Gage Epistemology QD Fus-qd0802-All

QUAESTIONES DISPUTATAE
FRANCISCAN UNIVERSITY OF STEUBENVILLE

QUAESTIONES
QUAESTIONES
DISPUTATAE
1235 UNIVERSITY BLVD
DISPUTATAE
STEUBENVILLE, OH 43952
WWW.QUAESTIONESDISPUTATAE.COM
SPECIAL ISSUE ON EPISTEMOLOGY

SPECIAL ISSUE ON EPISTEMOLOGY
A JOURNAL OF
PHILOSOPHICAL INQUIRY
AND DISCUSSION
Issue Edited by
Logan Paul Gage
VOLUME 8
N O. 2
ISSN: 2150-5756 VOLUME 8, NO. 2 SPRING 2018

Quaestiones Disputatae Quaestiones Disputatae is a journal of philosophy inspired by the medieval
dialectical form of the “disputed question”: a method of philosophical dis-
Editor: cussion aimed at addressing the relevant issues of the time. In the spirit of
Alex Plato, Franciscan University of Steubenville the medieval quaestiones disputatae, this journal addresses significant questions
and topics of contemporary philosophic interest. Recognizing that signifi-
Board of Advisors: cant philosophical contributions are found among philosophers of various
backgrounds and outlooks, the journal is, therefore, not limited to a particu-
John Deely, University of St. Thomas, Houston lar period of philosophy, nor is it representative of a specific philosophic
Alfred Fredosso, University of Notre Dame school or point of view.
Robert P. George, Princeton University
Stephen Gersh, University of Notre Dame Each issue has a special theme. An issue may focus on a topic in a particular
Christina M. Gschwandtner, University of Scranton area of philosophy, such as logic, philosophy of science, political philosophy,
Siobhan Nash-Marshall, Manhattanville College or aesthetics; on the work of a significant philosopher such as Proclus, Duns
Timothy Noone, Catholic University of America Scotus, or Husserl; on the examination and study of an important text of
Giorgio E. Pini, Fordham University philosophy such as the Enneads of Plotinus, or Formalism in Ethics and Non-
Nicholas Rescher, University of Pittsburgh Formal Ethics of Values by Max Scheler; or on the valuable contributions of a
Linda Zagzebski, University of Oklahoma particular philosophical school or movement such as Cambridge Platonism,
Scottish Realism, or Paduan Aristotelianism.
Subscription Information:
Quaestiones Disputatae aims at a genuine philosophia perennis that respects the
The Catholic University of America Press distributes Quaestiones Disputatae. tradition of philosophy but also recognizes philosophy to be a dynamic and
Those interested in subscribing may inquire at https://www.cuapress.org/ progressive discipline, one that is more than its history. In this way, it can be
journals/quaestiones-disputatae/, via email at jrnlcirc@press.jhu.edu, or by a forum for philosophers to discuss and share their work, and to participate
calling 410-516-6987 or 1-800-548-1784. Individual print subscriptions are in a meaningful dialogue with others.
$42.50 per year and institutional print subscriptions are $110 per year. Please
visit our website for additional prices. Single issues are $30 for individuals
and $70 per issue for institutions. No additional shipping and handling costs
for U.S. purchases, but please add $20 for international shipping. Discover, Contact Information:
MasterCard, Visa or American Express are accepted. Quaestiones Disputatae is All correspondence concerning Quaestiones Disputatae, submissions, or the
also available electronically through The Philosophy Documentation Center Annual Conference in Christian Philosophy hosted by Franciscan University
at http://www.pdcnet.org/qd. Checks may be made out to “The Catholic of Steubenville should be directed to:
University of America Press” and sent to:
Quaestiones Disputatae
JHUP Journals Division Alex Plato, Editor
P.O. Box 19966 Quaestiones Disputatae
Baltimore, MD 21211-0966 Franciscan University of Steubenville
1235 University Blvd
Steubenville, OH 43952
qdphilosophyjournal@gmail.com
Copyright © 2018 Quaestiones Disputatae

ISSN: 2150-5756
Quaestiones
Disputatae
Special iSSue on epiStemology
Logan PauL gage
(Franciscan university oF steubenviLLe),
sPeciaL guest editor
2018
2
Vol. 8, No. 2 Spring 2018
Special Issue on Epistemology

Editor’s Introduction
Logan Paul Gage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Newman on the Grounds of Faith

Frederick D. Aquino . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Complexly Based Beliefs and the Generality Problem for Reliabilism

Max Baker-Hytch. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Preemptionism and Epistemic Authority

Donald J. Bungum. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
The Unassertability of Contextualism

Martijn Blaauw and Jeroen de Ridder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Can Experience Fulfill the Many Roles of Evidence?

Logan Paul Gage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
The World, the Deceiver, and The Face in the Frost

Lydia McGrew . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Fine-Tuning and the Search for an Archimedean Point

Timothy McGrew . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Proper Functionalism and the Metalevel:

A Friendly Reply to Timothy and Lydia McGrew
Tyler Dalton McNabb. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Defeating Objections to Bayesianism by Adopting a

Proximal Facts Approach
Calum Miller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
The Myth of a Pure Virtue Epistemology

Joshue Orozco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Editor’s Introduction
Logan Paul Gage

Guest Editor
It is with great pleasure that I bring you this special issue of Quaestiones Dis-
putatae (QD). Looking back upon previous issues of this journal, what strikes
one most is the sheer diversity of topics covered: the thought of fascinating
figures such as phenomenologist Edith Stein and Thomist W. Norris Clarke,
historical essays on Neo-Platonism and Aristotle’s categories, and discus-
sions of distributism and Catholic bioethics. With this new set of essays, QD
explores yet another branch of philosophy: that of contemporary episte-
mology in the analytic tradition.
Analytic philosophy is still thought by many to ignore the deepest
concerns of the Western philosophical tradition; it is thought to be obsessed
with linguistic analysis rather than with being, meaning, and truth. To be
fair, much of early analytic philosophy could easily have given rise to these
stereotypes. Yet anyone familiar with analytic philosophy over the last half
century knows that this is a caricature at best. Bertrand Russell and A. J. Ayer
were analytic philosophers, but so are Alasdair MacIntyre, Alvin Plantinga,
and John Haldane. Contemporary work on ethics, metaphysics, and philoso-
phy of mind has taken up perennial questions of virtue and vice, the nature
of necessity, the existence of God, and free will. And the best of this work
has often been clear about its debt to the great thinkers of the past, such as
St. Thomas Aquinas.
Contemporary epistemology is no different. Current analytic episte-
mology focuses on the nature of knowledge, the nature of justification for
our beliefs, testimony, a priori knowledge, perceptual knowledge, and the like.
The present collection of essays spans a wide array of topics, but each essay
touches on an issue of traditional concern either directly or indirectly. Fred
Aquino explores the thought of John Henry Newman in relation to contem-
porary categories of religious epistemology. Tim McGrew and Calum Miller,
while utilizing technical tools of probability theory, touch on issues important
to philosophy of religion (namely, the fine-tuning of the laws of physics and
the multiverse). The essays of Max Baker-Hytch, Martijn Blaauw and Jeroen
de Ridder, Donnie Bungum, Lydia McGrew, Tyler McNabb, Josh Orozco,
and the present writer all bear on fundamental epistemological concerns such
© Logan Paul Gage, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)

4 editor’S introduction
as evidence, testimony, and important theories about knowledge and reason-

able belief such as evidentialism, contextualism, proper functionalism, and
virtue epistemology.
I am grateful to Alex Plato, the general editor of QD, for the freedom
to explore the topics found in this issue and for his many helpful comments
and suggestions on these essays. To Franciscan University and the readers of
this journal, I am grateful for your continued support of this philosophical
forum. Lastly, I want to express my deep gratitude to Baylor University for
its generous support during my graduate studies, which allowed me to travel
and present at numerous conferences. In reflecting upon the present volume,
such support, as well as that of hosting institutions like the Tyndale House at
Cambridge University, the European Epistemology Network, the American
Philosophical Association, and the American Catholic Philosophical Associ-
ation, has played such a vital role in connecting me with the contributors to
this volume. I am grateful to these many institutions but most of all to each
of the contributors for making this special issue possible.
Logan Paul Gage

Franciscan University of Steubenville
Newman on the Grounds of Faith
Frederick D. Aquino
Abilene Christian University
An epistemological issue that preoccupied John Henry Newman was the

conditions under which Christian belief can be considered rational. As
he sought to offer a broader and more refined account of faith and reason, he
focused, for example, on the informal nature of reasoning and on the role of
personal judgment in assessing evidence. In particular, his approach homed
in on how the mind actually works and the conditions under which people
reason within various contexts and fields of knowledge.
Along these lines, an important, though complex, issue involves clari-
fying Newman’s position on the grounds of faith. In this respect, Anthony
Kenny says the University Sermons contain some of Newman’s “very best work
on the nature and justification of faith.”1 However, Newman’s position on
the grounds of faith in some of the University Sermons is difficult to capture
and, perhaps, prone to misunderstanding (e.g., serm. 10). As I hope to show,
Newman provides greater clarification of his own position on the grounds
of faith in sermon 13 (see also serm. 14).
Accordingly, I will structure this essay in the following way. First, I will
identify some potential misunderstandings in the University Sermons concern-
ing Newman’s position on the grounds of faith. Second, I will show how the
distinction between implicit and explicit reason in sermon 13 shapes both
his rejection of a particular kind of hard rationalism (a religious belief is
rational if and only if it can be articulated or demonstrated formally) and
his alternative understanding of the grounds of faith. Implicit reason, for
Newman, is a spontaneous, unconscious, or unargumentative process of
reasoning by which people form beliefs without appealing to explicitly stated
grounds; explicit reason is a second-order activity that works out whether
beliefs are true rather than false; the former is unreflective, while the latter
has a reflective component.2 Third, I will argue constructively that Newman
1
Anthony Kenny, “Newman as a Philosopher of Religion,” in Newman: A Man
for Our Time, ed. David Brown (Harrisburg, PA: Morehouse, 1990), 101.
2
According to Michael Peterson, William Hasker, Bruce Reichenbach, and Da-
vid Basinger, Reason & Religious Belief: An Introduction to the Philosophy of Religion, 5th
© Frederick D. Aquino, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)

6 newman on the groundS of faith
is best construed as a soft rationalist insofar as he thinks that (1) faith is

subject to rational analysis, (2) its process of reasoning is cumulative, though
not necessarily formal, in nature, and (3) judgment plays an irreducible role
in evaluating evidence and forming arguments. Fourth, I will briefly place
Newman’s proposal within the contemporary debate between fideism and
rationalism.
Potential Misunderstandings
In the University Sermons, Newman offers some rich epistemological insights

