cutting the

space of taxonomical possibilities at its natural joints (if there even are any such natural

joints). Still, it is likely that we can expect a certain degree of homogeneity within

disciplines with respect to how inquirers will characterize their agendas. Moreover, if we

can expect there to be such hierarchies, and we see to it that we focus on the same level

of any given hierarchy in comparing how well two practices do with respect to answering

the questions contained on that level, then the possibility that every form of inquiry can

be described with reference to a great multitude of different questions does not seem to

threaten this particular use of the notion of a research agenda in evaluation.

118
4.1.3. Shared Agendas

So far, we have been concerned with what it is for one practice to be epistemically better

than another for an individual. As for what it is for one practice to be better than another

for a group of individuals, consider the following reformulations of (B), (O) and (S2):

(BG) A practice P1 is epistemically better than another practice P2 for a group of

subjects G iff P1 outperforms P2 for G, where

(OG) a practice P1 outperforms another practice P2 for G iff P1 yields (i) a higher

ratio and number, (ii) an equal ratio but a higher number, or (iii) a higher

ratio but an equal number of beliefs that are significant to the subjects in G,

and

(SG) a belief is significant to G iff it constitutes a sufficiently informative

answer to questions on G’s shared research agenda.

As these definitions make clear, what it is for one practice to be epistemically better than

another for a group is just a matter of aggregating the relevant ratios and numbers for the

group’s individuals, on the crucial assumption that the individuals in question share a

research agenda. In the interest of simplicity of exposition, I will stick to talking in terms

of epistemic betterness for an individual subject, and leave to the reader to draw the

analogous implications for groups of subjects in accordance with (BG)+(OG)+(SG), in

contexts where she deems that appropriate.

However, note that none of this is to suggest that we cannot imagine comparative

epistemic measures across agendas. For example, a measure that can be extracted from

119
the present account fairly easily is the following: A practice P1 is more robust than

another practice P2 iff P1 outperforms P2 for a wider range of agendas than the range of

agendas within which P2 outperforms P1. Note, however, that one practice being more

robust than another does not imply that the former is epistemically better than the latter,

for the simple reason that epistemic betterness is defined with reference to a single

agenda. This mirrors the intuition that a practice might be epistemically better than

another given a very specific set of questions, without, thereby, being successfully

applicable to a whole host of other questions, as is surely the case with several practices

within the sciences that involve very specialized forms of methods, practices, and

instruments. We will return to this issue briefly in §5.3 below. Beyond that, I will not

have anything to say about comparative epistemic measures that may apply across

agendas.

4.1.4. On Incomparability

On (O), one practice outperforms another if and only if the former yields (i) a higher ratio

and number, (ii) an equal ratio but a higher number, or (iii) a higher ratio but an equal

number of true belief that are significant to S than the latter practice does. But what about

scenarios in which none of (i) through (iii) holds? By mapping out all the nine (i.e., 32)

possibilities, we see that there are easy as well as hard cases. More specifically, consider

the following table, representing the comparison of one practice, P1, with another

practice, P2, where ‘+’ represents P1 receiving a higher score than P2 on the relevant

factor, ‘–‘ represents P1 receiving a lower score than P2 on the relevant factor, and ‘=’

represents P1 receiving the same score as P2 on the relevant factor:

120
 REL PWR
1. + + B
2. = + B
3. + = B
4. = = E
5. + – I
6. – + I
7. = – W
8. – = W
9. – – W

Rows 1 through 3 cover cases in which (i), (ii), or (iii) hold, and where P1, consequently,

is epistemically better than P2. Row 4 covers a case in which P1 and P2 are equally

epistemically good, and rows 7 through 9 cases in which P1 is epistemically worse than

P2, given the following two definitions:

(E) Two practices, P1 and P2, are equally epistemically good for a subject S iff

P1 and P2 yield equal ratios and numbers of beliefs that are significant to S.

(W) A practice P1 is epistemically worse than another practice P2 for a subject S

iff P1 yields (i) an equal ratio but a lower number, (ii) a lower ratio but an

equal number, or (iii) a lower ratio and a lower number of beliefs that are

significant to S than P2 does.

This leaves us with rows 5 and 6, involving cases where one of the two practices fares

better on one score, while the other fares better on the other score. What are we to say

about these cases? Consider the following:

121
 (I) Two practices, P1 and P2, are epistemically incomparable for a subject S iff

neither (B), (E) or (W) holds between P1 and P2 for S.

This way of understanding incomparability is fairly standard in the literature.7 On (I), the

cases covered by rows 5 and 6 are instances of epistemic incomparability. That is, while

one practice is epistemically better with respect to power, and one with respect to

reliability, there is no additional covering consideration in light of which it can be

inferred that one is epistemically better than the other simpliciter. Consequently, in these

cases of incomparability, there are no epistemic reasons to prefer one of the practices to

the other.

However, it should be noted that this is perfectly compatible with there being a

whole host of non-epistemic reasons to prefer one practice over another, even in a

scenario where they are incomparable from an epistemic point of view. For example, we

can easily imagine cases in which avoiding error is more important than arriving at many

true answers, due to the fact that the moral or practical costs of a mistaken judgment

about the matters at hand far outweighs the benefits of being able to make a judgment. In

such cases, reliability might trump power, and P1, hence, be preferable on non-epistemic

grounds to P2 in the scenario represented by row 5, despite the two practices being

epistemically incomparable. Similarly, we can imagine cases in which power trumps

reliability, as in contexts of intellectual exploration or speculation where it is more

7
See, e.g., Chang (1997). This is not to say that the idea that this is the proper way to
understand incomparability is unchallenged. For example, Chang (2002) argues for the
existence of a fourth relation (beyond better, worse, and equally good), i.e., parity—a
relation that I will not consider presently.

122
important that an answer is generated than that the proportion of false answers to total

answers generated is minimized. In such cases, P1 would be preferable on non-epistemic

grounds to P2 in the scenario represented by row 6.

Moreover, such paradigmatically non-epistemic considerations can not only

provide relevant covering considerations in cases of epistemic incomparability, but also

in cases of epistemic comparability. In the latter cases, however, such considerations are

not invoked in the absence of any epistemic reasons to opt for one practice over the other,

but in the presence of epistemic reasons that are overridden by non-epistemic

considerations. Still, qua non-epistemic considerations, they fall outside the scope of the

present investigation, beyond the extent to which they may serve to account for the

intuition that (a) epistemic incomparability is compatible with having reasons (as in: non-

epistemic reasons) to prefer one practice over another, and that (b) the epistemic reasons

present in cases of epistemic comparability may be overridden by non-epistemic

considerations. We will find reason to return to the latter point in the next chapter (in

§5.4).

4.2. On the Specialized Role of Epistemic Evaluation

Let us take stock at what has been argued so far. The account of epistemic betterness

developed above may be summed up as follows:

(B) A practice P1 is epistemically better than another practice P2 for a subject S

iff P1 outperforms P2 for S, where

123
 (O) a practice P1 outperforms another practice P2 for a subject S iff P1 yields (i)

a higher ratio and number, (ii) an equal ratio but a higher number, or (iii) a

higher ratio but an equal number of belief that are significant to S, and

(S2) a belief is significant to S iff it constitutes a sufficiently informative answer

to questions on S’s research agenda, where

(A) an answer a is a sufficiently informative answer to a question q iff (i) a is

true, and (ii) any normally functioning person that grasps the contents of

both a and q would accept a as an answer to q, provided that she takes a to

be true.

On this account, epistemic betterness is defined for practices, given a research agenda.

The account, thereby, tells us something about what inquirers should do, in so far as they

want to arrive at belief that answer their questions. In the remainder of this chapter, I will

address two concerns. First, while answering any questions on our agendas, clearly,

might be a good thing from an epistemic point of view, might it not be epistemically

better to answer some questions rather than others? In particular, might it not be

epistemically better to answer questions that we are more interested in than questions that

we are less interested in (ceteris paribus)? Second, what epistemic judgments can we

make with respect to what items should figure on agents’ agendas in the first place? Let

us consider each question in turn.

124
4.2.1. Degrees of Interest

According to (S2)+(A), it is a necessary and sufficient condition on a true belief being

significant to an inquirer that the belief (or, to be exact: its content) constitutes a

sufficiently informative answer to questions on her agenda. As we have seen, this makes

significance agent-relative in the sense that what true beliefs are significant is, in part, a

function of the questions posed by the inquirer. At the same time, significance is on the

present account an objective matter, at least in the following sense: A belief b1 is not

necessarily more significant for an agent than another belief b2 simply because she

considers the questions answered by b1 more interesting than the answered by b2.

This might be taken to present a problem. To see why, consider a scenario

discussed recently by Don Fallis (2006), in which there are two practices, P1 and P2,

available to an agent.8 P1 yields as its only output one believed proposition that answers

one question that the inquirer takes great interest in, viz. the question of how the world

could be made a good and happy place, while P2 yields as its only output two believed

propositions that answer two questions the inquirer takes some (but less) interest in, viz.

the questions when Hill Street Blues and Streets of San Francisco come on TV,

respectively. Assuming that the inquirers take sufficient interest in all three questions for

them to be featured on her research agenda, and that other things are equal—e.g., that the

believed propositions answer no more than one question each—it follows on the present

account that P2 is epistemically better for the subject than P1.

8
Fallis formulates his argument in terms of sets of doxastic states rather than in terms of
the processes or practices that generate these sets. In order to make his argument
applicable to the present account, however, I have reformulated it in terms of practices.

125
 Is this a problem? To get a better grip on the issues relevant to answering this

question, let us consider Fallis’ suggestion as for an alternative account, on which the

epistemic value of forming a true belief is weighted by the degree of interest that the

agent has in the truth in question.9 Worries about how to properly quantify degrees of

interest aside, Fallis’ framework would, thereby, enable us to say that P1 is epistemically

better than P2, due exactly to the fact that the inquirer deems the question answered by

way of P1 more interesting than the questions answered by P2. Fallis introduces his

account—as well as the above scenario—in a criticism of Goldman (1999). According to

Goldman, epistemic evaluation is restricted to epistemic performances that pertain to

questions that interest the inquiring agent, in much the same way as epistemic evaluation

on the present account is restricted to questions that are featured on the relevant inquirer’s

research agenda.10 As such, the presence of interest, clearly, plays a role on Goldman’s

account, even though degree of interest does not.

Goldman opts for an account in terms of the presence as opposed to degree of

interest in an attempt to balance two desiderata. On the one hand, he acknowledges that

“a good epistemic practice should not be indifferent to the relative amounts of interest

that a given agent takes in different questions” (Goldman 1999, p. 95). On the other hand,

he wants to settle interest with a modest as opposed to a pervasive role in evaluation,

where the latter “might go so far as to use the degree of interest in a question to give an

overall weighting to an agent’s veritistic [i.e., truth-related] success” (p. 95). Goldman

sees two problems with letting interests play a pervasive role in evaluation:

9
See Fallis (2006, pp. 181-182).
10
See Goldman (1999, pp. 88 and 94-96).

126
 First, one person’s questions of interest might be easier to answer than those of a
second. The fact that the first scores higher on the set of questions that interest her
does not demonstrate superior intellectual skill. Second, the intensity of interest a
person takes in a question may reflect factors that do not properly belong in an
epistemological analysis. The magnitude of interest may depend on the person’s
financial or mortal stake in the question. But that does not mean that his
succeeding or failing to answer those questions correctly should loom
disproportionately large in an epistemic analysis. This would lead down the path
toward abandoning the specialized, veritistic mission of epistemology in favor of
a more purely pragmatic enterprise (Goldman 1999, p. 95).