concerning the conditions under which Christian belief can be considered
rational. However, the polemical edge of these sermons may contribute to
(or result in) some misleading conclusions.3 To do constructive justice
to them, one has to tease out and develop the relevant epistemological hints.
As Newman himself points out, the discourses on faith and reason show a
progression and development of thought, especially in terms of providing
ed. (New York: Oxford University Press, 2013), 61, hard rationalism (or as they call
it, strong rationalism) holds that “for a religious belief system to be properly and rationally
accepted, it must be possible to prove that the belief system is true . . . [or] show that a belief
is true in a way that should be convincing to any reasonable person.” Some hard rationalists
include W. K. Clifford and John Locke; Peterson et al. argue that the contempo-
rary philosopher Richard Swinburne is very close to the definition at hand. In this
respect, William J. Abraham, An Introduction to the Philosophy of Religion (Englewood
Cliffs, NJ: Prentice Hall, 1985), 116, says that Swinburne and the soft rationalist
“disagree on whether the cumulative case for religious belief can be quantified or
codified. Swinburne thinks that it can be laid out according to the formal calculus of
confirmation theory.” A key issue for Newman is that rationality is not reducible to
formal argumentation. Some soft (or critical) rationalists include William Abraham
(An Introduction to the Philosophy of Religion, esp. ch. 9), Basil Mitchell (The Justification
of Religious Belief [New York: Seabury Press, 1973]), and Peterson et al. (Reason &
Religious Belief, 73).
3
For example, Newman’s seemingly quick dismissal of evidential theology in
sermon 10 confused and disturbed James Anthony Froude; see The Letters and Diaries
of John Henry Newman, Vol. 29: The Cardinalate: January 1879 to September 1881, ed.
Charles Stephen Dessain and Thomas Gornall, S.J. (Oxford: Clarendon Press, 1976),
351f. (henceforth cited as LD); LD xxxi, 300f.; and Geertjan Zuijdwegt, “Scepticism
and Credulity: Victorian Critiques of John Henry Newman’s Religious Apologetic,”
Journal for the History of Modern Theology 20, no. 1 (2013): 61–84. On the polemical na-
ture of the University Sermons, see the editors’ introduction in John Henry Newman,
Fifteen Sermons Preached before the University of Oxford: Between A.D. 1826 and 1843, ed.
James David Earnest and Gerard Tracey (Oxford: Clarendon Press, 2006); hence-
forth cited as US.
frederick d. aquino 7
precise definitions and clarifying the relationship between faith and rea-
son. They “are of the nature of an exploring expedition into an all but
unknown country, and do not even venture on a definition of either Faith or
Reason on starting. As they proceed, however, they become more precise, as
well as more accurate” (US pref.; see also US 10.45). For instance, one gets a
clearer sense of Newman’s own position on the grounds of faith in sermon
13 but not necessarily in sermon 10.
While Newman clearly thinks that faith is rational, some passages in
the University Sermons seem to complicate things on this front. For example,
Newman says that faith is “mainly swayed by antecedent considerations . . .
by previous notices, prepossessions, and (in a good sense of the word)
prejudices” (US 10.26). In terms of “when the mind savingly believes, the
reasoning which that belief involves, if it be logical, does not merely proceed
from the actual evidence, but from other grounds besides” (US 11.10). The
“desire” of faith “is its main evidence . . . Not that it has no grounds in
Reason, that is, in evidence; but because it is satisfied with so much less
than would be necessary” (US 10.34). These passages seem on the surface to
suggest that Newman thinks that faith is not based primarily on evidence but
instead on antecedent assumptions, hopes, and desires.
Is Newman, then, proposing that faith operates more in line with one’s
antecedent considerations or assumptions than in accordance with the rele-
vant evidence? Or is he simply trying to capture, and offer some correc-
tives of, a popular conception of the relationship between faith and reason
before providing a constructive alternative? The latter seems to be more
plausible; it makes sense and fits well with one of the aims of the Univer-
sity Sermons—namely, to examine existing accounts of faith and reason and
offer a broader and more plausible alternative. On the one hand, in making
a distinction between antecedent assumptions and the actual evidence,
Newman seems to be challenging a narrowly construed notion of evidence
as “easily stateable and widely accessible,” not necessarily disparaging an
appeal to evidence.4 Apart from the danger of pursuing religious matters
without a proper orientation, “the Evidences,” according to Newman, may
be of “great service to persons in particular frames of mind” (US 10.44). On
the other hand, he seems to be drawing attention to the fact that antecedent
considerations (in both religious and non-religious worldviews) play a role in
evaluating and interpreting the force of evidence, and so “in deciding what
is to be believed on the strength of it we are rightly influenced by consider-
ations other than those provided by the evidence.”5 Although the evidence
4
William J. Wainwright, Reason and the Heart: A Prolegomenon to a Critique of Pas-
sional Reason (Ithaca, NY: Cornell University Press, 1995), 81.
5
Basil Mitchell, Faith and Criticism (Oxford: Clarendon Press, 1994), 14.
itself “does not vary in force, the antecedent probability attending it does
vary without limit” according to a person’s intellectual complexion, training,
and “the temper of the mind surveying it” (US 10.36).
Newman recognizes that a person may believe p because of certain
wishes, desires, or dispositions. Though such wishes, desires, and dispositions
can certainly get in the way of deciphering correctly (or forming reliably)
true beliefs, they may have “positive epistemic value.”6 Newman thinks that
some antecedent assumptions or a properly disposed mind can put one in a
place to perceive things correctly or even perceive something that otherwise,
with different antecedent assumptions or a different mind-set, would not be
perceived at all.7 What matters is whether a particular kind of commitment or
a properly disposed mind puts one in a better place to evaluate evidence and
thus increases reliability in terms of acquiring true rather than false beliefs.
In addition, Newman does not really develop his own account of the rela-
tionship between faith and reason in sermon 10. Instead, he simply unpacks
and engages with a common or popular contrast of faith and reason as two
habits of mind, writing, “Next, I observe, that, whatever be the real distinction
and relation existing between Faith and Reason, which it is not to our purpose
at once to determine, the contrast that would be made between them, on a
popular view, is this,—that Reason requires strong evidence before it assents,
and Faith is content with weaker evidence . . . To conclude: It will be observed,
I have not yet said what Reason really is, or what is its relation to Faith, but
have merely contrasted the two together, taking Reason in the sense popularly
ascribed to the word” (emphasis mine; US 10.17, 45). So, Newman is concerned
in this sermon not with developing his own account of faith and reason but
instead with clarifying what a popular conception entails. According to this
popular interpretation, faith is simply “a moral quality,” “feeling, or sentiment”
that is “dependent upon Reason” (US 10.12, pref. 2; see also US 11.1). The
contrast, then, is between faith, which proceeds mainly with “conjectures or
6
Wainwright, Reason and the Heart, 73. Wainwright argues that passional com-
mitments are not necessarily detrimental to the process of assessing religious belief.
They may play a role in shaping our evaluation of the evidence; see also Mark Wynn,
“The Relationship of Religion and Ethics: A Comparison of Newman and Contem-
porary Philosophy of Religion,” Heythrop Journal 46 (2005): 438.
7
Thanks to Alex Plato for his help in clarifying this point. In “An Educated
Conscience: Perception and Reason in Newman’s Account of Conscience,” Studies
in the Literary Imagination 49, no. 2 (2016): 63–80, I show how Newman’s notion of
an educated conscience presumes that proper formation can in fact enhance one’s
moral perception. See also Frederick D. Aquino, “Trained Perception: A Construc-
tive Look at John Cassian,” in Sensing Things Divine: Towards a Constructive Account of
Spiritual Perception, ed. Frederick D. Aquino and Paul Gavrilyuk (Oxford: Oxford Uni-
versity Press, 2019), 1–25.
presumptions,” and reason, which proceeds mainly with “proofs” (US pref.
2; see also US 10.17, 12.3). Consequently, faith depends on and follows “a
distinct act of Reason beforehand,—Reason warranting, on the ground of
evidence, both ample and carefully examined, that the Gospel comes from
God, and then Faith embracing it” (US 11.1; see also US 10.17–20, 12.3).8
Newman rejects the claim that faith is essentially a matter of feeling and
sentiment.9 In fact, he argues that the popular view operates with a narrow
conception of reason and is insufficiently empirical. Alternatively, an adequate
account of faith and reason should “take things as we find them” rather
than impose an a priori notion of reason (or what one takes to be norma-
tive from one’s own perspective) on the actual processes of belief-formation
(US 12.12). In seeking to carve out a robust account of the rationality of
religious belief, Newman draws attention to how reason works in everyday
life. His approach “was to examine how in fact people made up their minds
on non-religious issues and argue that by the same standards religious beliefs
were justified.”10 In everyday life, most people are rarely afforded the oppor-
tunity of rigorously scrutinizing propositions before they assent to them, and
yet they seem to hold a host of rational beliefs. As a result, Newman seeks
to map a broader account. In so doing, he does not reject the role of reason
and evidence in the life of faith. Rather, as we will see, he challenges the
assumption that reason and/or evidence is synonymous with the capacity to
explain faith in formal arguments.11 More specifically, he calls into question
the hard rationalist claim that a religious belief is rational if and only if it can
8
In the preface of the third edition (1872) to the University Sermons (US), New-
man summarizes the popular view in the following way: “Faith is the judging on weak
grounds in religious matters, and Reason on strong grounds . . . by Faith is meant a
feeling or sentiment, by Reason an exercise of common sense; Faith is conversant
with conjectures or presumptions, Reason with proofs” (US pref. 2).
9
See John Henry Newman, Apologia Pro Vita Sua, ed. Martin Svaglic (Oxford:
Clarendon Press, 1967), 54.
10
Peter Forrest, “The Epistemology of Religion,” Stanford Encyclopedia of Phi-
losophy, April 23, 1997, updated May 26, 2017, https://plato.stanford.edu/entries/
religion-epistemology. See also Basil Mitchell, “Newman as a Philosopher,” in New-
man after a Hundred Years, ed. Ian T. Ker and Alan G. Hill (Oxford: Clarendon Press,
1990), 223–46; Frederick D. Aquino, “Epistemology,” in The Oxford Handbook of John
Henry Newman, ed. Frederick D. Aquino and Benjamin J. King (Oxford: Clarendon
Press, 2018), 375–94. Cf. Duncan Pritchard, “Wittgenstein on Faith and Reason: The
Influence of Newman,” in God, Truth, and Other Enigmas, ed. Miroslaw Szatkowski
(Berlin: De Gruyter, 2015), 197–216.
11
See US 13.9–15; and John Henry Newman, The Theological Papers of John Henry
Newman on Faith and Certainty, ed. Hugo M. de Achaval and J. Derek Holmes (Oxford:
Clarendon Press, 1976), 84.
be articulated (or demonstrated) formally. Thus, his dissatisfaction with exist-

ing accounts of rationality stems from several defects, such as the failure to
consider the ways in which the whole character of the person factors in the
evaluation of evidence, its restriction of faith to those who are intellectually
capable of following the relevant arguments, and the disconnect between the
notion of what human rationality ought to be like and how in fact concrete
individuals actually think.
Grounds of Faith
Some of the University Sermons, as we have seen, can be misleading. However,

it is in sermon 13 that Newman offers a deeper analysis of the natural work-
ings of reason and thus provides greater clarity on the grounds of faith and
the relationship between faith and reason. What seems to be rejected out of
hand is the claim that faith requires the operation of an explicit kind of rea-
soning. Though reason plays an important role in evaluating the process of
belief-formation, it does not follow that faith springs from a formal account,
nor does it follow that reason is reducible to an explicit kind of reasoning
and that faith is dependent upon this kind of reasoning. Being aware of
how reason operates, then, is not a precondition to having rationally accept-
able beliefs. Faith, as a tacit or implicit kind of reasoning, is “independent
of and distinct from what are called philosophical inquiries, intellectual sys-
tems, courses of argument, and the like” (US 11.16; see also US 11.9, 12.1).
In other words, reflection is “a natural faculty,” but it is not “an initial fac-
ulty.” However, faith is not groundless. It is “independent not of objects or
grounds . . . but of perceptible, recognized, producible objects and grounds.”
As a result, faith “admits, but does not require, the exercise” of explicit rea-
soning (US 13.4). In other words, the reasoning of faith includes but is not
reducible to explicit reasoning.12
12
In the Grammar of Assent, for example, Newman takes up and corrects two
epistemological claims: (a) one is rationally entitled to believe p if and only if one can
fully understand p and (b) one is rationally entitled to believe p if and only if one has
demonstrative proof of p. In this respect, the larger project of the Grammar of Assent
aims to show how Christian belief is rational without explicit awareness of how the
mind understands claims of faith and without demonstrative proof. See John Henry
Newman, Essay in Aid of a Grammar of Assent, ed. Ian T. Ker (Oxford: Clarendon
Press, 1985), 319, 321; henceforth cited as GA. Accordingly, the Grammar of Assent
challenges the notion that a necessary condition for rationally acceptable belief is
that one must have epistemic access to the features that make that belief rationally
acceptable. See John R. T. Lamont, “Newman on Faith and Rationality,” International
Journal of Philosophy of Religion 40 (1996): 63–84. Newman shows that people render
What, then, is Newman’s position on the grounds of faith? William