However, note that Goldman’s two objections to providing degrees of interests with a

pervasive role in epistemic evaluation apply even if reformulated in terms of the presence

(as opposed to degrees) of interest. As for easy answers, it is sufficient that a person only

is interested in easy questions for an epistemic evaluation in terms of the generation of

true belief to be skewed in her favor, when compared to someone that is only interested

in hard questions. (This issue will be discussed in more detail below.) As for intensity of

interest, just like a person’s magnitude of interest may depend on the her financial or

mortal stake in the question, so, too, for her having any interest in the question.

Consequently, it does not seem that Goldman’s objections serve to favor giving degrees

of interest a modest rather than a pervasive role—which is, I take it, exactly Fallis’ point.

Where does this leave the present account? Unlike Goldman, I have no quarrel

with Fallis’ account per se. However, within the context of the present framework,

weighting the significance of answering agenda items by the degree of interest in the

corresponding question would not so much amount to an unwelcome as to a somewhat

unnecessary feature. The reason is that we do not imagine the typical research agenda as

one that contains both (what tends to be considered) paradigmatically pressing concerns,

such as those pertaining to the variety of projects that concern the scientific and political

127
community, and questions about when different TV shows come on TV. More to the

point, we are imagining that the span of degrees of interest that we have for the different

items on a particular agenda is far more narrow than in the kind of scenario Fallis

imagines, for the simple reason that what ends up on the research agenda are the

questions we deem most interesting out of all the questions that we may choose to pursue.

Consequently, it is not that degree of interest plays no role in the present account. It does,

but rather than playing a role by way of weighting, it plays a role in that agenda items

correspond to those—and only those—questions the pursuits of which we consider most

worthy of our limited epistemic resources.

4.2.2. Banal and Harmful Pursuits

Understanding agendas in terms of that which we consider most worthy of pursuit,

however, should not be taken to put any substantial restrictions on what inquirers may

deem most worthy of pursuits. So, let us turn to the question of what epistemic judgments

we can make with respect to what items should figure on agents’ agendas. For one thing,

we can often provide people with epistemic reasons to pursue at least some challenging

questions, if doing so would serve to sharpen their cognitive tools and, thereby, enable

them to answer whatever questions already figure on their agenda with a greater

efficiency. For another, we can often also make a case for engaging with specific

questions (be they challenging or not), since doing so may enable us to address already

existing agenda items in more productive ways. By the same token, failing to ask certain

new questions may have (comparatively) detrimental effects on the extent to which the

inquirer will be able to answers questions already on her current agenda. In such cases,

128
the inquirer has straightforwardly epistemic reasons for expanding or otherwise revising

parts her agenda, in so far as doing so will make her epistemically better off with respect

to answering the questions she deems most worthy of pursuit.

But what normative, epistemic judgments (if any) can we make beyond this? We

may use this question to put pressure on the point made above in relation to Fallis’

account, where it was suggested that agendas typically do not involve both questions

about what most people would consider paradigmatically pressing concerns and fairly

banal questions about TV shows. That this is so may be granted by Fallis, but the

question remains: What if such a span of topics actually happened to be considered most

interesting by an inquirer, however atypical she, thereby, would be? By way of

illustration, consider a subject who can choose between utilizing two practices, P1 and P2.

P1 would enable her to answer a very small number of questions that figure on her agenda

about the causes, treatment, and prevention of cancer. P2, on the other hand, would

enable her to answer a great many questions, also figuring on her agenda, about a series

of utterly banal matters, primarily pertaining to the plots of a couple of soap operas that

she follows.11 Let us also assume that there is no overlap in the questions answered by the

two processes, and that the questions answered by P2 are so banal that no pursuit of

challenging questions would make any difference to her ability to answer them. If so, the

account defended here implies that P2 is epistemically better than P1 for the subject.

Does this amount to a reductio of the present account? I do not think that it does.

There are a number of ways in which the subject in question would be better off by

opting for P1. From a moral perspective, the questions answered by P1 have a potential to

11
Feldman (2001) invokes a similar scenario in a critique of Goldman (1999).

129
help people suffering from serious and often terminal diseases. From the perspective of

personal well-being, it might be that, if the subject is anything like most of us, engaging

with the questions addressed by P1 would provide her with a deeper sense of purpose and

intellectual stimulation than engaging with the questions addressed by P2 does. And so

on. However, in developing a notion of epistemic betterness, we are interested in a

specialized form of evaluation, namely the kind of evaluation that—if the observations

made above are on the right track—pertains to the generation of high ratios and numbers

of beliefs that answer whatever questions the relevant inquirers happen to be posing. As

such, it is to be expected that the present account will not capture aspects relevant to

other forms of evaluation, such as those just mentioned. Consequently, despite the fact

that opting for P2 makes the subject inferior in all aforementioned respects, she is not,

thereby, epistemically worse off, and suggesting that she is would simply amount to

exaggerating if not misconstruing the scope of epistemic evaluation.

A similar point can be made in relation to considerations about the epistemic

evaluation of harmful pursuits. By way of illustration, consider a scenario exactly like the

one above, the only difference being that the questions answered by P2 pertain to how to

bring about the maximum amount of gratuitous pain and suffering in the maximal amount

of people. For reasons perfectly analogous to those just discussed, the present account

would imply that the subject is better off by opting for P2 rather than for P1. Is this a

problem? In answering this question, it might be fruitful to consider an account on which

this would not follow. Recently, Michael Bishop and J. D. Trout (2005) have suggested

that “the significance of a problem for S is a function of the weight of the objective

reasons S has for devoting resources to solving that problem” (p. 95). As for the specific

130
nature of such objective reasons, they make clear that their “general account of epistemic

significance resides, ultimately, in judgments about what conduces to human well-being”

(p. 99). Bishop and Trout do not say explicitly whose well-being is relevant to whose

objective reasons. However, most of their comments—such as that the “priorities of the

excellent reasoner (and more generally, of the wise person) are set so that they may serve

as a means to human flourishing” (p. 101)—suggest that the objective weight of S’s

epistemic reasons is a function of increases in the amount of general well-being in the

world, rather than simply in the particular well-being of S.

Incorporating this notion of significance in the present account of epistemic

betterness would imply that the subject would be epistemically better off by opting for P1

rather than for P2. However, there are two problems with forging such a strong

connection between significance and well-being. First, what is the normative force of this

connection? Is it that we ought to only engage in endeavors that are highly conducive to

general well-being? If this is the idea, the account seems to not only promote an

extremely skewed division of cognitive labor, but also render the majority of what makes

up current scientific (not to mention philosophical) research insignificant. In fact, even a

more lax construal of conduciveness would be problematic, since there are substantial

sets of questions and problems that lack any connection to the promotion of well-being.

For example, is it always possible to find, somewhere in the infinity of integers, a pro-

gression of any length of equally spaced prime numbers? Are there three, four or eleven

dimensions to our universe? It is not clear that either of these questions may be related to

131
matters of human-well being, or that this ought to imply that the corresponding beliefs

should not be considered epistemically significant.12

That being said, it might be possible to rework Bishop and Trout’s account in a

way so that the normative force does not flow from an injunction to maximize well-being,

but from a desire to ensure that inquiry is not detrimental to well-being. Still, a

fundamental point remains, which brings us to the second problem: While there is,

clearly, something to be said for the fact that the subject should not opt for P2, it is not

obvious that she should be considered epistemically worse off by doing so. As noted

above, epistemic evaluation is concerned with evaluating practices along a specifically

epistemic dimension—a dimension that, I have argued, should be understood in terms of

beliefs the contents of which answer whatever questions happen to figure on the

inquirer’s agenda. As such, we should, as already noted, expect there to be aspects of

betterness not captured by the present framework, i.e., non-epistemic aspects. However,

rather than indicating a flaw in the present account, the fact that it does not account for

such aspects simply suggests that it does not account for more than it should, qua an

account of epistemic betterness.

12
Could the account be saved by appeal to the increase in well-being that may be
experienced by the researchers engaged in these ventures (ceteris paribus)? This would
imply that, if everyone engaging in foundational issues in mathematics, natural science,
and metaphysics (among other areas) ceased to experience any degree of well-being in
their pursuits, and no one else started doing so for them, the corresponding beliefs would
cease to be significant—and for that very reason. I suspect that this would give us a
notion of significance very different from what Bishop and Trout are looking for.

132
4.3. Conclusion

The previous two chapters argued that true belief is the only thing of non-instrumental

epistemic value. However, it does not follow from true belief being the only good of non-

instrumental epistemic value that all true beliefs are valuable thus. More specifically, the

present chapter argued that the epistemic value of a practice is a function of the extent to

which it yields beliefs that answer questions posed by relevant sets of inquirers. As such,

the account tells us something about what methods or practices inquirers should utilize, in

so far as they want answers to their questions. The remainder of the chapter was primarily

concerned with tracing the limits of this ‘should,’ and it was suggested that there are very

real limits to the extent to which we can make epistemic judgments about what questions

inquirers should pose, as opposed to about what practices to opt for, given a set of

questions. However, rather than indicating a flaw in the account, this simply suggests that

it is able to account for the specialized nature of epistemic evaluation.

133
 CHAPTER 5

EPISTEMIC AGENCY AND PATERNALISM:

WHEN LESS IS MORE

5.1. Reasons for Mandating Compliance

When conducting inquiry we do a variety of things, such as gather information, mull over

our data, and choose among different methods of investigation—all for the purpose of

attaining our epistemic goal. In so doing, we are expressing our epistemic agency, as was

suggested in chapter 1. Chapters 2 and 3 provided an account of the relevant goal, i.e., the

goal of believing truly, and chapter 4 developed a theory about what it is for some

practices to be epistemically better than others with respect to that goal, in terms of how

good of a job they do in generating beliefs that answer questions posed by the relevant

inquirers. The purpose of the present chapter is to illustrate how we may use the

resulting notion of epistemic agency to say something constructive about how the ways in

which we actually express our agency may be brought in line with how we should

express our agency.

To this end, consider the fact that it, intuitively, might seem that, the greater our

freedom to express our epistemic agency, the better. At the very least, it seems prima

facie plausible that gathering more information, spending more time mulling over our

data, or choosing from a wider selection of methods cannot be a bad thing from the point

of view of forming true belief with respect to whatever matters happen to be under

investigation. However, the present chapter argues that, when it comes to our freedom to

134
express such agency, more is not always better. For example, it will be argued that we

often are better off by gathering only a very limited amount of information, having our

selection of methods be greatly restricted, and spending our time less on reflecting than

on simply reading off a market price or the output of a simple algorithm. In fact, we are

in many situations better off from the point of view of attaining true belief by having our

agency be constrained in substantial ways, even if against our will.