Wainwright qualifies the kind of evidentialism that Newman seems to reject.
If evidentialism “requires stateable and publicly accessible evidence that
compels assent regardless of a person’s moral temper,” then Newman “is not
an evidentialist.” Nevertheless, Newman is convinced that “properly formed
religious beliefs are based on (sufficient) evidence, and that those whose
noetic faculties are functioning as they should will find this evidence convinc-
ing.”13 This accords well with his notion of a properly shaped disposition. In
other words, passional commitments may enable one to be properly open to
the evidential basis of faith.
In general, Wainwright’s observation seems to be a fair and apt appraisal
of Newman’s approach to the grounds of faith. He rightly points out that
Newman is rejecting a particular kind of evidentialism, not necessarily the
claim that faith must have grounds. However, it is not my concern in this
section to determine whether Newman is an evidentialist or what kind of
evidentialism his project may entail. Instead, I want to clarify some important
aspects of Newman’s understanding of the grounds of faith. Toward this
end, I will focus on three features in Newman’s account.
First, Newman does not think that faith is impervious to critical reflec-
tion. In fact, he argues that it is problematic to exempt faith from ratio-
nal analysis. Though faith is the “simple lifting of the mind to the Unseen
God, without conscious reasoning or formal argument, still the mind may be
allowably, nay, religiously engaged, in reflecting upon its own Faith; investi-
gating the grounds and the Object of it, bringing it out into words, whether
to defend, or recommend, or teach it to others” (US 13.3). However, it is
also clear that Newman does not think that everyone needs to be engaged in
this kind of rational reflection or that “all who have faith should recognize,
and be able to state what they believe, and why” (US 13.4). The last thing
Newman intends to convey is a narrow understanding of belief-formation,
a theory that categorically denies the validity of the implicit reasoning of
faith. In many ways, Newman presumes that people largely employ a kind
of implicit reasoning and most likely defer to (trust) those who are capable
of rendering apt judgments concerning the subject at hand.14
unconditional assents to propositions that fall short of demonstrative proof and

admit of nothing higher than probable reasoning.
13
Wainwright, Reason and the Heart, 81.
14
For the role that epistemic dependence plays in the formation and suste-
nance of faith, see The Theological Papers of John Henry Newman on Faith and Certainty,
26; and Frederick D. Aquino, Communities of Informed Judgment: Newman’s Illative Sense
and Accounts of Rationality (Washington, DC: Catholic University of America Press,
2004), esp. ch. 4. In contemporary terms, Newman seems to presume the division
Second, Newman broadens the scope of what counts as rational by

making a crucial distinction between implicit and explicit reasoning. Though
Newman grants an explicitly (second-order) rational process of reflection,
such accounts of belief-formation do not exhaust the operation of the
mind’s capacity for reasoning. In this respect, Newman rejects the claim that
all the grounds of a belief accord with an explicit awareness of the features
that make a belief reasonable or that these grounds need to be cognitively
accessible to the believer. A person may rationally believe p without providing
reasons for believing p. Newman, then, makes a distinction between (a) faith
having grounds for believing p and (b) articulating the grounds for believing p
(whether to oneself or to others). In other words, the operation of the mind
is not reducible to the capacity to formalize one’s reasoning.
Third, in making a distinction between implicit and explicit reasoning,
Newman does not think that all forms of implicit reasoning are necessarily
truth-conducive. More specifically, explicit reasoning is a second-order
activity that works out whether beliefs are true rather than false. It plays
an important role in the life of faith, offering a retrospective analysis of
experiences, practices, and antecedent considerations that factor into form-
ing Christian belief. In this sense, explicit reasoning is “the process of inves-
tigating our reasonings” (US 13.9). Notwithstanding, the point at hand is
that you cannot assume that a belief is necessarily irrational simply because
a person is unable to provide the explicit reasons for that belief. Reasoning,
however, is not synonymous with proficiency in logical argument; it is pres-
ent outside demonstrative proof (contra the popular view). Most people form
beliefs without epistemic access to ways in which the mind justifies beliefs.
Instead, they operate from a presumptive level of reasoning, acknowledging
the strong probability of reliable belief-forming processes (e.g., testimony,
memory, sense perception, and experience).
If explicit reasoning is the bar of reason, then most beliefs (religious and
non-religious) are not rational. More exactly, Newman employs a parity argu-
ment to challenge the notion that faith, as a presumptive mode of reason-
ing, is a unique phenomenon. He highlights the ways in which presumptive
reasoning factors into the formation of religious and non-religious beliefs.
So, faith is “not the only exercise of Reason, which, when critically examined,
would be called unreasonable, and yet is not so” (US 11.12). There is ample
empirical evidence that most people operate on the level of presumptive
of cognitive labor. Along these lines, an overlooked theme in the Grammar of Assent
is Newman’s thoughts on epistemic duty. In the context of his comments about the
proper functioning of the illative sense, Newman says that in the case of mature or
educated minds, fulfillment of epistemic duty is “an obligation, or rather a necessity.”
It requires the “exercise” of “mature judgment” (GA 126).
reasoning until “antecedent probabilities fail” (US 10.28). They dispense with
the need for demonstrative proof and follow the dictum that probability is
the way of life in everyday affairs.15
Overall, Newman’s point seems to be that the activity of the mind is
richer and subtler than the kind of reflexive activity in explicit reasoning. His
account of faith and reason includes implicit and explicit reasoning. Further-
more, Newman thinks that faith “cannot exist without grounds or without
an object” (US 13.4). As we have seen, he thinks it is important to assess the
original or implicit process of reasoning. Assumed here is the recognition that
not all grounds are necessarily adequate—that is, truth-conducive.16 However,
he rejects the conflation of grounds with the capacity to state those grounds
in argumentation. People should not be considered irrational simply because
they are incapable of offering an explicitly rational understanding of an implicit
process of reasoning. Reflexive analysis is not a prerequisite to being rational.
Newman as a Soft Rationalist
In an essay on the importance and relevance of Newman as a philosopher,

Basil Mitchell describes Newman as “a rationalist of some kind, albeit one
who is sensitive to the many different ways in which rationality can be man-
ifested.”17 Within his own context, Newman sought to chart a path between
fideism (faith is exempt from or impervious to rational analysis) and a par-
ticular kind of hard rationalism (a belief is rational if and only if it can
be articulated or demonstrated formally). In contemporary and more pre-
cise terms, Newman’s account of faith and reason in the University Sermons,
especially as expressed in sermons 13 and 14, fits what contemporary phi-
losophers call soft rationalism.18 Elements of soft rationalism include (1) the
willingness to subject faith to rational analysis, (2) an emphasis on the cumu-
lative nature of reasoning, and (3) the recognition of the irreducible role that
15
On the objection that Newman’s appeal to antecedent probabilities supplies
no intelligible rule for deciding between a genuine and a counterfeit revelation or
for adjudicating one’s beliefs in a publicly accessible manner, see Anthony Kenny,
“Newman and Victorian Doubt,” New Blackfriars 92, no. 1038 (2011): 159; and Jay
Newman, The Mental Philosophy of John Henry Newman (Waterloo, Ontario: Wilfrid
Laurier University Press, 1986), 139–50. For a response to Jay Newman, see Wain-
wright, Reason and the Heart, 75.
16
For a contemporary account of adequate grounds, see William P. Alston,
Beyond Justification: Dimensions of Epistemic Evaluation (Ithaca, NY: Cornell University
Press, 2005), esp. ch. 5.
17
Mitchell, “Newman as a Philosopher,” 240.
18
I think this is also the case in the Grammar of Assent.
judgment plays in evaluating evidence and forming arguments. With this in

mind, I will briefly spell out three features of soft rationalism and show how
they dovetail nicely with Newman’s thought.19
The first feature of soft rationalism is that faith is not exempt from or
impervious to rational scrutiny. Though there is a willingness to rationally
evaluate one’s own faith, such a process of evaluation is shaped by one’s
training, “attitudes, basic convictions and previous experiences.”20 It is not as
simple as proportioning our context-less beliefs to the evidence at hand. The
commitments we bring to the evaluation of evidence and argumentation are
religiously or non-religiously loaded.
As we have seen, Newman thinks that rational reflection plays an impor-
tant role in sizing up our beliefs. Faith has grounds, but not all grounds are
necessarily adequate. In the Grammar of Assent, for example, Newman makes
a comparable point by distinguishing simple from complex assent. Simple
assent refers to the process of assenting to things without appealing to
explicitly stated grounds or, to put it in contemporary terms, without fulfill-
ing some prior grounding requirement. Simple assent is “exercised uncon-
sciously” (GA 124), and so it involves the mind’s acceptance of propositions
without conscious reflection of the grounds upon which the assent is made.
Accordingly, we believe p without employing the full-blown powers of justifi-
cation (without providing reasons for believing p), or at least not all processes
of belief-formation are subject to our awareness. On the other hand, complex
assent involves deliberate recognition (or reflexive awareness) of the grounds
upon which an assent is made. Complex assent reflects the activity of more
mature or educated minds, “investigations into the argumentative proof of
the things to which they have given their assent” (GA 126). Newman thinks
that not all simple assents are necessarily reliable or conducive to truth, and
so they can be overturned upon further reflection. Complex assent differs
from simple assent in that it uncovers (reflexively or deliberately) cognitive
19
My constructive reading of Newman as a soft rationalist is shaped by Abra-
ham, Introduction, esp. ch. 9.; Peterson, Hasker, Reichenbach, and Basinger, Reason &
Religious Belief, esp. ch. 4; and C. Stephen Evans, “Critical Dialog in Philosophy of
Religion,” in Philosophy of Religion: Selected Readings, 3rd ed. (New York: Oxford Uni-
versity Press, 2007), 123–29. Though they use different terminology (soft rational-
ism, critical rationalism, and critical dialog, respectively), there seems to be sufficient
compatibility or overlap. In this essay, however, the operative term will be soft
rationalism.
20
Evans, “Critical Dialog,” 128, argues that it is “a mistake to think that this
process of testing can or should proceed from a totally neutral standpoint, the stand-
point of a person without any convictions.” See also Peterson, Hasker, Reichenbach,
and Basinger, Reason & Religious Belief, 70f.; and Abraham, Introduction, 106–10.
inaccuracies and determines whether a proposition yields true rather than

false beliefs (or judgments).
In addition, Newman retains a role for evidence and argument while under-
scoring the cognitive significance of commitment or antecedent assumptions.
Newman thinks that construing reason without commitment, as the hard ratio-
nalists seem to require, sets the standards too high for evaluating religious belief.21
In my estimation, this is what Newman’s parity argument seeks to accomplish.
What Newman calls into question, then, is an idealized (if not impossible) set
of rational demands, not reason itself. In one sense, Newman thinks that faith
has an evidential basis, though this basis is not necessarily spelled out in provid-
ing reasons.22 Rather, he emphasizes the subtle and informal nature of reason-
ing and the ways in which our reasoning is in some sense shaped by antecedent
assumptions, attitudes, experiences, and convictions.
A second feature of soft rationalism is the recognition that religious
belief ought to be evaluated as “a rounded whole rather than taken in stark
isolation.” A religious worldview is a “complex, large scale system of belief ”
that must be viewed as a “whole” before it is evaluated.23 In determining
whether a belief is rationally acceptable, then, we consider various pieces of
evidence (instead of relying on one piece alone) and thereby seek to form a
cumulative case. Newman illustrates the cumulative nature of the reasoning
process with the example of a cable. A cable is composed of several strands;
individually, each is weak and insufficient to support a belief, but collectively,
they are as “sufficient as an iron rod” (e.g., mathematical or strict demonstra-
tion). The cable symbolizes “moral demonstration, which is an assemblage
of probabilities, separately insufficient for certainty, but, when put together,
21
On this point, Abraham, 106, says that the hard rationalist’s demand for evi-
dence that is “beyond doubt and dispute” sets the standard too high for evaluating
religious belief; see also Evans, “Critical Dialog,” 127; and Peterson, Hasker, Re-
ichenbach, and Basinger, Reason & Religious Belief, 69.
22
Peterson, Hasker, Reichenbach, and Basinger, in Reason & Religious Belief (69),
describe two kinds of critical rationalism—namely, critical evidentialism and criti-
cal anti-evidentialism. Critical evidentialists hold that in order for religious beliefs
to be rational, one “must be prepared to provide reasons and arguments in favor of
the beliefs” that one holds. In other words, having reasons is necessary to being a
reasonable religious believer. Critical anti-evidentialists do not think that one must
provide arguments in favor of their religious beliefs. They think that they have an
“adequate basis” for their beliefs “without needing any arguments in their favor.”
However, if reasonable objections are raised against any of their beliefs, they “need
to answer those objections—but they feel no need to give positive reasons in support
of their beliefs.” A fruitful area of inquiry would involve determining where to locate
Newman’s thought on the spectrum of critical rationalism.
23
Abraham, Introduction, 104.
irrefragable.” People who refuse to depend on the durability of a cable and

demand “an iron bar, would, in certain given cases, be irrational and unreason-
able.” The same applies to people who say that a religious belief is rationally
acceptable if and only if that belief can be supported by “rigid demonstra-
tion” rather than by “moral demonstration.”24
With this cumulative process of reasoning, “indirect evidence and weak
arguments, which alone would bear little weight, may be interwoven into a
fabric that strongly supports a conclusion. Each element derives warrant
from its place in the whole.”25 In talking about the illative sense in the Gram-
mar of Assent, for example, Newman says that it provides “a mental compre-
hension of the whole case, and a discernment of its upshot, sometime after
much deliberation, but, it may be, by a clear and rapid act of the intellect,
always, however, by an unwritten summing-up, something like the summa-
tion of the terms, plus and minus of an algebraic series” (GA 189). The illative
sense, for Newman, is principally a natural faculty of judgment, an informal
and tacit ability for reasoning, developing intellectual skills, and making apt
judgments. It does not merely combine various probabilities but shows how
they mutually support one another and form the best possible explanation.
A third (and related) feature is the irreducible role that judgment plays in
evaluating evidence and forming arguments. As William Abraham points out,
disputes in some fields of knowledge (e.g., history, literary exegesis, philoso-
phy, jurisprudence) require a kind of “informal reasoning,” “judgments of
plausibility,” and “an irreducible element of personal judgment, which can
be trained and rendered more sensitive.” As a result, rational evaluation is
not reducible to “strict demonstration” or “strict probability arguments.”26
The appeal to judgment is not to some personal feeling or some “new kind
of evidence which somehow adds extra weight to the whole.” Instead, the
emphasis is placed on how “the evidence is recognized and weighed in
24
LD xxi, 146. See also GA 187: “It is the cumulation of probabilities, inde-
pendent of each other, arising out of the nature and circumstances of the particular
case which is under review; probabilities too fine to avail separately, too subtle and
circuitous to be convertible into syllogisms, too numerous and various for such con-
version, even were they convertible. As a man’s portrait differs from a sketch of him,
in having, not merely a continuous outline, but all its details filled in, and shades and
colours laid on and harmonized together, such is the multiform and intricate process
of ratiocination, necessary for our reaching him as a concrete fact.”
25
Catherine Z. Elgin, Considered Judgment (Princeton: Princeton University Press,
1996), 13. See also William J. Abraham, “Cumulative Case Arguments for Christian
Theism,” in The Rationality of Religious Belief, ed. William J. Abraham and Steven W.
Holtzer (Oxford: Clarendon Press, 1987), 17–37.
26
Abraham, Introduction, 106; see also Evans, “Critical Dialog,” 128.
the scales of rational evaluation” but not constrained by or reducible to