The case for this claim starts out with two psychological observations. First, as

noted already above, it is well known that we suffer from a variety of cognitive biases

when reasoning under uncertainty.1 In the psychological literature, a cognitive bias

typically designates a systematic tendency to form judgments in a manner that fails to

accord with principles that (given certain normative assumptions) may be gleaned from

logic, probability theory, statistics, and related disciplines. Understood thus, certain

biases may very well be adaptive, and, as such, yield accurate judgments in many

situations, despite being arrived at by way of heuristics that fail to accord with

aforementioned principles.2 In light of this, a more useful way to understand what it is for

something to be a bias is in terms of systematic tendencies to form inaccurate belief—be

it due to a failure to form belief in accordance with principles gleaned from logic,

probability, statistics, etc., or to substantial differences between the cognitive challenges

that faced our evolutionary ancestors and the cognitive challenges that face us today.

Second, any attempt at de-biasing—i.e., any attempt to either correct or sidestep

our systematic tendencies to form inaccurate beliefs through corrective or substitutive

strategies, respectively—has to take into account another fact touched upon above,

1
See Gilovich et al. (2002).
2
See, e.g., Gigerenzer et al. (1999).

135
namely that each and every one of us tends to underestimate the extent to which we, as

opposed to everyone else, actually are prone to such biases.3 This is relevant since there

are (at least) two ways in which an attempt at de-biasing may fail. On the one hand, it

may fail to improve our reliability (within the relevant domain), even if the methods

prescribed are utilized properly. On the other hand, the methods prescribed may fail to be

utilized at all. Let us refer to the latter possibility as involving problems of defection.

Problems of defection have received ample attention within the literature on

predictive modeling in connection with the so-called “broken leg problem.”4 The

problem is illustrated with reference to an imagined prediction model (more on such

models below) that is highly reliable in predicting a person’s weekly attendance at a

movie, but that should be disregarded upon finding out that the person in question has a

fractured femur. There is, certainly, something to be said for being sensitive to

information not taken into account by whatever models one happens to be relying on in

prediction, especially in light of the plausible observation that no domain-specific method

will be reliable in all domains. The problem is just that, in a wide range of domains where

we rightly care about accuracy of judgment—including the domains of medical and

financial decision making—people tend to see far more broken legs than there really are,

and, thereby, end up defecting from reliable models far more often than they should, from

an epistemic point of view.5

Evidence suggests that that major culprit behind defection is that we are

systematically overconfident about the accuracy of our judgments. For example,

3
See Pronin et al. (2002).
4
See Meehl (1954).
5
See Dawes et al. (2002).

136
Whitecotton (1996) found a correlation between confidence ratings and defection from

models known to be highly reliable in an earnings-forecasting study with professional

financial analysts, as did Arkes et al. (1986) in a similar study. The suspicion that this

correlation consisted in a relation of causation has been substantiated more recently in

Sieck and Arkes (2005). Several studies have suggested that overconfidence is a very

recalcitrant phenomenon that is mitigated neither by accuracy incentives6 nor by simple

motivational declarations.7 What have been shown to reduce overconfidence to some

degree, however, are rigorous regiments of feedback.8 Building upon this fact, Sieck and

Arkes found that such feedback not only lowered people’s overconfidence in their

judgments in so far as they did not rely on the statistical models offered, but also led to

greater reliance on those models (with a resulting improvement in performance), as

compared to a control group. This provides us with reason to believe that overconfidence

is one (albeit not necessarily the only) cause of many cases of defection.

This has implications for how we should go about the business of de-biasing.

Since defection is the result of such a widespread phenomenon as overconfidence, we can

expect that a great majority of us will be prone to defection. Moreover, since defection is

due to a psychological feature of us, as opposed to a feature of the methods typically

invoked in attempts at de-biasing, we can also expect that a great majority of us will be

prone to defection, no matter the particular characteristics of the methods in question.

Granted, it might be suggested that the empirical results in question suggests one way to

come to terms with problem of defection, namely by focusing our attention more directly

6
See Arkes et al. (1986).
7
See Lord et al. (1984).
8
See Arkes et al. (1987).

137
on strategies that temper people’s confidence in their own accuracy. However, the

recalcitrant nature of overconfidence, together with the rigorousness of the feedback

schedule required to come to terms with it, renders the practical prospects of mitigating

the former by way of the latter somewhat dim.

For this reason, a different strategy will be pursued presently. More specifically,

since the above results suggest that a great majority of us will be prone to defection, no

matter the particular methods involved, the most promising approach to de-biasing will

be one that simply does not provide us with the option of not using the methods in

question. When the option of not using the relevant methods is removed thus, i.e., when

the inquirers, in effect, are left with no option but to utilize the methods in question, we

are mandating compliance with those methods. More specifically, I submit that the

aforementioned research on our dual tendency for bias and overconfidence makes

plausible the following:

(C) For every judgment domain D, such that we (a) care about accuracy in

judgment within D, and (b) have methods available that (if used properly)

have been shown to make those making judgments within D better off from

an epistemic point of view than do the methods involved in prevailing

practices, we have pro tanto reason for mandating compliance with the

former methods within D.

A couple of comments are in order. First, the question of what it is for a method to make

someone better off from an epistemic point of view will be postponed until the next

138
section, although the discussions in the previous chapter should provide substantial hints

as for how this locution will be spelled out. Second, no necessary and sufficient

conditions will be given for what constitutes a judgment domain. Instead, subsequent

discussions will rely on paradigmatic examples, such as political polling and medical

diagnosis and prognosis. While working with paradigmatic examples rather than clear-cut

definitions, clearly, will leave room for some degree of ambiguity, that will present no

problem presently, as long as it can be agreed that the domains in question contain many

examples of judgment tasks that we typically want to get right, no matter the particular

ways in which the domains may be further specified. The presence of such tasks is

crucial—and, moreover, provides the motivation for including (a) in (C)—since it is not

clear that the mere availability of epistemically beneficial methods gives us any reason to

implement them, unless we care about accuracy within the domain(s) in question. But

when we do care thus, and relying on the methods in question have been shown to make

us better off from an epistemic point of view as compared to the methods involved in

prevailing practices—as per (b) in (C)—we have pro tanto reasons to mandate

compliance.

The remainder of the chapter is concerned with spelling out the implications of

(C). §5.2 argues that the relevant mandate is epistemically paternalistic, and briefly

discusses a number of domains in which such paternalism is already practiced in society.

§5.3 shows that the precedents in question do not imply that a commitment to the relevant

kind of paternalism is trivial, by discussing two domains that satisfy (C) but where the

mandates in question are still to be imposed, and doing so would be potentially

controversial. §5.4 addresses the worry that, if we restrict ourselves to pro tanto

139
reasons—i.e., genuine but not necessarily decisive reasons, since the overall balance of

reasons may direct us to act differently—it might turn out that we have non-epistemic

reason not to impose paternalistic constraints on agency, even in cases where the

practices in question are epistemically beneficial. In response to this, it is argued that the

reasons involved in the cases considered actually amount to reasons all things considered.

The paper concludes in §5.5 that there, hence, are several cases in which we have not

only pro tanto but all things considered reasons for having our epistemic agency be

constrained on a wider scale than it currently is.

5.2. What is Epistemic Paternalism?

The idea that constraints on our agency can make us better off is not unique to the present

investigation. For example, Jon Elster (2000) mounts a persuasive case for agents often

being able to improve their lives in a variety of respects by precommitting themselves to

certain future actions, either directly or by severely restricting the number and kinds of

options that will be available to them at a future time. However, the mandate in (C)

should not be understood in terms of such self-binding. If it were only for the fact that we

suffered from cognitive bias, self-binding might have provided a viable option for de-

biasing. However, given our tendency to also underestimate the extent to which we are at

all prone to bias, there is simply no reason to believe that we—as in: each and every one

of us—will see any reason to bind ourselves to the methods in question (and even if we

did, we would subsequently consider ourselves having ample reason to break that

commitment, as in aforementioned cases involving defection from prediction models). In

140
short, due to our dual tendencies for bias and overconfidence, we cannot rely on

ourselves for doing what is best for us, from an epistemic point of view.9

For this reason, and as indicated above, a different and largely neglected approach

to de-biasing will be considered, involving epistemically paternalistic practices. One of

the few previous discussions of such practices is that of Goldman (1992b).10 Goldman

suggests that certain forms of information control practiced in society are motivated with

reference to how they protect us from our cognitive failings, but he neither discusses the

important role our tendencies for overconfidence play in motivating such protection, nor

dwells on the issue of defining the relevant kind of epistemic paternalism. Since we

discussed the former issue in §5.1, let us consider the latter. Judging by the vast literature

on paternalism in social and political philosophy, the conceptual core of paternalism

involves an interference with the liberty of another without her consent but for her own

good.11 In light of this, we may define what it is for a practice to be epistemically

paternalistic as follows:

(P) A practice is epistemically paternalistic if and only if it

(i) constrains inquirers’ freedom to conduct inquiry in whatever way they see

fit;

(ii) is implemented without the consent of those constrained; and

9
Elster (2000, ch. 2) does discuss other-binding, but only in terms of cases where the
binding in question is not for the benefit of those bound, as in the kind of de-biasing that
concerns us here, but simply for the protection of the binder(s).
10
However, see Sunstein and Thaler (2008) for a recent defense of what they refer to as
libertarian paternalism, invoked as a tool for ridding ourselves of a variety of behaviors
that are systematically detrimental to our well-being. See also Trout (2009 and 2005),
who will be discussed further below.
11
See Young (2008) and Dworkin (2002) for two overviews of relevant discussions.

141
 (iii) is prescribed for the purpose of making those constrained better off from an

epistemic point of view.

A couple of comments are at place. (i) is meant to capture the interference of paternalistic

practices. I will postpone further discussion of this aspect until we are considering real-

world cases below. (ii) is meant to capture the arrogance, if you will, of paternalistic

practices. Note, however, that this condition does not require acting against the consent

of those constrained. For example, the satisfaction of (ii) is compatible with the inquirer

in question appreciating the merits of the relevant methods and happily utilizing them.

What is crucial as far as (ii) is concerned is not that the inquirer is forced to do something

that she does not want to do, but simply that no effort has been made to seek her consent

on the issue.12

Condition (iii) is meant to capture what makes the relevant form of paternalism

epistemic.13 The kind of improvement relevant here can be defined in terms of the notion

of epistemic betterness developed in the previous chapter. More specifically, we may

think of what it is for someone to be made better off in the relevant sense in terms of how

the relevant paternalistic practice outperforms prevailing practices by mandating

compliance with methods that make her more reliable, without making her worse off in

some other epistemically relevant respect, e.g., with respect to question-answering power.