“some sort of formal calculus.”27
As we have seen, Newman emphasizes the subtle and informal nature
of reasoning and the ways in which our reasoning is shaped by antecedent
considerations. Our reasoning proceeds “from starting-points, and with
collateral aids, not formally proved, but more or less assumed, the process
of assumption lying in the action of the Illative Sense, as applied to primary
elements of thought respectively congenial to the disputants” (GA 239). More
exactly, judgment plays a crucial role in weighing various pieces of evidence, in
seeing the interrelatedness of diverse and complex considerations that go into
the epistemological assessment of beliefs, and in determining which of these
considerations are significant and relevant to the issue at hand (e.g., in making
cumulative case arguments for religious beliefs). Along these lines, Newman’s
account of the illative sense in the Grammar of Assent offers extraordinary
explanatory power in accounting for why people differ over intellectual issues
on the basis of their training, first principles, and so on. However, this personal
mode of reasoning does not “supersede the logical form of inference, but is
one and the same with it; only it is no longer an abstraction, but carried out
into the realities of life, its premisses being instinct with the substance and the
momentum of that mass of probabilities, which, acting upon each other in
correction and confirmation, carry it home definitely to the individual case,
which is its original scope” (GA 189f.). In particular, determining the converg-
ing probabilities depends on personal judgment, and so the cumulative force
of evidence on any matter may affect people differently.
Given these three features, Newman fits the description of a soft rational-
ist. Though Newman initially engages with a popular conception of faith and
reason in sermon 10, he does not offer a constructive alternative. However,
what he seeks to do in sermon 13 is broaden our conception of what counts
as rational and thus show that faith is a kind of reasoning. As a result, his
proposal does not “reject the claims of reason as such” but demands a subtler
“appreciation of the way reason works, not only in relation to religious truth
but also in respect of all matters of serious importance.”28 In fact, Newman
thinks that faith has grounds and is subject to rational analysis, though he
rejects a hard kind of rationalism that says a belief is rational if and only if it
can be articulated or demonstrated formally.
27
William J. Abraham, “The Existence of God,” in The Oxford Handbook of Sys-
tematic Theology, ed. John Webster, Kathryn Tanner, and Iain Torrance (Oxford: Ox-
ford University Press, 2007), 34; see also William P. Alston, “The Distinctiveness of
the Epistemology of Religious Belief,” in The Rationality of Theism, ed. G. Brüntrup
and R. K. Tacelli (Dordrecht: Kluwer, 1999), 246.
28
Mitchell, “Newman as a Philosopher,” 226.
Concluding Remarks
Having established that Newman is a soft rationalist, I will briefly spell out
two implications. First, Newman’s account of faith and reason needs to be
constructively incorporated into contemporary discussions of soft (or criti-
cal) rationalism and into the larger debate (traditionally conceived) between
rationalism and fideism. His version of soft rationalism wonderfully illus-
trates a broader construal of reason. It rejects some of the more problematic
elements of fideism29 while not committing him to unrealistic epistemic stan-
dards for belief-formation as found in hard rationalism.
Second, as we have seen, Newman employs a parity argument in his
account of faith and reason. Two ways of understanding this kind of argu-
ment have emerged in the scholarly literature. One claims that religious belief
is not immune to rational evaluation, and so, in this sense, it is exempt from
the charge of fideism. However, religious as well as non-religious beliefs
presuppose fundamental arational commitments—that is, commitments that
are not rationally grounded. Duncan Pritchard calls this position quasi-fideism
and draws attention to the connection between Newman and Wittgenstein
on this front.30 An alternative way of reading Newman’s parity argument, as
shown in this essay, is to contend that the epistemic standards for assess-
ing faith as rational are too high. When our concept of rationality includes
implicit as well as explicit reasoning, we find out that faith, as a presumptive
form of reasoning, is not unique, especially since this kind of reasoning is
found in commonly held beliefs. Insofar as Newman thinks that faith is a
mode of reasoning, it seems that his account of faith cannot be construed as
arational or groundless. Nevertheless, a fruitful (and related) inquiry would
involve clarifying the nature of commitment and its role in our discussion of
faith and reason in light of Newman’s broader account of rationality.31
29
For example, the notion that faith is exempt from external standards of ra-
tionality. For the distinction between irrational fideism (pitting faith against reason)
and rational fideism (faith can appropriately take over where reason leaves matters of
ultimate concern unresolved), see https://plato.stanford.edu/entries/fideism.
30
See Pritchard, “Wittgenstein on Faith and Reason,” 197–216; Duncan
Pritchard, “Faith and Reason,” Royal Institute of Philosophy Supplement 81 (2017): 101–18.
31
Thanks to Logan Gage, Alex Plato, and Chance Juliano for helpful comments
on an earlier version of this essay.
Complexly Based Beliefs and the
Generality Problem for Reliabilism1
Max Baker-Hytch
Wycliffe Hall, Oxford University
Introduction
This essay argues that certain cases involving what I shall term complexly based
belief, where a belief is formed via complex inference to the best explanation,
pose a serious difficulty for reliabilist theories of epistemic justification or
warrant. Many of our most important beliefs appear to be of this character.
The problem, in short, is that in such cases we cannot identify any belief-
forming process type that is such as to yield an intuitively correct verdict on
the epistemic status of the agent’s belief. If this is correct, then no proposed
solution to the generality problem can succeed.
The Generality Problem for Process

Reliabilism and Modal Reliabilism
In broad terms, the generality problem is the problem of finding a principle

or rule for determining which general type of belief-forming process should
be regarded as the relevant one for assessing the reliability of a given belief.
Process reliabilism can be understood in the following way, where “warrant”
denotes that property enough of which turns a true belief into an item of
knowledge:2
(PR) S’s belief p is warranted to the extent that the causal process that
produced that belief is reliable.3
1
I am grateful to C’Zar Bernstein, Ben Page, Logan Gage, and Alex Plato for
very helpful comments on earlier versions of this essay.
2
Here I am following Alvin Plantinga’s usage in Warrant: The Current Debate
(Oxford: Oxford University Press, 1993).
3
Alvin Goldman is the foremost advocate of an account along these lines. See
his “What Is Justified Belief ?,” in Justification and Knowledge, ed. George S. Pappas
© Max Baker-Hytch, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)
20 complexly BaSed BeliefS
The reason that the generality problem arises for (PR) is as follows. Any given
belief is the result of a particular causal sequence or process. For instance,
Jim’s belief that the dinner is ready is the end result of the following causal
sequence: first, Jim put the dinner in the oven and set the timer, which caused
him to have the belief that the dinner will be ready when the timer goes off; then later,
the timer went off, which in conjunction with Jim’s other beliefs caused Jim
to have the belief that the dinner is ready. This is a rough-and-ready description
of the causal process that led to Jim’s belief. We ask the question “Is the
process reliable?” If by “the process” we mean the sequence of events that
caused Jim’s belief, then it isn’t meaningful to ask whether “the process” is
reliable because reliability is a tendency, a matter of having a high success
rate across a suitably defined reference class. If “the process” denotes the
particular sequence of events that caused Jim’s belief, then we can’t sensibly
ask whether it is reliable because there isn’t any sense in which this particular
sequence of events produces a true belief in most cases. There is only one
case for us to consider.
Hence when we ask whether “the process” that produced Jim’s belief
was reliable, we must really be asking about the reliability of the general type
of process instantiated by the particular causal sequence that led to Jim’s
belief. What, then, is the general type that is instantiated here? Immediately
one realizes that there are many candidates: inference from perception and memory,
hearing the oven timer go off, perceiving informative signals from domestic appliances, and
so on. All of these general types are instantiated in the token causal process
that produced Jim’s belief. The generality problem, then, is the problem of
devising a principled way of selecting the uniquely relevant process type for
any given token belief.
The generality problem also arises for modal reliabilist accounts of
warrant.4 A modal reliabilist account of warrant appeals to the notion that
the agent wouldn’t have formed a false belief in certain counterfactual
(Dordrecht: D. Reidel, 1979). See also Kelly Becker, Epistemology Modalized (London:
Routledge, 2007); John Greco, Achieving Knowledge: A Virtue-Theoretic Account of Epis-
temic Normativity (Cambridge: Cambridge University Press, 2010). These latter two
accounts of warrant include a process reliability condition but more besides.
4
The generality problem for modal reliabilism receives less discussion than the
generality problem for process reliabilism, but the problem as it pertains to the sensi-
tivity condition has been discussed by Mark Alfano, “Sensitivity Theory and the Indi-
viduation of Belief-Formation Methods,” Erkenntnis 70 (2009). Timothy Williamson
acknowledges that “the generality problem arises for my safety account because, like
process reliabilism, it faces the question ‘How similar must counterfactual process-
es [bases] be to count towards reliability/safety?’” (“Reply to Alvin Goldman,” in
Williamson on Knowledge, ed. Patrick Greenough and Duncan Pritchard [Oxford: Ox-
ford University Press, 2009], 100.)
max Baker-hytch 21
circumstances. The sensitivity condition and the safety condition are both
modal reliability conditions. They differ in specifying different counterfactual
circumstances in which error needs to be avoided by the agent. The sensitiv-
ity condition requires that the agent would not have believed p in the nearest
possible world(s) in which p is false.5 Depending on the proposition, the
nearest world in which p is false may be very distant from the actual world.
For example, suppose that in the actual world, we are not brains in vats being
subjected to massive deception; reality is roughly as we suppose it to be. The
nearest world in which this proposition is false—that is, the nearest world in
which we are brains in vats being subjected to massive deception—is a very
distant world. Naturally, given that in such a world we are being subjected
to massive deception, we fail to be aware of the truth that we are being
so subjected. Hence my belief that I am not a brain in a vat fails to satisfy
the sensitivity condition. The safety condition for knowledge merely requires
that the agent avoids false belief about p in nearby possible worlds, regard-
less of the content of p.6 Hence given that the nearest world in which I am a
brain in a vat who doesn’t realize he is a brain in a vat is a very distant world,
my true belief that I am not a brain in a vat can satisfy the safety condition.7
Modal reliability conditions for warrant are often formulated in a way
that obscures the fact that they face the generality problem. The following
is typical:
(MR) S’s belief p is warranted only if S wouldn’t have falsely believed p in