Understanding improvement specifically in terms of such pareto improvements in

reliability enables us to ignore cases of epistemic incomparability, as discussed in the

12
This is in line with Dworkin (2002).
13
It is compatible with a practice being epistemically paternalistic that it is motivated by
non-epistemic in addition to epistemic considerations.

142
previous chapter. To see why, consider two practices, P and P+, where the latter is a

pareto improved version of the former. If P+ is a pareto improved version of P, then P+

will better than P with respect to reliability, and equally good or better with respect to

question-answering power. In either scenario, P+ outperforms P, as defined in the

previous chapter.14

A further qualification is needed. In the previous chapter, we defined the relevant

reliability ratios not in terms of true belief per se, but in terms of significant true belief,

where a belief is significant if and only if it constitutes a sufficiently informative answer

to questions on the relevant agent’s research agenda. In light of this, we will need to

make two simplifying assumptions. First, since significance is a function of the relevant

agents’ research agenda(s), we will need to assume that those involved in the compared

practices are posing the same questions, at least as far as the relevant practices are

concerned. Second, we also need to assume that all agents involved function normally, in

that they would be able to identify sufficiently informative answers as such when they

grasp them. To make matters tractable, let us include these factors under a ceteris paribus

clause, and say that what it is for those constrained to be made better off from an

epistemic point of view is for the relevant paternalistic practice to yield a pareto

improvement with respect to reliability for those constrained, ceteris paribus.15

If we understand epistemic paternalism along the lines of (P), there does seem to

be practices already in place in society that would qualify as epistemically paternalistic.

14
The reader may return to the table in §4.1.4 to confirm that this is so.
15
Why not drop the pareto qualification and simply talk about an improvement with
respect to reliability, ceteris paribus? Because that would not count as improvements
cases in which a practice makes the relevant agents better off with respect to reliability
and power. Such improvements would constitute pareto improvements but not ceteris
paribus improvements with respect reliability.

143
Goldman (1992b) discusses four such practices, pertaining to (a) the control of legal

evidence on part of U. S. judges (particularly with respect to character evidence), (b)

curriculum selection in education, (c) the Federal Trade Commission’s actions against

false or deceptive advertisement, and (c) deliberate attempts within the news media to

simplify matters that are to be presented to the public. In the interest of adding to the

literature rather than simply repeating what has already been said, I will not dwell on

these practices. It suffices to note that they all serve to constrain peoples’ freedom to

conduct inquiry in whatever way they see fit by way of restrictions on information

access, without seeking their consent, for the purpose of making them better off

epistemically by protecting them from their own ignorance or bias in contexts where we

care about accuracy. As such, the practices in question are epistemically paternalistic, on

the above definition.

Any attempt to elaborate on the need for epistemic paternalism needs to answer to

a series of worries, three of which will be considered presently. Consider, first, the point

that the motivation for setting up epistemically paternalistic practices, as outlined above,

is a set of statistical facts about our cognitive tendencies. As such, there are likely to be

individual exceptions, in the sense that there will be some people who do not exhibit the

tendencies in question, even if the relevant statistics suggest that they will constitute a

minority. In light of this, the worry goes, is it really defensible to be paternalistic not only

with respect to the majority that is likely to benefit epistemically, but also with respect to

the few subjects who might either not be made better off, or that in some rare instances

144
perhaps even will be made worse off from an epistemic point of view for being imposed

upon thus?16

Two lines of response are available here. On the first one, we press the point that

the minorities are just that, i.e., minorities, and that the few that are not made better off or

that are made worse off, most likely, constitute statistical outliers. However, the problem

with this line of reasoning is that it seems to presuppose that, unless the minority is made

up of outliers, the relevant paternalistic practices would not be defensible. That seems to

be granting too much, which brings us to the second line of response: In mandating

compliance, the goal is not to make each and every one of those constrained better off; it

is simply to bring about a positive and hopefully significant epistemic net effect in the

relevant population of inquirers. Granted, the smaller the proportion of subjects that do

not exhibit the dual tendency of bias and overconfidence, the greater the net effect. But it

is sufficient for motivating mandated compliance with epistemically virtuous methods

that those who do exhibit the relevant tendencies are in a majority and we, consequently,

can expect there to be such a net effect, irrespective of whether those who do not exhibit

the relevant tendencies constitute outliers or simply a minority.

In so far as this response is deemed unsatisfactory, there might be a somewhat

different worry lurking in the background, pertaining less to the concern for maximizing

net epistemic effects among populations of inquirers, and more to a regard for the

integrity of the individual inquirers. This brings us to the second worry, according to

which epistemic paternalism is objectionable simply on account of being paternalistic.

The closest we come to a statement to this effect in the epistemological literature is in

16
Thanks to Adrian Staub for raising this question.

145
Larry Laudan’s (2006) treatment of evidence control in law. Laudan is explicitly

concerned with the epistemic aspects of this practice. Moreover, he describes it as being

paternalistic, and clearly intends the term as a pejorative. However, when we turn to his

case against the practice in question, we see that Laudan’s complaint is that restricting

jurors’ access to certain kinds of evidence (such as character evidence) “is not a

promising recipe for finding out the truth” (p. 25). That is, even if Laudan is right in

claiming that we have reason to be skeptical about the epistemic merits of prevailing

systems of evidence control—and the “if” is important, since Laudan, unfortunately, pays

no attention to the body of empirical literature concerned with how jurors process

character evidence (e.g., Hunt and Budesheim 2004; Lupfer et al. 2000)—his complaint

is not that the practice should be given up because it is paternalistic but because it does

not work, as far as enabling jurors to arrive at accurate verdicts is concerned.17

Evaluating existing paternalistic practices for their epistemic efficacy is an

important task. But it is not one that will be undertaken here, and for the reason just

outlined: an argument demonstrating that a particular practice is not fulfilling its intended

epistemic purpose is not an argument against paternalism; it is an argument for

implementing a different practice (paternalistic or not). For this reason, §5.4 will return to

a more promising way to spell out aforementioned worry, namely in terms of us having

non-epistemic reason not to impose paternalistic constraints on agency, even in cases

where the practices in question are epistemically beneficial. Before considering that

possibility, however, the next section will consider a third worry. The worry in question

pulls in a completely different direction from the ones just considered, which suggested

17
See also Goldman (1999), who raises concerns that are very similar to Laudan’s.

146
that a commitment to epistemic paternalism is, in a sense, too controversial simply in

virtue of being paternalistic. The third and final worry is that the fact that we already

practice epistemic paternalism in society, as evidenced by aforementioned examples,

indicates that a commitment to epistemic paternalism is not controversial enough, in that

it carries no substantial implications as for how we ought to set up our epistemic

practices.

5.3. Two Cases for Expansion

Above, we illustrated the idea of epistemic paternalism with reference to contexts in

which we already seem to practice such paternalism in society. This could be an

indication to the effect that a commitment to epistemic paternalism is already sufficiently

ingrained in the ways in which we set up epistemic practices for any attempt to argue that

we should be epistemically paternalistic to provide any reason to do things differently. It

is the burden of this section to argue that a commitment to epistemic paternalism is not

trivial in this sense. To make this point, it will be argued that a commitment to (C)

provides pro tanto reason for practicing epistemic paternalism on a wider scale than we

are currently doing. This point will be argued with reference to two domains clearly

satisfying (C), but where the relevant mandates are still to be imposed and the relevant

impositions would be potentially controversial.

147
5.3.1. Prediction Models in Diagnosis and Prognosis

In a recent treatment of the issue of de-biasing, J. D. Trout (2005) considers a variety of

“institutional assistance,” as he calls them, for coming to terms with our sometimes quite

biased ways. The kinds of assistances he considers include taking advantage of our

sensitivity to differences in framing by couching health care information in terms of gains

rather than losses in order to encourage preventive medical testing, changing driving

behavior by way of chevrons, and allowing employees to direct a portion of their future

salary increases toward retirement savings and thereby using our tendencies for inertia

and procrastination to our advantage.

Interestingly, however, Trout goes out of his way to argue that, since the measures

“assist the person in effectuating their long-term goals or plans, there is no relevant

interference, and so [such] regulation is not paternalistic” (p. 409).18 There is certainly

something to be said for the practices not amounting to paternalism, but not because they

promote our long-terms goals. After all, many paradigmatic instances of paternalistic

practices—e.g., the enforcement of laws for wearing a helmet when riding a

motorcycle—are motivated exactly on the basis that they promote our long-term goals.

What makes the practices Trout considers non-paternalistic is rather that no one is, in any

relevantly strong sense, constrained to do (or from doing) anything.

For present purposes, however, the kinds of assistance not considered by Trout

are more relevant than the ones he does consider. In particular, Trout does not discuss a

kind of assistance that he has discussed at length elsewhere (with Michael Bishop in their

2005), and where an illustrative and straightforwardly paternalistic case can be made. The

18
The same approach is taken by Trout more recently in his (2009, p. 210).

148
practice in question is that of using prediction models as an alternative to relying on the

judgments resulting from the unaided reflections of clinicians in medical diagnosis and

prognosis. The last fifty years of empirical research on predictive modeling has not only

demonstrated the superiority of the former over the latter, but also provided ample

evidence of inquirers being very reluctant to comply.19 Again, this has received some

attention within recent epistemological literature, primarily due to Bishop and Trout’s

discussion. However, what has not been acknowledged is that the literature in question

also provides pro tanto reason for mandating the use of such models in medical diagnosis

and prognosis—or so I will argue.20

The first thing to note here is that we typically care about accuracy in medical

diagnosis and prognosis, particularly given the potential human costs of inaccurate

judgment within these areas. As such, the relevant domains satisfy (C) (a). In light of

this, it should be considered good news that aforementioned models enable us to increase

our reliability significantly in matters of such diagnosis and prognosis, as compared to

reliance on unaided clinical judgments. Moreover, there is no reason to believe that such

a mandate would, in turn, decrease our question-answering power, i.e., that it would

enable us to answer fewer questions correctly. Here, it is important that we do not

confuse power with what the previous chapter referred to as robustness (see §4.1.3).

Granted, it is likely that unaided clinical judgment constitutes a more robust method than

does reliance on prediction models, since there are many questions for which prediction

19
See Dawes et al. (2002).
20
Despite resisting the term ‘paternalism’ in his (2005) and (2009), Trout (in
correspondence) suggests that there is an implicit case for paternalism in Bishop and
Trout (2005). If so, the following can be seen as an attempt to spell out the relevant case
explicitly.

149
models are yet to be developed and, hence, many questions for which the former would

outperform the latter. But power, like epistemic betterness, is a property defined for a

specific set of questions, i.e., an agenda. In other words, when comparing the power of

different methods, we need to hold fixed the questions at issue, such as the questions

pertaining to some particular diagnostic or prognostic problem for which a model is

available. Given such a set of questions, the claim is that the relevant prediction model

will not yield less true belief than do the corresponding unaided clinical judgment

processes. Consequently, there is good reason to believe that relying on prediction

models (whenever available) will lead to a pareto improvement in reliability for the

inquirers involved, ceteris paribus, and that the relevant domains, hence, satisfy (C) (b).