the nearest world in which p is false (SenSitivity) / in nearby possible
worlds (Safety).
To see modal reliabilism in action, consider the following case. Suppose that
Lizzie forms the belief that it is raining outside on the basis of the testimony
5
Accounts of warrant that include a sensitivity condition have been defended
by Fred Dretske, “Conclusive Reasons,” Australasian Journal of Philosophy 49, no. 1
(1971); and Robert Nozick, Philosophical Explanations (Cambridge: Harvard University
Press, 1981), 172–79.
6
Accounts of warrant that include a safety condition have been defended by
Timothy Williamson, Knowledge and Its Limits (Oxford: Oxford University Press,
2000); and Duncan Pritchard, Epistemic Luck (Oxford: Oxford University Press, 2005);
“Anti-luck Virtue Epistemology,” Journal of Philosophy 109, no. 3 (2012).
7
That sensitivity accounts do not allow for knowledge that we are not brains
in vats whereas safety accounts do allow for such knowledge is often held by safety
theorists to be a decisive advantage. See Ernest Sosa, “How to Defeat Opposition to
Moore,” Philosophical Perspectives 33, no. 13 (1999); and Duncan Pritchard, “Sensitivity,
Safety, and Antiluck Epistemology,” in The Oxford Handbook of Skepticism, ed. John
Greco (Oxford: Oxford University Press, 2009).
of her trustworthy roommate Laura, who has just come in from the pouring
rain. Is Lizzie’s belief sensitive? Presumably it is, because in the nearest pos-
sible world in which it isn’t raining, her roommate Laura doesn’t tell her that
it is raining and Lizzie doesn’t form the belief that it is. Is Lizzie’s belief safe?
Again, presumably it is, because Laura couldn’t easily have been wrong about
the weather she recently witnessed, and hence there are no nearby worlds in
which Lizzie errs in regard to whether it is raining. Hence Lizzie’s belief is
safe. So far, so good. No mention of belief-forming process types.
However, there are various cases that require (MR) to be finessed in
certain ways in order to deliver intuitively correct verdicts. Suppose that
Lizzie very nearly consulted her phone’s weather app instead of asking her
roommate Laura, and suppose that, unbeknownst to Lizzie, this weather app
has a glitch that causes it to display the forecast for this time last week, when
it rained constantly. In nearby worlds in which it isn’t raining and in which
Lizzie consults the weather app instead of her roommate, Lizzie winds up
believing mistakenly that it is raining outside. What’s more, in the nearest
world in which it isn’t raining, Lizzie’s roommate stayed out longer, and so
Lizzie consults her weather app instead of her roommate. So in the nearest
world in which it is not raining, Lizzie mistakenly believes that it is raining;
and in some nearby possible worlds in which it is not raining, Lizzie mistak-
enly believes that it is raining. Hence Lizzie’s belief is neither sensitive nor
safe. But it is counterintuitive to deny Lizzie’s belief the status of knowl-
edge. After all, in the actual world, her belief is based on the testimony of a
trustworthy informant who told her the truth. Evidently, then, (MR) needs
to be refined with reference to the way or method by which an agent formed
her belief:8
(MR*) S’s belief p is warranted only if S wouldn’t have falsely believed p

in the nearest world in which p is false (SenSitivity) / in nearby pos-
sible worlds (Safety), given the way S actually formed her belief.
What this refinement says is that when we survey the nearby possible
worlds / the nearest not-p world, although we allow a range of conditions to
vary across worlds, including the weather, we are to hold fixed the method
Lizzie actually used in forming a belief about whether it is raining. But what
is the method Lizzie used? Precisely which features of the situation are we
to hold fixed? It is at this point that the generality problem presents itself
because there are different levels of generality at which we can characterize
the method of belief-formation that gets held fixed across possible worlds.
What we are holding fixed, of course, is the instantiation of a certain type of
8
As acknowledged by Pritchard, “Anti-luck Virtue Epistemology,” 257–58.
max Baker-hytch 23
causal belief-forming process. At the broad end of the spectrum, we could

characterize Lizzie’s method as trusting testimony, and at the narrow end, as
something much more specific, such as trusting Laura’s testimony about weather
conditions that Laura has witnessed in the very recent past. In between these two
extremes are indefinitely many other possible characterizations.
The crucial point is that which of these methods (i.e., process types) we
select as the relevant one can make a crucial difference to whether the token
belief satisfies modal reliability conditions. Suppose we select trusting testimony
as the relevant method that is to be kept fixed across possible worlds for the
purposes of evaluating modal reliability. Accordingly, nearby possible worlds
in which Lizzie receives testimony about the weather from other testifiers
besides Laura are also counted as relevant for the safety of Lizzie’s belief.
That significantly increases the scope for failure in regard to the safety condi-
tion. But if we select trusting Laura’s testimony about weather conditions that Laura
has witnessed in the very recent past as the method of belief-formation, then that
considerably limits the scope for failure in regard to the safety condition,
since it takes into account only those worlds in which Lizzie forms a belief
by way of one particular person’s testimony under quite specific conditions.
Proposals for Solving the Generality Problem
Earl Conee and Richard Feldman have suggested that a satisfactory solu-
tion to the generality problem must meet three desiderata.9 First, it must be
principled. In other words, it must involve the application of a general rule
for selecting the relevant process type in any given case; it cannot involve ad
hoc classifications made on a case-by-case basis. Second, it must yield intui-
tively correct judgments about the level of epistemic warrant enjoyed by any
given token belief. Third, it must “remain true to the spirit of the reliabilist
approach.”10 That is, it shouldn’t rely on nonreliabilist notions such as the
strength of an agent’s evidence, except insofar as such notions are ultimately
spelled out in terms of the success rate of the relevant process type at pro-
ducing true beliefs across a suitably defined range of circumstances.
Several proposals have been put forward as to how to solve the generality
problem. Each of these proposals aims in effect to satisfy the requirements
outlined by Conee and Feldman. One could roughly divide these proposals
9
Earl Conee and Richard Feldman, “The Generality Problem for Reliabilism,”
Philosophical Studies 89 (1998), 1–29; “Typing Problems,” Philosophy and Phenomenological
Research 65, no. 1 (2002): 98–105; Evidentialism: Essays in Epistemology (Oxford: Oxford
University Press, 2004), 135–59.
10
Conee and Feldman, “Generality Problem,” 4.
into three camps. Taking their cue from the cognitive sciences, some relia-
bilists propose that the solution to the generality problem is to be found in
the descriptions of belief-forming processes afforded by our best empirical
accounts of human cognition. William Alston and Ralph Baergen have devel-
oped this approach.11 The thought is that for any given token belief, empirical
accounts of cognition will enable us to fix upon the one relevant process
type for evaluating the belief ’s epistemic status. Another approach, hinted
at by Alvin Goldman but more fully developed by Kelly Becker,12 proposes
that the relevant process type is “the narrowest, content-neutral process that
is causally operative in belief production.”13 The idea is that for any given
token belief, one should select a type that is as narrow as possible within
the following two constraints: First, the description of the type must not
mention the content of the token belief. So for Tom’s belief that the sun
is shining, we should not select the type forming a belief about whether the sun is
shining. Second, the description of the relevant type must only make refer-
ence to factors that were actually causally relevant to the production of
the token belief. So for instance, the fact that Tom formed his belief on
a Tuesday is not causally relevant (let us suppose), and so that fact should
be omitted from the description of the relevant process type. Yet another
solution that has been put forward in the recent literature is a contextual-
ist one, according to which features of the conversational context deter-
mine the truth conditions for a statement of the form “S’s belief that p was
formed by a reliable process.” John Greco and Mark Heller have defended
this approach,14 according to which conversational context fixes both the
width of the relevant process type and the range of circumstances across
which the type success ratio is measured in order for a given belief to count
as being produced by a reliable process.
Each of these proposals purports to offer a general rule for identifying
the relevant process type for any given token belief. A number of critiques
of these proposals are already extant, and I won’t repeat them.15 In essence,
11
William P. Alston, “How to Think about Reliability,” Philosophical Topics 23,
no. 1 (1995); Ralph Baergen, Contemporary Epistemology (Fort Worth, TX: Harcourt
Brace, 1995).
12
Alvin Goldman, Epistemology and Cognition (Cambridge: Harvard University
Press, 1986), 50–51; Kelly Becker, “Epistemic Luck and the Generality Problem,”
Philosophical Studies 139 (2008).
13
Becker, “Epistemic Luck,” 363.
14
Mark Heller, “The Simple Solution to the Problem of Generality,” Noûs 29,
no. 4 (1995); Greco, Achieving Knowledge, 76–80.
15
See especially Conee and Feldman, “Generality Problem”; “Typing Prob-
lems”; but also Richard Swinburne, Epistemic Justification (Oxford: Clarendon Press,
2001), 13–20.
max Baker-hytch 25
the objection is that these various proposals fail to do enough to narrow

down the range of process types to just one single relevant type for each
given token belief. I am sympathetic to this objection. But what I shall argue
now is that there are certain sorts of cases wherein no process type that we
might single out as the relevant one will deliver an intuitively correct verdict
on the epistemic status of the belief.
Complexly Based Beliefs and Modal Reliabilism
By speaking of “complexly based belief,” what I mean to denote is a case

in which an agent’s belief is based on inference from a complex evidence
set. For the purposes of my argument, I don’t need a watertight definition
or analysis of complexly based belief because I am simply using the term to
gesture loosely at a cluster of cases that exhibit features that I shall argue are
problematic for reliabilism. Complexly based belief is to be contrasted with
simply based belief, such as forming a belief that it is 12 p.m. based on the
evidence of one’s wristwatch reading “12 p.m.”
Consider the following fictional case of complexly based belief. A histo-
rian, let’s call him Richard, is investigating the question of whether Alberich,
a king who at one time reigned over an ancient kingdom called Theusia, was
responsible for setting free a people group known as the Mythrians, who had
once been the slaves of the Theusians. Richard has assembled the following
relevant evidence:
E1: A careful Theusian chronicler writing approximately a century after

the time of Alberich reports that “around a century ago, the Myth-
rian slaves were freed and thereupon migrated into the region of
Dorn, where they made their lasting home.”
E2: Modern archaeological surveys of the area corresponding to the
ancient region of Dorn show a sudden appearance of coins and
pottery in the archaeological stratum corresponding to the decades
following Alberich’s reign. Many of the coins and pottery items
bear an emblem depicting a king, accompanied by the words “We
thank Alberich.”
E3: A fragment of a report from another Theusian chronicler, writing
shortly after the reign of Alberich, states that “Alberich will not be
remembered kindly by the people of Theusia, for he needlessly gave
away the kingdom’s most valuable commodities.”
E4: A pamphlet from around the time of Alberich rehearses what
appears to be a Theusian proslavery slogan: “Mythrian bodies are
our commodities.”
Richard reflects on these data and comes to the conclusion that Alberich did
indeed emancipate the Mythrian slaves. Specifically, he reasons in the follow-
ing way: To start with, the first chronicler’s report, E1, should be regarded
as probably accurate (in the absence of contrary evidence) in view of this
chronicler’s high level of accuracy on matters where historians have been
able to independently check his claims. The time period during which E1 says
the Mythrians were freed is chronologically consistent with Alberich having
freed them. What’s more, the report states that the Mythrians migrated into
the regions of Dorn and settled there after their emancipation. E2, the sud-
den appearance of pottery and coins in the region of Dorn in the archae-
ological stratum corresponding to the period shortly after Alberich’s reign,
is very probably due to that migration of Mythrians that E1 reports. The
fact that many of these coins and pottery items bear emblems expressing
gratitude to Alberich strongly indicates that the Mythrians were grateful to
Alberich for something of deep importance to them—something important
enough for them to have committed it to communal memory by imprint-
ing many of their household items. What might Alberich have done for the
Mythrians to make them so enduringly grateful to him? The fact that E1
tells us that the Mythrians were set free from slavery, combined with the fact
that only an act so radical as emancipation would be likely to leave a people
group so enduringly grateful to the king of the nation that formerly held
them captive, makes it pretty probable that Alberich’s emancipation of them
was the cause of their deep gratitude to him. So already, just given E1 and
E2, it is probable that Alberich emancipated the Mythrians. E3 and E4, taken
together, only serve to increase that probability. E3 states that Alberich gave
away the kingdom’s most valuable commodities. What commodities might
those be? Slaves are typically thought of by their slaveholders as commod-
ities. But this conjecture is confirmed by E4, which shows that proslavery
writers in Theusia around the time of Alberich specifically used the language
of “commodities” to refer to Mythrian slaves, making it quite probable that
the “commodities” that E3 blames Alberich for giving away are indeed Myth-
rian slaves. All in all, then, the cumulative weight of E1–E4 firmly supports
the hypothesis that Alberich freed the Mythrian slaves. Richard’s conclusion
appears strongly warranted. If we suppose that Richard’s conclusion is cor-
rect, then it is plausible that Richard knows that Alberich freed the slaves.
Now suppose that if Alberich had not freed the Mythrian slaves, his fail-
ure to do so would have been due to his being assassinated by a proslavery
activist before he had the chance to enact his lifelong goal of emancipating
the Mythrians. And suppose that if he had been so assassinated, his family
would have inscribed his tomb with the words “Alberich—king, husband,
father. His life’s purpose was to free the Mythrian slaves.” And suppose that
if Alberich had been murdered by a proslavery assassin, this would have so
max Baker-hytch 27
enraged the Mythrian slaves that they would have revolted, and many of
them would have successfully escaped Theusia and formed a new commu-
nity in the region of Dorn and would have erected stone monuments there
declaring “Freedom for Mythria—at last!” A later Theusian chronicler would
have reported that the Mythrian slaves gained freedom at around this time.
The nearest world to the actual world in which Alberich did not free the
Mythrian slaves is a world in which the foregoing things occur. Let’s call this
world w. Not only is w the nearest world to the actual world in which Alberich
didn’t free the Mythrians, but it is also extremely close to the actual world: the
assassination very nearly occurred. In the actual world, the assassin’s cross-
bow bolt missed Alberich’s head by mere millimeters. Thus in a wide range
of nearby worlds, Alberich was assassinated and events unfolded as they did
in w.
In w and other nearby worlds in which the assassination attempt
succeeded, the historian Richard is confronted with the following pieces of
evidence:
E5: An inscription on the tomb of Alberich, bearing the words

“Alberich—king, husband, father. His life’s purpose was to free the
Mythrians.”
E6: Several stone monuments in the region of Dorn, dating to shortly
after the death of Alberich, inscribed with the words “Freedom for
Mythria—at last!”
E7: A report from a careful Theusian chronicler writing about a century
after the reign of Alberich, stating that “around a century ago, after
they gained their freedom, the Mythrians migrated into the region
of Dorn.”
On the basis of surveying this evidence, Richard concludes that Alberich

freed the Mythrian slaves, though he holds his conclusion with slightly less
confidence than in the actual world. He reasons that E7 makes the time
during which the Mythrian slaves gained freedom chronologically consistent
with Alberich’s having freed them and that this is supported by the existence
of E6, the freedom monuments built shortly after Alberich in the region of
Dorn, the very region where E7 says some of the Mythrians settled after
gaining freedom. The fact that the Mythrians gained freedom around the time
of Alberich is well-explained by the hypothesis that Alberich freed them, and
this hypothesis is supported by the tomb inscription, E5. The evidence that
Alberich freed the slaves is moderately good in world w, but in fact in world
w, Alberich did not free them.
This scenario creates trouble for both sensitivity and safety accounts
of warrant. For sake of ease, let p be the proposition that Alberich freed the
Mythrian slaves. The historian Richard’s actual-world belief in p is counted

sensitive provided that Richard refrains from believing p in the nearest world
in which p is false and in which Richard uses the same belief-forming method
as he did in the actual world. The key question, then, is whether Richard’s
belief-forming method in world w is the same as his method in the actual
world. The answer to this question, of course, hangs on how we character-
ize the method that Richard uses. And what I want to suggest is that none
of the ways of characterizing Richard’s method that we might try can give
the intuitively correct result that Richard’s actual-world belief is an instance
of knowledge. The difficulty can be boiled down to the following dilemma.
Either: We characterize Richard’s method in a way that doesn’t make
reference to particular items of evidence—namely, E1–E4. The avoidance
of reference to particular items of evidence can be thought of as establish-
ing a minimum level of generality at which we may characterize Richard’s
method. The narrowest (i.e., least general) method we could select, on this
approach, would be something like inference to the best explanation from archaeolog-
ical evidence in the region of Dorn from close to the time of the events in conjunction with
Theusian manuscript evidence from close to the time of the events. Given that we avoid
reference to particular items of archaeological and manuscript evidence in
characterizing Richard’s method, we are forced to regard w as a world in
which Richard uses the same method as he uses in the actual world. After all,
Richard’s inference to p from items E5–E7 certainly involves making infer-
ences to the best explanation from both archaeological evidence from Dorn
and Theusian manuscript evidence from close to the time of Alberich. Hence
we are forced to regard w as the nearest not-p world in which Richard uses
the same method as he uses in the actual world. We are also forced to regard
w as a nearby world in which Richard uses the same method as he uses in the
actual world. Given that in w Richard’s belief p is false, his actual-world belief
p fails to meet either the sensitivity condition or the safety condition. Hence
when we characterize the method in terms that fail to mention particular
items of evidence, Richard’s actual-world belief p is implausibly deemed not
to be an instance of knowledge by the sensitivity and safety accounts.
Or: We characterize Richard’s method in a way that does make reference
to particular items of evidence. The inclusion of particular items of evidence
in the method can be seen as establishing a maximum level of generality at
which we may characterize Richard’s method. Given that we include one
of the particular items of evidence in our characterization of Richard’s
method, there is no good reason not to include all the particular items of
evidence that Richard actually used in drawing his conclusion. On this
approach, then, we classify Richard’s method as drawing inferences from evidence
E1–E4 or as something even narrower that makes reference to the particu-
lar manner in which Richard evaluated items E1–E4. Given this way of
max Baker-hytch 29
characterizing Richard’s method, w is not counted as a world in which Richard

uses the same method as in the actual world because in world w, Richard’s
particular evidence was E5–E7 rather than E1–E4. So for the purposes of
evaluating sensitivity, we are to consider what Richard ends up believing in
the nearest not-p world in which he possesses items of evidence E1–E4 and
draws inferences from these very pieces of evidence. What we are asking,
then, is whether Richard would still believe p in the nearest world in which
he has the very same (excellent) evidence for p as he actually has, and yet p is
false. Let’s call this world x. The answer to our question is surely that Richard
does still believe that p in such a world. Similarly, if we asked whether I would
still believe that I have hands in the nearest world in which I have the same
(excellent) visual evidence for that proposition and yet it is false that I have
hands, the answer will of course be that I do still believe that I have hands
in that world. So given that we classify Richard’s method in a way that keeps
fixed the particular (very strong) evidence he actually used, we are forced
to conclude that Richard’s actual-world belief p fails to meet the sensitiv-
ity condition and hence, contrary to intuition, that it is not an instance of
knowledge.
How about the safety condition? Well, safety only deems relevant those
nearby worlds in which the agent uses the same method as she actually used.
Plausibly, world x is not a nearby world. That is, things would have to have
gone pretty differently than they actually did in order for a scenario in
which Alberich didn’t free the slaves to leave behind such a highly mislead-
ing set of clues that point so strongly to the conclusion that he did free
them: the chronicler’s report about the period of Alberich’s reign that states
that the Mythrian slaves migrated into Dorn after being freed, in conjunction
with the pottery and coins appearing shortly after the reign of Alberich in
the regions of Dorn with an emblem of a king and the words “We thank
Alberich,” together with the other chronicler’s report that castigates Alberich
for giving away the kingdom’s most valuable commodities and the pamphlet
indicating that the term commodities was a way of talking about slaves. Indeed,
it is hard to imagine quite how such evidence could have arisen given that
Alberich didn’t free the slaves. That world x is distant from the actual world
is good news for the safety theorist: it can be ignored for the purposes of
evaluating the safety of Richard’s actual-world belief that p. But so too can
almost all of the nearby worlds for the purposes of evaluating safety. The
reason is that in almost all of the nearby worlds, the assassination attempt on
Alberich succeeded, which means that in all such worlds, Richard’s evidence
base comprises different particulars (viz., E5–E7) than in the actual world.
On an approach to method-classification that holds fixed particular items
of evidence, Richard’s method in the actual world is counted as a different
method than the one he uses in all of these nearby worlds in which the
assassination attempt succeeded. And so Richard’s actual-world belief satis-

fies the safety condition by virtue of the fact that only extremely close worlds
are taken to be relevant for determining the safety of his belief—namely,
those worlds in which the assassination attempt failed and thus his evidence
base contains the very same particulars as in the actual world. In other words,
the range of counterfactual circumstances across which we are testing the
modal reliability of Richard’s method is exceptionally narrow.
It is easy to see why the safety condition is trivialized given an approach
to method-classification that holds fixed the detailed particulars of the agent’s
evidence base. Suppose, for instance, that Susan believes that Shakespeare =
Francis Bacon, not because she has done any historical research into the matter
but because she accepted the say-so of her friend Carmen, who revels in
all manner of conspiracy theories. Suppose that in fact it is true that Shake-
speare = Francis Bacon. It is intuitive that Susan doesn’t know this to be so.
But given that we hold fixed a method that includes the detailed particulars
of Carmen’s evidence—namely, trusting Carmen’s testimony that Shakespeare =
Francis Bacon (or something even narrower), there is no nearby world—
indeed, no possible world at all—in which Susan arrives at a false belief
by using this method. Susan’s belief is counted as safe on this approach to
method-classification that keeps fixed the detailed particulars of the agent’s
evidence base.
Cases of complexly based belief like the one involving Richard’s histori-
cal inferences pose a serious difficulty for modal reliabilism, given that modal
reliabilism crucially relies on the notion of a method of belief-formation.
What the difficulty boils down to is that in a case of complexly based belief,
the intricacies of the agent’s evidence base need to be taken into account
in the characterization of her belief-forming method; otherwise we end up
classifying together cases that are really quite different from one another as
regards the epistemic status of the agent’s belief. But once we take the step
of including the intricacies of the agent’s evidence base in the characteriza-
tion of her method, we run into the difficulty that holding fixed such a tightly
specified method trivializes the safety condition by drastically narrowing the
range of possible worlds that are deemed relevant for evaluating the safety
of the agent’s belief. As regards sensitivity, holding fixed a highly detailed
method of belief-formation will invariably yield the result that the agent’s
belief is not sensitive because, however good the agent’s evidence, she would
of course still believe that p in the nearest world in which p is false, and yet
she possesses just the same evidence as she does in the actual world.
max Baker-hytch 31
Complexly Based Beliefs and Process Reliabilism
Let us now turn to consider the situation as regards process reliabilism. Recall
that whereas modal reliability is a matter of whether the agent would have
avoided error under certain counterfactual circumstances, process reliability
is a matter of the success-to-error ratio of the relevant belief-forming pro-
cess type that is instantiated by the agent’s belief or, in other words, a matter
of how often the relevant process type yields true beliefs rather than false
beliefs across a suitably defined range of actual and possible circumstances.
Of course, there is the far-from-trivial issue of how wide a range of circum-
stances we are to take into consideration when evaluating the truth-ratio of
a given process type. For example, if the process type under consideration
is eyesight of medium-sized objects, it will of course make a big difference to our
estimation of the truth-ratio of this process type whether we measure it
across a range of circumstances that includes situations where the lighting
is poor and the objects are very distant from the eyes or whether instead we
measure its truth-ratio across a narrower range of circumstances in which
the lighting is always good and the objects are always close to the eyes. As
has been noted elsewhere,16 this problem of fixing the relevant range of
circumstances across which the process type’s truth-ratio is measured can be
seen as simply another facet of the generality problem. After all, the range of
conditions across which a given type’s reliability is measured can simply be
built into the description of the type itself. For example, rather than simply
talking about the reliability of the process type eyesight, we can talk about the
reliability of the process type eyesight under such-and-such conditions. The issue
of fixing the appropriate breadth of possible circumstances can simply be
subsumed, then, within the larger issue of fixing the appropriate level of
generality for evaluating the reliability of a belief-forming process.
Feldman has characterized the generality problem in terms of finding a
way of identifying the relevant process type such that the type that is selected
is neither too narrow that it is unrepeatable and hence it becomes meaning-
less to speak of its reliability (the “single case” problem) nor too broad that
various token beliefs that vary in their degree of epistemic warrant are all
classified under the same relevant process type (the “no distinction” prob-
lem).17 The approach that is implicit in much of the literature that develops
reliabilist accounts of warrant is one in which the selection of the relevant
process type is guided by intuition on a case-by-case basis. For cases of
simply based belief, this approach often works, at least, in the sense that
16
See Klemens Kappel, “A Diagnosis and Resolution to the Generality Prob-
lem,” Philosophical Studies 127 (2006).
17
Richard Feldman, “Reliability and Justification,” The Monist 68 (1985).
it is typically quite easy in a given case of simply based belief to identify a

process type that is not too narrow as to be unrepeatable and not too broad
as to run into the “no distinction” problem and yields an intuitively correct
verdict on the level of warrant enjoyed by the agent’s belief. Suppose, for
instance, that Lawrence arrives at the belief that there are fairies because he
really likes the idea and is apt to believe that his fantasies are reality. For this
sort of case, it is quite easy to fix upon a process type—something like wishful
thinking—which is not so narrow as to be unrepeatable and yet not so broad
as to subsume cases that vary widely in their level of epistemic warrant. The
process type wishful thinking presumably has a very low truth-ratio, which fits
our intuitive judgment that Lawrence’s belief has a very low degree of epis-
temic warrant. Conee and Feldman have argued that this sort of case-by-case
approach to selecting the relevant process type is unacceptable by virtue of
being unprincipled and ad hoc.18 However, what I want to suggest is that
even if we set aside that rather serious objection, process reliabilism runs into
another difficulty when it comes to cases of complexly based belief such as
the one I introduced earlier in the essay—namely, that no level of generality
that we might select is able to yield intuitively correct verdicts on such cases.
Or to put it another way, when it comes to these sorts of cases, there is no
space to pass between the twin perils of the “single case” problem and the
“no distinction problem.”
So let us return to the case of Richard the historian evaluating archae-
ological and manuscript evidence and arriving at the conclusion that King
Alberich liberated the Mythrian slaves. Again, we can draw the following
dividing line down the middle of the various process types instantiated by
Richard’s belief: either we select a type that makes no mention of the particu-
lar items of evidence on which Richard based his belief, or we select a
type that does mention the particulars. Let’s begin with the first approach.
Again, the narrowest we can go without mentioning evidential particulars is
something like the type inference to the best explanation from archaeological evidence
in the region of Dorn from close to the time of the events in conjunction with Theusian
manuscript evidence from close to the time of the events. The problem is that this
type is too broad: it is instantiated by token beliefs that vary in their level of
epistemic warrant. This process is instantiated by Richard’s actual belief that
p, which is based on excellent evidence (namely, E1–E4). But it is also instan-
tiated by Richard’s belief that p in the many nearby worlds in which Alberich
was assassinated before he could free the slaves and in which the slaves
rebelled and then migrated into the region of Dorn. In these worlds, Richard
arrives at the belief that p on the basis of moderately good evidence (namely,
E5–E7). Intuitively, there is a difference between the level of warrant enjoyed
18
Conee and Feldman, “Generality Problem.”
max Baker-hytch 33
by Richard’s actual-world belief, based as it is on excellent evidence, and the