It might be objected that the answers provided by way of prediction models will

not be as informative as those yielded by unaided clinical reflection. Two things need to

be noted here. First, even if a case could be made to the effect that unaided clinical

judgments generally are more informative than the outputs of prediction models, it is

sufficient for the two methods to be on a par on this score that they are both sufficiently

informative to answer the relevant questions, whether or not one is more informative than

the other. Second, it is questionable that a case can be to the effect that unaided clinical

judgments generally are more informative than the outputs of prediction models. It

certainly cannot be ruled out that unaided clinical reflection often will yield judgments

containing more information than what can be gleaned from prediction models, which

typically yield a simple binary or numerical output. However, even if that is so, the fact

that the latter, nevertheless, tend to outperform the former suggests that whatever

additional information tends to be provided through unaided clinical judgment generally

150
is not of a kind that increases accuracy. And when it comes to diagnosis and prognosis, it

is accuracy—not linguistic flourishes—that we are ultimately after.

Keeping these points in mind, we see that the judgment domains of medical

diagnosis and prognosis, clearly, fall within the scope of (C), which suggests that we

have pro tanto reason for mandating compliance with the models in question. Moreover,

the practice of implementing such a mandate would be epistemically paternalistic. For

one thing, the practice would, as already noted, be motivated with reference to how

relying on said models makes inquirers epistemically better off relative to prevailing

practices (i.e., reliance on unaided clinical judgment). Consequently, the practice would

satisfy (P) (iii). Furthermore, a practice that mandated the use of predictive models (when

available) would put real restrictions on clinicians’ ability to pursue diagnoses and

prognoses in whatever manner they happen to see fit. Consequently, the practice would

satisfy (P) (i). As noted above, this is not to rule out the possibility that the relevant

inquirers may happily utilize the models, perhaps on account of appreciating their merits.

However, considering the well-documented prevalence of defection, and the fact that

inquirers have been shown to disapprove (even while lacking epistemic reason for doing

so), the practice of implementing the mandate can be motivated without any concern for

whether the inquirers themselves actually consent to its implementation. Consequently,

the practice would also satisfy (P) (ii).

In conclusion, we have pro tanto reason to mandate compliance in utilizing those

prediction models that have been shown to make us epistemically better off, as compared

to prevailing practices. Moreover, the practice of mandating such compliance would be

epistemically paternalistic, in virtue of not only being motivated with reference to how it

151
would make clinician’s epistemically better off but also constraining their freedom to

conduct inquiry in whatever way they see fit, without seeking their consent in so doing.

Given that such a mandate is yet to be instituted, a commitment to (C), hence, provides

pro tanto reason to practice epistemic paternalism on a wider scale than we are currently

doing.

5.3.2. Prediction Markets as an Alternative to Social Deliberation

The second domain in which we have pro tanto reason to implement paternalistic

practices can be illustrated with reference to the fact that, in any reasonably complex

society or organization, it will often be the case that only a minority of the people

involved will be informed on any given issue. In those domains where we care about

accuracy, how are we to mine such informed minorities for information? One intuitively

plausible and historically prominent way to do so is by way of social deliberation,

understood as the process of a group of people exchanging reasons for and against a (set

of) conclusion(s) for the purpose of reaching a social verdict on the issue in question.

Does such a process do a good job of eliciting information from informed minorities?

To answer this question, we need to consider what happens to informed minorities

in contexts of social deliberation. There are two possibilities. First, assume that the

members of the informed minority do not disclose their information. After all, there are

many circumstances under which people refrain from disclosing what they know (or what

they take themselves to know), either in light of the informational pressure coming out of

whatever happens to be the majority position (and the assumption on part of the minority

that they, not the majority, are probably mistaken), or the social pressure associated with

152
the risk of social sanctions against dissenters.21 In other words, in the event that there is a

diversity of opinions, and the informed members find themselves in a minority, there is a

real risk that the information they have in their possession will not even be submitted for

deliberation.

Second, assume that the members of the informed minority do disclose their

information. What is likely to be the impact of that information on the deliberating

group? Given that the informed members make up a minority, the impact is likely not to

be particularly great, due to what is typically referred to as the common knowledge effect.

(Since the effect in question applies to information submitted for deliberation in general,

and not just to information believed in a manner that amounts to knowledge, a more

appropriate designation might have been “the common information effect”.) More

specifically, social psychologists Daniel Gigone and Reid Hastie sum up the relevant

research in terms of how “[t]he influence of a particular item of information [on the

judgment of a group] is directly and positively related to the number of group members

who have knowledge of that item before the group discussion and judgment” (Gigone

and Hastie 1993, p. 960).

In other words, what makes a difference when it comes to having an impact on

group judgments is not so much quality of information as quantity of people bringing a

particular piece of information to the table. This would be great news, were there a robust

correlation between the information had by the majority and the truth. While such a

correlation might hold with respect to certain easy questions that have demonstrable

answers, it clearly does not hold with respect to a great many of the questions that we

21
See Sunstein (2006) for a discussion of the relevant evidence.

153
typically deliberate over, such as questions about politics, public policy, public health,

scientific research, and so on. Consequently, it does not seem that social deliberation

provides a particularly promising method for eliciting and aggregating dispersed pieces

of information for the purpose of arriving at informed judgments on a wide variety of

questions that we care about answering. At best, it seems a prima facie promising method

for aggregating information that is already widely shared—which is not the kind of role

we are concerned with presently.

This gives us reason to look elsewhere for a more promising method for eliciting

information from informed minorities. One strategy that has received a lot of attention

recently is that of setting up a prediction market. A prediction market is a market that is

designed and run for the primary purpose of (a) mining and aggregating information

scattered among traders, and subsequently (b) using this information in the form of

market values to make predictions about future events. As it turns out, the market prices

of such markets provide surprisingly accurate information about the events bet on.22

Moreover, there is evidence to the effect that accuracy of the market prices that

arise in prediction market are to a great extent due to small subsets of traders, typically

referred to as marginal traders.23 Beyond investing more money in the markets and

spending more time trading than the average trader, marginal traders also earn

significantly higher returns. While it might be thought that this is a mere artifact of

marginal traders simply investing more money than non-marginal ones, further

considerations suggest that the most plausible interpretation of these returns is that the

22
See Bragues (2009), Wolfers and Zitzewitz (2006), and Berg et al. (2003) on political
prediction markets, Pennock et al. (2001) on commercial forecasting, Chen and Plott
(2002) on sales forecasting, and Servan-Schreiber et al. (2004) on sports forecasting.
23
See Forsythe et al. (1999).

154
former simply are more informed and, consequently, place better bets than the latter.

After all, by virtue of their high trading frequency and higher than average investments,

marginal traders have a disproportionally high influence on the price signal. Given that

the price signals of prediction markets tend to amount to good predictions, this suggests

that the accuracy of the predictions of such markets are, to a great extent, due exactly to

the informed investments of marginal traders. In fact, there is further and more direct

evidence for this claim. Separating out the subset of marginal traders that are particularly

active in setting bid and ask prices, as compared to those traders (marginal or not) who

mostly accept others’ prices, Oliven and Rietz (2004) found that the average error rate for

the latter on the 1992 vote share market of the Iowa Electronic Markets—one of the

oldest and most prominent electronic prediction markets—was almost six times that of

the former (8% across all trades, versus 47%).

The role played by such marginal traders in prediction markets has implications

for how to think about the respective merits of social deliberation and prediction markets.

More specifically, above discussion suggests that prediction markets do a good job of

mining information exactly under the conditions where social deliberation does not, i.e.,

in contexts of informed minorities. In fact, the evidence just reviewed gives us pro tanto

reason to institute a paternalistic mandate for using information gleaned from prediction

markets rather than that coming out of social deliberation, in cases where informed

minorities are present. To see why, consider, first, the fact that many of the issues that we

care to deliberate about socially, including issues pertaining to governmental policy and

organizational strategy, are ones where we care about accuracy. Consequently, the

domains in question satisfy (C) (a). Next, consider the relative epistemic merits of relying

155
on prediction markets as opposed to on social deliberation. More specifically, consider

the following two processes:

(1) Forming beliefs on the basis of the judgments of deliberating groups; and

(2) forming beliefs on the basis of the price signals of prediction markets.

Given what we know about the workings of social deliberation and prediction markets,

respectively, we have good reason to believe that, if informed minorities are present,

someone forming belief in the manner described in 2 will form a higher ratio of true

belief than someone forming belief in the manner described in 1 (ceteris paribus).

Furthermore, in analogy with what was argued in relation to the power of prediction

models above (and keeping in mind the distinction between power and robustness), there

is no reason to believe that basing ones beliefs on the price signals of prediction markets,

whenever available, will lead to a decrease in question-answering power.

Consider, also, the issue of whether the answers gleaned from prediction markets

are sufficiently informative. While the social verdicts of deliberating groups may,

certainly, provide more information than what can be gleaned from the price signals of a

prediction market, the fact that the someone relying on the latter as opposed to on the

former is likely to be epistemically better off suggests that prediction markets do a

significantly better job than do deliberating groups in harnessing the right information

and, in turn, making for accurate answers. As such, the relevant worry about

informativeness is not that prediction markets provide insufficiently informative outputs;

it is that socially deliberating groups might not satisfy a necessary condition on any

156
sufficiently informative answer, as defined in the previous chapter, i.e., that the judgment

in question be an accurate one.

In other words, there is good reason to believe that relying on prediction markets

(whenever available) will lead to a pareto improvement with respect to reliability for the

inquirers involved, ceteris paribus, and that the relevant domains, hence, satisfy (C) (b).

As such, the many domains typically associated with social deliberation fall within the

scope of (C), suggesting that we have pro tanto reason for mandating reliance on the

outputs of relevant prediction markets, whenever available. More than that, a practice

involving such a mandate would be epistemically paternalistic. As already noted, the

mandate would be motivated with reference to the epistemic merits of relying on

prediction markets. As such, it would satisfy (P) (iii). However, it should be stressed that

the practice of enforcing the mandate would not involve a ban on social deliberation.

Instead, it would involve a mandate on those making judgments within the relevant

domains to use information gleaned from prediction markets (when such markets are

available) rather than information produced by way of social deliberation. Since this

would put restrictions on the freedom of those making judgments to use whatever kind of

information they deem relevant, the practice would satisfy (P) (i). Moreover, since there

is far from a guarantee that those restricted thus would welcome the restriction, defection

is a real possibility, not the least given the risk that inquirers will exhibit overconfidence

in their assessments of whether they will be able to judge what kind of information is

most probative. As above, this is not to deny that some of those restricted may actually

welcome the restriction; it is simply to suggest that the practice can be motivated without

157
any concern for whether the inquirers themselves actually approve of its implementation.

As such, the practice satisfies (P) (ii).

In conclusion, many of the judgment domains typically associated with social

deliberation satisfy (C), and we, consequently, have pro tanto reason to mandate reliance

on prediction markets as opposed to on the outputs of social deliberation within the

relevant domains, whenever such markets are available. More than that, the practice of

mandating such reliance would be epistemically paternalistic, in virtue of not only being

motivated with reference to how it would make those making judgments within the

relevant domains better off, but also constrain their freedom to conduct inquiry in

whatever way they see fit, without seeking their prior consent. Given that such a mandate

is yet to be instituted on the scale considered presently, we thereby have yet another

example of how a commitment to (C) provides pro tanto reason to practice epistemic

paternalism on a wider scale than we are currently doing.

5.4. Countervailing Reasons

The previous section argued that we have pro tanto reason for practicing epistemic

paternalism on a wider scale than we are currently doing, and illustrated this point with

reference to the use of prediction models in medical diagnosis and prognosis, and the

reliance on prediction markets as a tool for mining informed minorities for information.

The present section returns to a version of the worry discussed in §5.2, to the effect that

paternalistic practices are objectionable simply on account of us having non-epistemic

158
reasons not to impose paternalistic constraints on agency, even if such constraints turn

out to be epistemically efficacious.

In order to address this worry, we need to say something about the conditions

under which pro tanto reasons amount to reasons all things considered. The following is

a sufficient condition:

(S) A pro tanto reason to ϕ amounts to a reason to ϕ all things considered if

there are no countervailing reasons, i.e., there are no reasons not to ϕ that

are strong enough to offset the pro tanto reason to ϕ.

The final point to be argued in this chapter is that the reasons pertaining to the

paternalistic practices discussed in §5.3 amount to reasons all things considered, when the

relevant mandates are properly qualified. Let us consider the two cases in turn.

5.4.1. Reasons for Not Relying on Prediction Models

Opting for the output of a statistical model rather than relying on the judgment of

experienced clinicians might seem to amount to stripping the relevant patients of their

individuality. At least that is how it feels. But does that feeling give us an ethical reason

against mandating the use of prediction models? It is not clear that it does, considering

the ethical costs of relying on methods that consistently have been shown to be equally

reliable at best, and significantly less reliable at worst. In this respect, little has changed

since Robyn Dawes wrote the following in a classical review article:

159
 No matter how much we would like to see this or that aspect of one or another of
the studies reviewed in this article changed, […] no matter how ethically
uncomfortable we may feel at ‘reducing people to mere numbers,’ the fact
remains that our clients are people who deserve to be treated in the best manner
possible. If that means—as it appears at present—that selection, diagnosis, and
prognosis should be based on nothing more than the addition of a few numbers
representing values on important attributes, so be it. To do otherwise is cheating
the people that we serve (Dawes 1979, p. 581).

In other words, it seems that ethical reasons might even speak for a strict reliance on

prediction models, given the potential ethical costs of relying on a method that has been

shown to be equally reliable at best, and significantly less reliable at worst.

Another potential worry is that relying on prediction models will be more costly

and time-consuming than relying on unaided clinical judgments. However, the typical

prediction model tends to be developed exactly for the purpose of decreasing cost and

reducing the time needed to arrive at a diagnosis or prognosis. By way of a representative

example, consider the model developed by Breiman et al. (1993) to classify potential

heart attack patients according to risk, in order to reduce the number of unnecessary

admittances of low risk patients to coronary care units. Beyond being more reliable than

both unaided clinicians and several complex statistical classification methods, the model

only requires the clinician to answer a maximum of three diagnostic, yes/no questions.

By way of a further example, consider the Ottawa Ankle Rule, developed by

Stiell et al. (1992) as a tool for diagnosing ankle fractures for the purpose of reducing the

number of unnecessary x-rays. In using the rule, the clinician needs to answer a

maximum of three diagnostic questions in order to determine whether an x-ray is needed.

Despite its simplicity, the rule has a very low rate of false negatives, yet allows for a

reduction of up to 36% of ankle x-rays, compared to prevailing practices. Still, Graham et

160
al. (2001) reports that, while 96% of emergency physicians in the United States are

familiar with the rule, 67% never use it, or only use it sometimes.

In other words, considerations about cost and time-efficiency might just be

speaking for a strict reliance on prediction models, whenever available. This, of course, is

compatible with there being high costs associated with introducing (as opposed to using)

the models in question.24 This brings us to the third worry: There might be cases in which

introducing the model will exceed the benefits of using them. Needless to say, this

possibility cannot be ruled out a priori, even if the benefits of increased accuracy and

decreased cost and time used that result from opting for prediction models make it likely

that even a significant start-up cost can be offset over time. Still, it seems reasonable to

require that the mandate should not apply in cases where the costs of introducing the

models in question clearly exceed the benefits of using them. The relevant weighing of

costs and benefit will, most likely, have to be done on a case-by-case basis, in a manner

that is, hopefully, sufficiently sensitive to the variety of concerns that may be brought to

bear on the issues at hand to, at the very least, reveal cases in which even aggregate

benefits are clearly outweighed by relevant costs.

In conclusion, the overall balance of epistemic, ethical, and practical

considerations suggest that we have reason to institute a paternalistic mandate on using

prediction models as opposed to relying on unaided clinical judgments, unless the cost of

introducing the models in question clearly outweighs even the long-term benefits—

epistemic and otherwise—of doing so. Given this qualification, and in the absence of any

24
See Bishop and Trout (2005) for a good discussion.

161
further countervailing reasons, (S) implies that the pro tanto reason involved amounts to

a reason all things considered.

5.4.2. Reasons for Not Relying on Prediction Markets

Turning now to prediction markets, three concerns deserve mention. First, prediction

markets require a fairly demanding infrastructure, and there might be many situations

where constraints on cost and other available means (not the least with respect to

attracting a sufficiently informed subset of traders) will make setting up a reliable

prediction market practically impossible. However, the mandate that concerns us here is

not one on setting up prediction markets, but rather on using the information coming out

of such markets, when available. The epistemic merits of using such information might

also give us pro tanto reason to set up more such markets. That, however, is a different

matter from the one that we have been concerned with here.

Second, social deliberation is often taken to be what renders governmental

policies politically legitimate, particularly in so far as the policies in question involve the

use of force against those representing a minority opinion. In light of this, there might be

worries about whether a greater reliance on prediction markets as opposed to on social

deliberation on part of governments—and perhaps on part of social institutions more

generally—would serve to erode the legitimacy of the policies in question.25

A couple of things need to be noted in response to this worry. The idea that

deliberation is the source of legitimacy is particularly prominent within theories of

deliberative democracy, in turn “rooted in the intuitive ideal of a democratic association

25
Thanks to Robert Talisse for raising this issue.

162
in which the justification of the terms and conditions of association proceeds through

public argument and reasoning among equal citizens,” to quote Joshua Cohen (1997, p.

72). The fact that the ideal is taken to apply to the justification of the terms and

conditions of association is important, since it suggests that the ideal is not meant to

apply to all instances of policy making, but only to those policies that settle fundamental

issues about what terms and conditions of democratic association would be just or fair.

On this interpretation, the deliberative ideal applies only to what John Rawls (1997) in

his account of public reason refers to as “constitutional essentials and matters of basic

justice” (p. 95), pertaining to issues such as “who has the right to vote, or what religions

are to be tolerated, or who is to be assured fair equality, or to hold property” (p. 94).

Needless to say, these are not the kinds of issues that we are imagining that the price

signals of prediction markets are to settle. Rather, prediction markets would enter at the

point where questions about constitutional essentials and matters of basic justice are

settled, and the relevant questions about governance are predictive questions about what

policies, as a matter of empirical fact, are likely to generate what goods, as defined by

whatever turns out to be the proper account of justice.

That point aside, imagine that the ideal in fact does apply more generally, in that

it applies to any procedure invoked in arriving at policies that involve the use of force

against minorities. In this context, it is crucial to note that what we are dealing with here

is an ideal, that, as such, leaves open how we can best approximate this deliberative

conception of democratic interaction in practice. As Cohen puts the point, “[t]he

appropriate ordering of deliberative institutions depends on issues of political psychology

and political behavior; it is not an immediate consequence of the deliberative ideal” (p.

163
85). Consequently, it cannot be ruled out that a proper realization of the deliberative ideal

might look very different from the ideal itself. To name but one respect in which this will

be so, consider Jürgen Habermas’ (1975) conception of ideal deliberation as involving a

context in which “no force except that of the better argument is exercised” (p. 108). As

we have seen, actual social deliberation is very different from such ideal deliberation in

that several forces are exercised beyond that of the better argument, including the force

exerted by the sheer the number of proponents that a particular argument happens to

have, or the social or informational pressures that may suppress the arguments of

minorities from even being submitted for deliberation. As such, social deliberation is

under many circumstances simply not conducive to the inclusion of all participants’

perspectives, nor consequently to the treatment of all deliberating citizens as equal,

contrary to the picture of democratic deliberation associated with aforementioned

intuitive ideal.26 In particular, and for reasons discussed above, the epistemic perspectives

of minorities are bound to lose out in contexts of social deliberation simply on account of

being minorities.

As we have seen, prediction markets do a much better job than social deliberation

in taking into account the reasons and perspectives provided by minorities, at least in so

far as they are informed on the relevant issues. Indeed, in contexts involving such

minorities, it seems that setting up a prediction market will actually provide a better

approximation of the deliberative ideal than social deliberation would do. This contention

might seem paradoxical, but should come as no surprise given (a) the idealized nature of

26
This is not to suggest that equality would be sufficient (as opposed to merely
necessary) for realizing the relevant ideal. If that were the case, a simple coin toss would
to the trick, as pointed out by Estlund (2008).

164
the intuition driving deliberative conceptions of democracy, and (b) the tendency for bias

and overconfidence that we are subject to as actual deliberators. Moreover, keeping this

in mind, it is not clear why the minority should feel any less compelled to accept

decisions based on information gleaned from prediction markets as opposed to from

social deliberation, contrary to the worry raised above. In the event that the members of

the minority are uninformed, they will run the risk of being ignored irrespective of

whether they are trading on a prediction market or attempting to partake in deliberation

(in the former case, by virtue of being uninformed; in the latter case, simply by virtue of

being members of a minority). More than that, if the minority happens to be informed, its

members might even have more reason to feel compelled thus in light of decisions taken

on the basis of markets rather than on deliberation, since they have a greater chance of

having their voices be heard in the former than in the latter.

Third, if there is a mandate on using information from prediction markets, and

more decisions, as a result, will be based on the price signals of such markets, we can

expect attempts to influence those decisions through market price manipulations. How

great of a threat is this? There is plenty of evidence of people investing in political

prediction markets with the intention of boosting the perceived likelihood of their

candidate winning the relevant office. However, as pointed out by Rhode and Strumpf

(2009), there is little evidence that political prediction markets can be systematically

manipulated beyond short periods of time. By way of example, political betting markets

on Wall Street in the late 19th and early 20th century involved millions of dollars, and

many attempts at boosting candidates by way of speculative investments. These

165
speculations resulted in price change, but the prices returned to their pre-attack levels

within days.