level of warrant enjoyed by his belief in a range of nearby worlds, which is
based on moderately good evidence. And yet we shall be forced to say that
these beliefs all have the very same degree of warrant as one another if our
selection of the relevant process type involves no mention of the particular
items of evidence that were used.
On the other hand, if we opt to select a process type that makes mention
of evidential particulars, the broadest type we can select is something like
drawing inferences from E1–E4. Naturally we avoid the problem we encountered
with the first approach, given that Richard’s actual-world belief instantiates
this process type, but his belief in nearby worlds that is based on E5–E7
does not instantiate this type. The problem we now encounter is that the
process type we have selected is too narrow; it is doubtful that we can mean-
ingfully speak of its reliability. We run into the single case problem. As Feld-
man articulates it, the single case problem is the problem that “if relevant
types are characterized very narrowly then the relevant type for some or
all process tokens will have only one instance (namely, that token itself).”19
But Kelly Becker rightly points out that this way of putting things leaves
it unclear whether this is supposed to be “a problem for any process actu-
ally used only once, no matter how individuated, or . . . a problem just for
those processes so narrowly individuated that they could be used only once.”20
Becker is inclined to think that the latter is the real source of trouble. Now,
in the way I handled things on the former approach (the approach on which
we select a process type that makes no mention of evidential particulars), I
already assumed that a process type could be considered repeatable even if
it is used only once in the actual world, provided that it could in principle
be used in other possible circumstances than the precise ones in which it is
actually used.21 The process type drawing inferences from E1–E4 is only used
once in the actual world, we may suppose. But it can be considered repeatable
provided there are possible worlds in which this process type is employed
under circumstances that differ somewhat from those in which it is actu-
ally used. The difficulty, however, is that because Richard’s having the very
particular evidence base he actually has (viz., E1–E4) is so exquisitely sensi-
tive to the way that events unfolded around the reign of Alberich, we simply
can’t vary the circumstances very much at all without the result that Richard
winds up with somewhat different evidence. In other words, the range of
possible circumstances in which Richard counts as using the type drawing
inferences from E1–E4 is extremely narrow, too narrow for it to be meaningful
19
Feldman, “Reliability and Justification,” 160–61.
20
Becker, “Epistemic Luck,” 361.
21
Becker (ibid.) argues for this understanding of repeatability.
to speak of the type’s being either “reliable” or “unreliable” across such a

narrow range of possible circumstances.
When we talk about a belief-forming process being “reliable,” we mean
that the process type in question achieves a high success rate when deployed
across a certain range of circumstances. The notion of reliability is trivialized
if the range of circumstances across which we measure the success rate of a
process type is extremely small. What counts as too small is surely vague, but
I think we have an intuitive sense that it ceases to be meaningful to speak of
“reliability” if the range of circumstances across which we measure the type’s
success rate is extremely narrow. This point applies not just to the reliability
of belief-forming processes but also to reliability discourse more generally.
Would we wish to come to a judgment about whether a certain car is reli-
able at starting if we are only allowed to take into account its success rate at
starting when it is inside a garage where the temperature is between 18 and
22 degrees Celsius and when the car has just been given a full checkup by a
mechanic? Probably not.
More generally, the situation will be like this for many cases of complexly
based belief. The reason for this, in short, is that these sorts of cases involve
an agent basing her belief on a detailed evidence base, which is such that the
details really matter when it comes to evaluating the belief ’s degree of warrant.
What’s more, in these sorts of cases, very slight changes in the circumstances
that yielded the agent’s very particular evidence base will lead to the agent’s
having a somewhat different set of detailed evidence. This creates the follow-
ing dilemma for the process reliabilist. Either the process reliabilist char-
acterizes the agent’s belief-forming process in a way that glosses over the
particular details of the agent’s evidence base, which leads to the result that
cases that intuitively differ as regards their degree of warrant end up being
classified under the same relevant process type, yielding the counterintuitive
judgment that these various beliefs possess the same degree of warrant. Or
the process reliabilist characterizes the agent’s belief-forming process in a
way that takes account of the very particular details of the agent’s evidence
base but then runs into the problem that the range of circumstances in which
such a detailed process type can be employed is exceptionally narrow; too
narrow for it to be meaningful to speak of the type’s reliability. In short, when
it comes to these sorts of cases, there is no level of generality to be found
that avoids both the no distinction problem and the single case problem.
Conclusion
I have argued that certain cases involving what I have termed “complexly
based belief ” pose a serious difficulty for both process reliabilism and modal
max Baker-hytch 35
reliabilism. What characterizes these cases is that they involve complex

inferences to the best explanation. The vast majority of our scientific, phil-
osophical, and historical beliefs are of this sort. The problem is that, with
such beliefs as these, we cannot identify any process type that is at the right
level of generality so as to yield an intuitively correct verdict on the epis-
temic status of the belief. As we saw earlier, various proposals have been put
forward in the literature for solving the generality problem. Objections to
these solutions typically allege that these solutions don’t really narrow things
down to just one process type for any given token belief. But if the thesis of
this essay is correct, then the situation for reliabilism is considerably worse
than that because any solution to the generality problem will be premised
upon the assumption that for any given token belief, there is at least one pro-
cess type that, if selected as the relevant type, would yield an intuitively cor-
rect verdict on the epistemic status of the token belief. If that is not in fact
the case, then the generality problem is insoluble. The reliabilist may then
retreat to the claim that reliabilism gives an adequate account of epistemic
warrant only for cases of simply based belief—cases of forming a belief
on the basis of the checking of a clock, or a glance out of the window, and
other similar examples that are the staple of reliabilist epistemologies—but
concede that, as Nicholas Wolterstorff puts it, “the very idea of identifying
methods of belief-formation and determining their reliability has little appli-
cation when it comes to the deep questions of philosophy and religion, and
of many other areas of human life. The idea lacks purchase.”22
22
Nicholas Wolterstorff, “The Significance of Inexplicable Disagreement,” in
Religious Faith and Intellectual Virtue, ed. Laura Frances Callahan and Timothy O’Con-
nor (Oxford: Oxford University Press, 2014), 327.
Preemptionism and Epistemic Authority
Donald J. Bungum
University of Mary
1. Introduction
Your friend is a nutritionist. She tells you that a serving of kale contains
900 percent of your recommended dietary allowance of vitamin K. This
sounds like a lot of vitamin K to you, but you already knew that kale is one
of those superfoods, and you trust your friend. On your friend’s authority,
then, you come to believe that a serving of kale contains 900 percent of
your RDA of vitamin K.
Let K stand for the proposition that kale contains 900 percent of your
RDA of vitamin K, and suppose that you have come to believe K in an
epistemically exemplary way. There is now a bit of a puzzle concerning
the basis for your belief that K. On the one hand, it seems plausible to say
that the basis for your belief is both your friend’s testimony and also all the
other evidence that you possess that is relevant to K, including your prior
knowledge that kale is healthy. This view is plausible because, after all, you
have come to believe K in an epistemically exemplary way, and it seems less
than exemplary for you to fail to base your belief in K on evidence that you
possess that is relevant to K.
On the other hand, it also seems plausible to say that the basis for
your believing K is just your friend’s testimony alone. This latter view
is plausible because we said that you came to believe K on your friend’s
authority, and it seems to display a lack of trust in your friend’s authority
to base your belief in K on anything besides your friend’s testimony.
For example, if I tell you that p, but you come to believe p because my
testimony is the nineteenth instance of testimony that p that you have
received this week, then you do not seem to believe me that p is true, but
rather you believe p on the basis of your own evidence and your own
assessment of it.
Preemptionism is a view concerning the proper basis for beliefs held on epis-
temic authority. According to preemptionists, when a subject believes what
another person says on that person’s authority, the authoritative testimony
© Donald J. Bungum, Quaestiones Disputatae, Vol. 8, No. 2 (Spring 2018)

donald J. Bungum 37
“replaces” a subject’s own consideration of the evidence.1 Arnon Keren, for

example, says that believing an epistemic authority “involves seeing his saying
or judging that p as a reason for not basing one’s opinion on p on some other
evidence available to one.”2 Linda Zagzebski, another preemptionist, says
that “I am believing on authority only if I take the authority’s belief or testi-
mony that p as replacing my other reasons for and against p.”3 The preemp-
tionist thinks that the recipient of authoritative testimony should “set aside”
her other reasons for belief,4 letting the view of the epistemic authority take
the place of her own.5
1
This is not to say that authoritative testimony must always be believed on
the testifier’s authority. For example, when my testimony that p is the nineteenth
instance of testimony that p that you have received this week, supposing that it is
reasonable for you to view me as an authority, it seems open to you either to believe
that p on my authority or else to use my testimony as inductive evidence that p (along
with the other instances of testimony that p). Preemptionism is a thesis specifically
about cases of belief on another person’s authority. For discussion of the distinc-
tion between believing on authority and using testimony as inductive evidence, see
Paul Faulkner, “Agency and Disagreement,” in Social Epistemology and Epistemic Agency:
Decentralizing the Epistemic Agent, ed. P. Reider (London: Rowman & Littlefield, 2016),
75–90.
2
Arnon Keren, “Trust and Belief: A Preemptive Reasons Account,” Synthese
191, no. 12 (2014): 2602.
3
Linda Zagzebski, “Replies to Christoph Jäger and Elizabeth Fricker,” Episteme
13, no. 2 (2016): 187.
4
Preemptionists justify this strong view of epistemic authority by claiming
that authoritative testimony provides subjects with a special kind of reason for
belief—namely, a preemptive reason for belief. According to the preemptionist, a pre-
emptive reason for believing p is two things: First, like ordinary reasons for belief, a
preemptive reason for belief is a consideration that favors believing p. For example,
according to the preemptionist, receiving authoritative testimony that p favors believ-
ing that p just like seeing smoke favors believing that there is fire. According to the
preemptionist, however, a preemptive reason for belief is also a reason not to believe
p for any other reason. This is what makes preemptive reasons for belief different
from ordinary reasons for belief. Thus according to the preemptionist, seeing smoke
differs from receiving authoritative testimony that p because seeing smoke does not
rule out checking for fire in other ways (e.g., asking, smelling, etc.), but receiving
authoritative testimony that p does rule out believing p for reasons besides the testi-
mony itself. So if your nutritionist friend tells you that kale contains a lot of vita-
min K, the preemptionist will say not only that you should believe that kale contains
a lot of vitamin K but also that your friend’s testimony should become the sole reason
for which you believe kale contains a lot of vitamin K.
5
For the most important recent defense of preemptionism in epistemology,
see Linda Trinkaus Zagzebski, Epistemic Authority (Oxford: Oxford University Press,
2012). Zagzebski’s preemptionism is inspired by Joseph Raz’s work on preemptionism
38 preemptioniSm and epiStemic authority
Is preemptionism plausible? That is, when you receive authoritative