More recently, a series of random investments made by Rhode and Strumpf

themselves in the 2000 Iowa Electronic Markets’ presidential market led to large initial

price changes, but the prices reverted to their initial levels in a few hours. These are, of

course, exactly the results we should expect to get in a market where at least some of the

investors utilize accurate information for the purpose of profiting from the influx of

liquidity provided by the manipulators. That said, while available empirical evidence

indicates that manipulation may not be as urgent a problem as it might seem at first

glance, the possibility of such manipulation suggests that the mandate to rely on

prediction markets should be suspended in the presence of evidence for large-scale

strategic investments.

In conclusion, the overall balance of epistemic, practical, political, and ethical

considerations come out in favor of a paternalistic mandate on using information gleaned

from prediction markets rather than information coming out of social deliberation, in

cases where we (a) have reason to believe that informed minorities are present, and (b)

lack any evidence for large-scale strategic investments in the relevant markets. Given this

qualification, and in the absence of any further countervailing reasons, (S) implies that

the pro tanto reason involved amounts to a reason all things considered.

166
5.5. Conclusion

When conducting inquiry, we gather information, mull over our data, choose among

different methods of investigation, and so on—all for the purpose of attaining true belief

relevant to the questions that we pose. In so doing, we are expressing our epistemic

agency. Above, it was argued that, when it comes to our freedom to express such agency,

more is not always better. These points were argued with reference to how our dual

tendency for bias and overconfidence gives us pro tanto reason for mandating

compliance with methods that have been shown to lead to pareto improvements in

reliability. Moreover, it was argued that mandating compliance thus would be

epistemically paternalistic, and that we have not just pro tanto but all-things-considered

reason to practice such paternalism on a wider scale than we are currently doing. For

example, in many situations, we are better off by gathering only a very limited amount of

information, having our selection of methods be greatly restricted, and spending our time

less on reflecting than on simply reading off a market price or the output of a simple

prediction model. Consequently, when it comes to our freedom to express our epistemic

agency, more is not always better. In fact, less is often so much more.

167
 BIBLIOGRAPHY

Alston, W., (2005), Beyond “Justification”: Dimensions of Epistemic Evaluation, Ithaca:

Cornell University Press.

Arkes, H. R., Dawes, R. M., and Christensen, C., (1986), “Factors influencing the use of
a decision rule in a probabilistic task,” Organizational Behavior and Human
Decision Processes 37, pp. 93–110.

Arkes, H., Christensen, C., Lai, C., and Blumer, C., (1987), “Two Methods of Reducing
Overconfidence”, Organizational Behavior and Human Decision Processes 39,
pp. 133–144.

Baehr, J., (2009), “Is There a Value Problem?” in A. Haddock, A. Millar, and D.
Pritchard (2009), pp. 42-59.

Belnap, N., and Steel, T., (1976), The Logic of Questions, New Haven: Yale University
Press.

Berg, J., Nelson, F., and Rietz, R., (2003), “Accuracy and Forecast Standard Error of
Prediction Markets,” manuscript, July 2003.

Bishop, M., (2000), “In Praise of Epistemic Irresponsibility: How Lazy and Ignorant Can
You Be?,” Synthese 122, pp. 179-208.

Bishop, M., and Trout, J. D., (2005), Epistemology and the Psychology of Human
Judgment, New York: Oxford University Press.

Bohman, J., and Rehg, W., (1997), Deliberative Democracy: Essays on Reason and
Politics, Cambridge: The MIT Press.

BonJour, L., (1985), The Structure of Empirical Knowledge, Cambridge: Harvard

University Press.

Bradley, B., (2006), “Two Concepts of Intrinsic Value,” Ethical Theory and Moral
Practice 9, pp. 111-130.

Bragues, G., (2009), “Prediction Markets: The Practical and Normative Possibilities for
the Social Production of Knowledge,” Episteme 6 (2), pp. 91-106.

Breiman, L., Friedman, J. H., Olshen, R. A. and Stone, C. J., (1993), Classification and
Regression Trees, Boca Raton: Chapman & Hall.

168
Burnyeat, M. F., (1980), “Socrates and the Jury: Paradoxes in Plato’s Distinction
Between Knowledge and True Belief,” Proceedings of the Aristotelian Society
suppl. vol. 54, pp. 173-191.

Chang, R., (1997), ed., Incommensurability, Incomparability, and Practical Reason,

Cambridge: Harvard University Press.

Chang, R., (2002), “On the Possibility of Parity,” Ethics 112, pp. 659-688.

Chen, K., and Plott, C., (2002), “Information Aggregation Mechanisms: Concept, Design
and Implementation for a Sales Forecasting Problem,” CalTech Social Science
Working Paper No. 1131.

Chisholm, R. M., (1968), “Lewis’ Ethics of Belief,” in P. Schilpp, ed., The Philosophy of
C. I. Lewis, LaSalle: Open Court.

Chisholm, R. M., (1977), Theory of Knowledge, 2nd ed., Englewood Cliffs: Prentice-Hall.

Chisholm, R. M., (1991), “Firth and the Ethics of Belief,” Philosophy and
Phenomenological Research 51 (1), pp. 119-128.

Clifford, W. K., (1866), “The Ethics of Belief,” in L. Stephen and F. Pollock, ed.,
Lectures and Essays by the late William Kingdon Clifford, 2nd ed., London:
Macmillan and Co., pp. 339-363.

Cohen, J., (1997), “Deliberation and Democratic Legitimacy,” in Bohman and Rehg
(1997), pp. 67-91.

Conee, E., and Feldman, R., (2008), “Evidence,” in Q. Smith, ed., Epistemology: New
Essays, New York: Oxford University Press, pp. 83-104.

Cooper, J. M., (1997), Plato: Complete Works, Indianapolis: Hackett.

David, M., (2001), “Truth as the Epistemic Goal”, in M. Steup, ed., Knowledge, Truth,
and Duty, New York: Oxford University Press.

David, M., (2005), “Truth as the Primary Epistemic Goal: A Working Hypothesis,” in M.
Steup and E. Sosa, (2005), pp. 296-312.

Dawes, R., (1979), “The Robust Beauty of Improper Linear Models in Decision
Making,” American Psychologist 34 (7), pp. 571-582.

Dawes, R., Faust, D., Meehl, P., (2002), “Clinical versus Actuarial Judgment,” in
Gilovich et al. (2002), pp. 716-729.

169
DePaul, M. R., (2009), “Ugly Analyses and Value,” in A. Haddock, A. Millar, and D.
Pritchard (2009), pp. 112-138.

Descartes, R., (1641/1984), “Meditations on First Philosophy,” in J. Cottingham, R.

Stoothoff, and D. Murdoch, eds., The Philosophical Writings of Descartes vol. II,
Cambridge: Cambridge University Press.

Dworkin, G. (2002) “Paternalism,” in E. N. Zalta (ed.), The Stanford Encyclopedia of

Philosophy; accessed from http://plato.stanford.edu/entries/paternalism/ in
November, 2009.

Elster, J., (2000), Ulysses Unbound: Studies in Rationality, Precommitment, and

Constraints, Cambridge: Cambridge University Press.

Estlund, D. M., (2008), Democratic Authority: A Philosophical Framework, Princeton:

Princeton University Press.

Fallis, D., (2006), “Epistemic Value Theory and Social Epistemology,” Episteme 2 (3),
pp. 177-188.

Feldman, R., (1985), “Reliability and Justification,” the Monist 68, pp. 159-174.

Feldman, R., (1988), “Epistemic Obligations,” Philosophical Perspectives (2), pp. 235-
356.

Feldman, R., (1999), “Methodological Naturalism in Epistemology,” in J. Greco and E.

Sosa, eds., The Blackwell Guide to Epistemology, Malden: Blackwell, pp. 170-
186.

Feldman, R., (2001), “Review of Alvin Goldman’s Knowledge in a Social World,”

British Journal for the Philosophy of Science 52, pp. 163-168.

Feldman, R., (2002), “Epistemological Duties,” in P. Moser, ed., The Oxford Handbook
of Epistemology, New York: Oxford University Press, pp. 361-384.

Feldman, R., (2003), “Chisholm’s Internalism and its Consequences,” Metaphilosophy 43

(5), pp. 603-620.

Field, H., (2001), Truth in the Absence of Fact, New York: Oxford University Press.

Fine, G., (2003), Plato on Knowledge and Forms: Selected Essays, New York: Oxford
University Press.

Fine, G., (2004), “Knowledge and True Belief in the Meno,” Oxford Studies in Ancient
Philosophy XXVII, pp. 41-81.

170
Fine, G., (2008), The Oxford Handbook of Plato, New York: Oxford University Press.

Forsythe, R., Rietz, T. A., and Ross, T. W., (1999), “Wishes, Expectations and Actions:
A Survey on Price Formation in Election Stock Markets,” Journal of Economic
Behavior and Organization 39, pp. 83-110.

Gigerenzer, G., Todd, P. M., and the ABC Research Group, eds., (1999), Simple
Heuristics that Make Us Smart, New York: Oxford University Press.

Gigone, D., and Hastie, R., (1993), “The Common Knowledge Effect: Information
Sharing and Group Judgment,” Journal of Personality and Social Psychology 65
(5), pp. 959-974.

Gilovich, T., Griffin, D., and Kahneman, D., (2002), Heuristics and Biases: The
Psychology of Intuitive Judgment, Cambridge: Cambridge University Press.

Goldman, A., (1978), “Epistemics: The Regulative Theory of Cognition,” The Journal of
Philosophy 75 (10), pp. 509-523.

Goldman, A., (1986), Epistemology and Cognition, Cambridge: Harvard University

Press.

Goldman, A., (1992), Liaisons: Philosophy Meets the Cognitive and Social Sciences,
Cambridge: The MIT Press.

Goldman, A., (1992a), “Foundations of Social Epistemics,” in Goldman (1992), pp. 179-
208.

Goldman, A., (1992b), “Epistemic Paternalism: Communication Control in Law and

Society,” in Goldman (1992), pp. 209-226.

Goldman, A., (1999), Knowledge in a Social World, New York: Oxford University Press.

Goldman, A., (forthcoming), “Toward a Synthesis of Reliabilism and Evidentialism? Or:

Evidentialism’s Problems, Reliabilism’s Rescue Package,” in T. Dougherty, ed.,
Evidentialism and Its Discontents, New York: Oxford University Press.

Goldman, A., and Olsson, E. J., (2009), “Reliabilism and the Value of Knowledge,” in A.
Haddock, A. Millar, and D. Pritchard (2009), pp. 19-41.

Graham, I. D., Stiell, I. G., Laupacis, A., McAuley, L., Howell, M., Clancy, M., Durieux,
P., Simon, N., Emparanza, J. I., Aginaga, J. R., O’Connor, A., and Wells, G.,
(2001), “Awareness and use of the Ottawa Ankle and Knee Rules in 5 countries:
Can Publication Alone be Enough to Change Practice?” Annals of Emergency
Medicine, 37(3), pp. 259–266.

171
Greco, J., (2009), “The Value Problem,” in A. Haddock, A. Millar, and D. Pritchard
(2009), pp. 313-321.

Haack, S., (1993), Evidence and Inquiry: Towards Reconstruction in Epistemology,

Oxford: Basil Blackwell.

Haack, S., (2001), “’The Ethics of Belief’ Reconsidered,” in M. Steup, ed., Knowledge,
Truth, and Duty, New York: Oxford University Press, pp. 21-33.

Habermas, J., (1975), Legitimation Crisis, translated by T. McCarthy, Boston: Beacon

Press.

Haddock, A., Millar, A., and Pritchard, D., eds., (2009), Epistemic Value, New York:
Oxford University Press.

Heil, J., (1984), “Doxastic Incontinence,” Mind 93 (369), pp. 56-70.

Hempel, C., (1965), “Aspects of Scientific Explanation,” in Aspects of Scientific

Explanation and Other Essays in the Philosophy of Science, New York: Free
Press, pp. 331-496.

Hunt, J. S., and Budesheim, T. L., (2004), “How Jurors Use and Misuse Character
Evidence,” Journal of Applied Psychology 89 (2), pp. 347-361.

James, W., (1907), “The Will to Believe,” in The Will to Believe and Other Essays in
Popular Philosophy, New York: Longmans Green and Co., pp. 1-31.

Jaworski, W., (2009), “The Logic of How-Questions,” Synthese 166, pp. 133-155.

Jones, W. E., (1997), “Why Do We Value Knowledge?” American Philosophical

Quarterly 34 (4), pp. 423-439.

Kitcher, P., (1992), “The Naturalists Return,” The Philosophical Review 101 (1), pp. 53-
114.

Klein, P., (1981), Certainty, Minneapolis: University of Minnesota Press.

Kornblith, H., (1994), Naturalizing Epistemology, 2nd ed., Cambridge: The MIT Press.

Kornblith, H., (2002), Knowledge and Its Place in Nature, New York: Oxford University
Press.

Kornblith, H., (forthcoming), On Reflection, New York: Oxford University Press.

Korsgaard, C., (1996), The Sources of Normativity, Cambridge: Cambridge University

Press.

172
Kvanvig, J., (2003), The Value of Knowledge and the Pursuit of Understanding,
Cambridge: Cambridge University Press.

Kvanvig, J., (2009), “The Value of Understanding,” in A. Haddock, A. Millar, and D.

Pritchard (2009), pp. 95-111.

Latus, A., (2000), “Our Epistemic Goal,” Australasian Journal of Philosophy, 78 (1), pp.
28-39.

Laudan, L., (1977), Progress and Its Problems: Towards a Theory of Scientific Growth,
Berkeley: University of California Press.

Laudan, L., (2006), Truth, Error, and Criminal Law: An Essay in Legal Epistemology,
Cambridge: Cambridge University Press.

Locke, J., (1706/1996), “Of the Conduct of the Understanding,” in R. W. Grant and N.
Tarcov, eds., Some Thoughts Concerning Education and Of the Conduct of the
Understanding, Indianapolis: Hackett.

Lord, C. G., Preston, E., and Lepper, M. R., (1984), “Considering the Opposite: A
Corrective Strategy for Social Judgment,” Journal of Personality and Social
Psychology 47 (6), pp. 1231-1243.

Lupfer, M. B., Weeks, M, and Dupuis, S., (2000), “How pervasive is the negativity bias
in judgments based on character appraisal?” Personality and Social Psychology
Bulletin 26, pp. 1353–1366.

Lynch, M., (2004), True to Life: Why Truth Matters, Cambridge: The MIT Press.

Lynch, M., (2009a), “The Values of Truth and the Truth of Values,” in A. Haddock, A.
Millar, and D. Pritchard (2009), pp. 225-242.

Lynch, M., (2009b), “Truth, Value and Epistemic Expressivism,” Philosophy and
Phenomenological Research 79 (1), pp. 76-97.

Meehl, P. (1954), Clinical versus Statistical Prediction: A Theoretical Analysis and a

Review of the Evidence, Minneapolis: University of Minnesota Press.

Mill, J. S., (1861/2001), “Utilitarianism”; reprinted in G. Sher, ed., Utilitarianism and the
1868 Speech on Capital Punishment, 2nd ed., Indianapolis: Hackett Publishing
Company, 2001, pp. 1-64.

Moran, R., (2001), Authority and Estrangement, Princeton: Princeton University Press.

173
Oliven, K., and Rietz, T. A., (2004), “Suckers are Born but Markets are Made: Individual
Rationality, Arbitrage, and Market Efficiency on an Electronic Futures Market,”
Management Science 50 (3), pp. 336-351.

Olsson, E. J., (2005), “The Impossibility of Coherence,” Erkenntnis 63, pp. 387-412.

Olsson, E. J., (2007), “Reliabilism, Stability, and the Value of Knowledge,” American
Philosophical Quarterly 44 (4), pp. 343-355.

Pennock, D., Lawrence, S., Nielsen, F. Å., and Giles, C. L., (2001), “Extracting
Collective Probabilitistic Forecasts from Web Games,” Proceedings of the
Seventh ACM SIGKDD International Conference on Knowledge Discovery and
Data Mining, pp. 174-183.

Plantinga, A., (1993), Warrant and Proper Function, New York: Oxford University
Press.

Pritchard, D., (2007), “Recent Work on Epistemic Value,” American Philosophical

Quarterly 44 (2), pp. 85-110.

Pritchard, D., (forthcoming), “Knowledge and Understanding,” in D. Pritchard, A. Millar,

and A. Haddock, eds., The Nature and Value of Knowledge: Three Investigations,
New York: Oxford University Press.

Pronin, E., Lin, D. Y., and Ross, L., (2002), “The Bias Blind Spot: Perceptions of Bias in
Self versus Others,” Personality and Social Psychology Bulletin 28, pp. 369-381.

Rabinowicz W., and Rønnow-Rasmussen, T., (2000), “A Distinction in Value: Intrinsic

and for Its Own Sake,” Proceedings of the Aristotelian Society 100, pp. 33-51.

Ramsey, F. P., (1931), “General Propositions and Causality,” in F. P. Ramsey, The

Foundations of Mathematics and Other Logical Essays, London: Routledge and
Kegan Paul, pp. 237-255.

Rawls, J., (1997), “The Idea of Public Reason,” in Bohman and Rehg (1997), pp. 93-141.

Rhode, P. W., and Strumpf, K. S., (2009), “Manipulating Political Stock Markets: A
Field Experiment and a Century of Observational Data,” manuscript, February
2009.

Roberts, R. C., and Wood, W. J., (2007), Intellectual Virtues: An Essay in Regulative
Epistemology, New York: Oxford University Press.

Rønnow-Rasmussen, T., (2002), “Hedonism, Preferentialism, and Value Bearers,” The

Journal of Value Inquiry 36, pp. 463-472.

174
Sartwell, C., (1992), “Why Knowledge is Merely True Belief,” Journal of Philosophy 89,
pp. 167-180.

Scott, D., (2006), Plato’s Meno, Cambridge: Cambridge University Press.

Servan-Schreiber, E., Wolfers, J., Pennock, D., and Galebach, B., (2004), “Prediction
Markets: Does Money Matter?” Electronic Markets 14, pp. 243–51.

Sieck, W., and Arkes, H., (2005), “The Recalcitrance of Overconfidence and its
Contribution to Decision Aid Neglect,” Journal of Behavioral Decision Making
18, pp. 29-53.

Sober, E., (2001), “What is the Problem of Simplicity?” in A. Zellner, H. Keuzenkamp,

and M. McAleer, eds., Simplicity, Inference and Modelling: Keeping It
Sophisticatedly Simple, Cambridge: Cambridge University Press, pp. 13-31.

Sosa, E., (1991), “Knowledge and Intellectual Virtue,” in Knowledge in Perspective:

Selected Essays in Epistemology, Cambridge: Cambridge University Press, pp.
225-244.

Sosa, E., (2004), “Replies,” in J. Greco, ed., Ernest Sosa and his Critics, Malden:
Blackwell, pp. 329-333.

Sosa, E., (2007), A Virtue Epistemology: Apt Belief and Reflective Knowledge, Volume I,
Oxford: Clarendon Press.

Steup, M., and Sosa, E., (2005), eds., Contemporary Debates in Epistemology, Malden:
Blackwell Publishing.

Stich, S., (1990), The Fragmentation of Reason: Preface to a Pragmatic Theory of

Cognitive Evaluation, Cambridge: The MIT Press.

Stiell, I. G., Greenberg, G. H., McKnight, R. D., Nair, R. C., McDowell, I, and
Worthington, J. R., (1992), “A Study to Develop Clinical Decision Rules for the
Use of Radiography in Acute Ankle Injuries,” Annals of Emergency Medicine
21(4), pp. 384-390.

Sunstein C., (2006), “Deliberating Groups versus Prediction Markets (or Hayek’s
Challenge to Habermas),” Episteme 3 (3), pp. 192-213.

Sunstein, C., and Thaler, R., (2008), Nudge: Improving Decisions about Health,
Happiness, Wealth, and Happiness, New Haven: Yale University Press.

Swinburne, R., (1999), Providence and the Problem of Evil, New York: Oxford
University Press.

175
Taylor, C. C. W., (2008), “Plato’s Epistemology,” in G. Fine (2008), pp. 165-190.

Taylor, S. E., (1989), Positive Illusions: Creative Self-Deception and the Healthy Mind,
Basic Books.

Trout, J. D., (2005), “Paternalism and Cognitive Bias,” Law and Philosophy 24, pp. 393-
434.

Trout, J. D., (2009), The Empathy Gap: Building Bridges to the Good Life and the Good
Society, New York: Viking.

Whitcomb, D., (forthcoming), “Epistemic Value,” in A. Cullison, ed., The Continuum

Companion to Epistemology, Continuum.

Whitecotton, S. M., (1996), “The Effects of Experience and Confidence on Decision Aid
Reliance: A Causal Model,” Behavioral Research in Accounting 8, pp. 194–216.

Williams, B., (1978), Descartes: The Project of Pure Enquiry, Highlands: Humanities
Press.

Williamson, T., (2000), Knowledge and Its Limits, New York: Oxford University Press.

Wolfers, J., and Zitzewitz, E., (2006), “Prediction Markets in Theory and Practice,”
Research Paper No. 1927, Stanford Graduate School of Business Research Paper
Series.

Wolterstorff, N., (1996), John Locke and the Ethics of Belief, Cambridge: Cambridge
University Press.

Wolterstorff, N., (2005), “Obligation, Entitlement, and Rationality,” in M. Steup and E.

Sosa (2005), pp. 326-338.

Young, R., (2008), “John Stuart Mill, Ronald Dworkin and Paternalism,” in C. L. Ten,
ed., Mill’s ‘On Liberty’: A Critical Guide, Cambridge: Cambridge University
Press.

176