testimony, should you really set aside all your other reasons for belief ? In
this essay, I will be arguing that you should, but there are obviously some
reasons to think that you should not. For example, suppose that, in the earlier
case, you also know that vitamin K comes from chlorophyll and that kale’s
deep-green color comes from chlorophyll. If you know these things, then
you have some decent independent evidence that kale contains a lot of
vitamin K. According to the preemptionist, however, once your friend tells
you that K, you should set aside all your own reasons for believing K. But
why should you set aside this knowledge about kale and chlorophyll? Isn’t
this knowledge a perfectly good piece of evidence for K? Is it even psycho-
logically possible to set aside such evidence?
Christoph Jäger has recently used such questions to motivate what
he calls the “unhinging objection” to preemptionism about epistemic
authority.6 His objection is that preemptionism requires the subject to
replace her own reasons with the authoritative testimony even when her
own reasons are epistemically adequate. Jäger argues that it is unreasonable
for a subject to “unhinge” her belief from epistemically adequate reasons,
since, as a rule, it is epistemically better to base one’s belief on more rather
than fewer good reasons. Jäger concludes that preemptionism compels a
subject to undergo “epistemic regression” when receiving authoritative
testimony. In light of cases like the one in which you know about kale
and chlorophyll, Jäger’s worry seems especially apt, since preemptionists
typically say nothing about whether the replacing power of authoritative
testimony depends on whether a subject’s own reasons are adequate or
inadequate.
In this essay, I defend preemptionism in the face of the unhinging
objection. In section 2, I set out the unhinging objection in more detail.
in ethics, especially Joseph Raz, The Morality of Freedom (Oxford: Clarendon Press,
1986), 42; Joseph Raz, “Authority and Justification,” in Authority, ed. Joseph Raz
(Oxford: Basil Blackwell, 1990), 10. For other recent defenses of preemptionism in
epistemology, see Arnon Keren, “Zagzebski on Authority and Pre-emption in the
Domain of Belief,” European Journal for Philosophy of Religion 6, no. 4 (2014): 61–76.
For criticisms of preemptionism in epistemology, see Trent Dougherty, “Zagzebski,
Authority, and Faith,” European Journal for Philosophy of Religion 4 (Winter 2014): 47–59;
Sarah Wright, “Epistemic Authority, Epistemic Preemption, and the Intellectual Vir-
tues,” Episteme 13, no. 4 (December 15, 2016): 555–70; Katherine Dormandy, “Epis-
temic Authority: Preemption or Proper Basing?,” Erkenntnis (July 19, 2017), Online
First; Michel Croce, “Expert-Oriented Abilities vs. Novice-Oriented Abilities: An
Alternative Account of Epistemic Authority,” Episteme (May 23, 2017), Online First.
6
Christoph Jäger, “Epistemic Authority, Preemptive Reasons, and Under-
standing,” Episteme 13, no. 2 (2016): 167–85.
donald J. Bungum 39
In section 3, I describe a general strategy that the preemptionist can use

to answer the unhinging objection. In section 4, I examine differences
between authoritative and lay belief, and in sections 5 and 6, I examine
two ways in which the general strategy can be filled out in order to defend
preemptionism. I argue that authoritative testimony enables recipients
to carry out “precision-dependent inferences,” and I show how this fact
renders recipients’ other reasons unsuitable bases for belief. I conclude that
authoritative testimony allows a recipient to live an elevated epistemic life,
one that is properly based on authoritative testimony but not on a subject’s
own reasons.
2. A Difficulty for Preemptionism: The Unhinging Objection
It will be useful to begin by setting out Jäger’s unhinging objection in a little

more detail. Recall that, according to preemptionism, a subject who receives
authoritative testimony that p should base his belief that p on the authorita-
tive testimony alone and not on any of his own reasons for believing p. As
Jäger points out, however, this view has the consequence that the recipient of
authoritative testimony must set aside her own reasons even when those rea-
sons are epistemically strong. But to set aside epistemically strong reasons seems
unreasonable. Jäger motivates his worry with the following case:
cardinal: Francis is not very good at identifying birds, but

male cardinals are distinctive, and Francis can identify them
fairly reliably. While out for a walk with his friend Sophia, Fran-
cis spies a red bird in a tree. It seems to be a cardinal to him,
but since he is not entirely confident, he asks Sophia, who, as
it happens, is an ornithologist (and Francis knows this). Sophia
confirms that it is a cardinal.7
7
Adapted from Jäger, “Epistemic Authority,” 174. Trent Dougherty offers a
similar case: “I see grocery bags on the floor and conclude that Jim went to the mar-
ket. Jill tells me he did. I rightly believe more firmly when Jill’s testimony is added to
my experience.” See Dougherty, “Zagzebski, Authority, and Faith,” 52. Dougherty
takes this case to show that reasons of different kinds (e.g., perceptual reasons, tes-
timonial reasons) can yet provide epistemic support simultaneously and aggregately,
and he thinks that preemptionists have completely failed to show how we can “look
a reason ‘square in the face’ and remain (non-degenerately) unmoved” by that reason
(ibid.). In the following text, I explain how we can remain nondegenerately unmoved
by a reason when that reason fails to support the precision-dependent inferences that
we can reasonably make.
Prior to asking Sophia, Francis is somewhat confident that the bird in the tree
is a cardinal. For the sake of argument, let us grant that Francis’s confidence
is rational, and let us suppose that Francis’s confidence is based on his having
a certain visual experience, say, seeming to see a crested red bird of a certain
size.8 Nevertheless, preemptionism says that authoritative testimony that p
replaces a subject’s other reasons for believing p, so according to preemp-
tionism, Sophia’s testimony that the bird is a cardinal replaces Francis’s visual
experience as a reason to believe that the bird is a cardinal. Jäger takes this
latter consequence to be implausible, since, by hypothesis, Francis is a fairly
reliable judge of cardinals, so his visual experience of a cardinal provides
fairly strong epistemic support that the bird before him really is a cardinal.
Jäger puts the point well when he asks, “Wouldn’t the agent be epistemically
better off if he based his belief both on the fact that the authority shares
his belief and on his own reasons (or set of reasons)? Basing a belief on two
good reasons is better than basing it on one.”9
3. Defending Preemptionism: A General Strategy
The preemptionist’s distinctive claim is that the recipient of authoritative tes-

timony that p should not base his belief that p on his own reasons for p. Jäger
has provided a reason why this distinctive claim might seem problematic.
Nevertheless, the following is a general strategy by which the preemptionist
might defend his distinctive claim. He might argue as follows:
Generalized Preemptionist Strategy

(Causal Claim) By virtue of receiving authoritative testimony that p, if a
recipient of authoritative testimony has any of his own reasons for
believing p, those reasons come to possess feature X.
(Basing Claim) Any reason for believing p possessing feature X is not a
suitable basis for believing p.
8
For the purposes of this essay, the precise details about justification and justi-
fiers are irrelevant, since the preemptionist claims only that authoritative testimony
replaces one’s other reasons for belief, however reasons and justification are under-
stood. I will use the terms “evidence for p” and “reasons for belief in p” inter-
changeably to mean “considerations favoring belief in p.” On this account, Francis’s
perceptual experience and Sophia’s testimony in cardinal are both reasons for Fran-
cis to believe that there is a cardinal in the tree.
9
Jäger, “Epistemic Authority,” 176.
donald J. Bungum 41
The generalized strategy that I am proposing for the preemptionist is com-

posed of two claims: a causal claim and a basing claim. The causal claim
says that a subject’s receiving authoritative testimony causes his other rea-
sons to come to have some property, and the basing claim says that reasons
with that property are not suitable bases for belief. The main idea behind
the general strategy is that authoritative testimony that p is not “neutral”
with respect to the epistemic adequacy of a subject’s other reasons that
p. The general strategy insists that, unlike an observation of an individual
white swan, which leaves one’s other observations of other white swans
intact and (epistemically) unaffected, authoritative testimony somehow
changes the epistemic significance of one’s other reasons, making them
unsuitable bases for belief.
It is clear that the preemptionist’s causal and basing claims are satisfied in
at least some situations. Consider the following case:
Silo: Josephine is a local, and Mara is a visitor. Josephine and

Mara are driving down a country highway together. Along the
way, Mara points and says to Josephine, “Look, that’s a grain
silo.” Josephine replies, “Well, remember we’re in North Dakota.
That’s either one of the sixty grain silos in the region, or it’s one
of the forty ballistic missile silos in the region.”
Let p be the proposition “That’s a grain silo.” In Silo, Mara first believes p
on the basis of her own perceptual experience of the object, which appears
to her as a grain silo. Mara then acquires a new reason to believe p—namely,
Josephine’s testimony about the number of grain and missile silos in the
surrounding region. But while Josephine’s testimony gives Mara some reason
to believe p, Josephine’s testimony also brings it about that Mara’s perceptual
experience comes to possess a certain feature—namely, a recognized inability
to discriminate between nearby situations in which p and ~p.10 Thus the pre-
emptionist’s general causal claim is satisfied in Silo. Moreover, it is widely
held that if a subject recognizes a reason as being unable to discriminate
between the truth and falsity of some proposition r, then it is not proper
10
There is, of course, the possibility that Mara will fail to recognize the bearing
of Josephine’s testimony on Mara’s perceptual evidence. But even on this possibility,
so long as Mara is a properly functioning epistemic agent, we should say that Mara
should recognize the bearing of Josephine’s testimony on Mara’s perceptual evidence.
And it is plausible that, at least in some sense, a subject should not base a belief that
p on a reason that she should recognize as nondiscriminating between p and ~p, even
if she does not actually recognize this. So this possibility does not undermine the
central point of the argument.
for the subject to use that reason as the basis for his belief that r.11 Conse-
quently, the preemptionist’s basing claim is also satisfied in Silo.
Nevertheless, we need to make one refinement to the preemptionist’s
causal claim. In Silo, Josephine’s testimony does cause Mara’s perceptual
experience to come to possess some new, epistemologically relevant feature.12
It is therefore true to say that Mara’s perceptual experience comes to possess
this feature by virtue of her receiving Josephine’s testimony. But it is false
to say that Mara’s testimony comes to possess this feature by virtue of the
authoritativeness of Josephine’s testimony. For it is not because Mara respects
Josephine as an authority or receives Josephine’s testimony as authoritative
that her perceptual experience comes to be recognized as nondiscriminat-
ing. Rather, it is because of the content of Josephine’s testimony that Mara’s
perceptual experience comes to have this feature: having accepted Josephine’s
testimony as true, we expect Mara to be able to recognize and infer by her own
lights that there are a wide class of cases in which perceptual experiences
such as hers in Silo do not pick out genuine grain silos. If the preemptionist,
however, wants to tie the replacing power of authoritative testimony to its
nature as authoritative, then we will need to make the following qualification
to the causal claim previously described:
(Causal Claim*) By virtue of the authoritativeness of received authoritative

testimony that p, if a recipient of that testimony has any of his own
reasons for believing p, those reasons come to possess feature X.
The case of Silo does not satisfy the revised causal claim, since it is not
by virtue of the authoritativeness of Josephine’s testimony that Mara’s
11
For example, consider a standard case of undermining defeat: you see a line
of widgets that appears to be red, and then you learn that the widgets are illuminated
by red lights. Your perceptual experience becomes recognized as unable to discrim-
inate between the widgets being red or their not being red, and this is what renders
your perceptual experience and unsuitable basis for belief.
12
I do not mean to say that Josephine’s testimony causes Mara’s perceptual ex-
perience to come to possess a new intrinsic, qualitative feature. I mean only to say that
Josephine’s testimony causes Mara’s perceptual experience to come to possess a new
relational feature. An example can help illustrate: Suppose that you have a memory
of having eaten alligator at the county fair when you were young. You then learn
that the “alligator” was in fact chicken. This new knowledge does not change the
intrinsic, qualitative features of the perceptual experience (e.g., taste, color, feel, etc.).
The new knowledge does, however, change whether the perceptual experience can
reasonably be used to support other propositions (e.g., “I’ve eaten alligator before”).
In the text, Josephine’s testimony changes Mara’s perceptual experience only in this
latter, extrinsic sense of change.
donald J. Bungum 43
perceptual evidence is recognized as nondiscriminating but rather by virtue

of the content of that testimony.
In light of the modified general strategy, we now have three questions to
put to the preemptionist about epistemic authority. First, is there any feature
that a subject’s own reasons for p come to possess whenever that subject
receives authoritative testimony that p? Second, does this feature make the
subject’s own reasons for p an unsuitable basis for believing p? Third, does
this feature come to the subject’s reasons by virtue of the very authorita-
tiveness of the testimony received? In what follows, I argue for affirmative
answers to all three questions.
4. Differences between Authoritative and Lay Belief
In order for the preemptionist successfully to deploy the general strategy,

the preemptionist must show that there is some feature that (1) attaches to a
subject’s own reasons for p, (2) renders those reasons an unsuitable basis for
believing that p, and (3) is caused by the authoritativeness of the testimony
that p. In order to examine whether there is any such feature, it will be useful
to identify some of the key epistemological differences between authoritative
and lay belief.
Consider, for example, the relevant differences between Sophia and
Francis in cardinal. Zagzebski rightly points out, “The ornithologist would
not be fooled by a cardinal look-alike.”13 But why, exactly, would Sophia, as
an ornithologist, not be fooled by a cardinal look-alike?14
I submit that Sophia is less likely to be fooled by a cardinal look-alike
because her epistemic position is superior to Francis’s in three ways: First,
she has a much better understanding of which factors are rightly taken to be
indications that a bird is a cardinal. For example, she will be aware of whether
the crucial factors are head shape, beak shape, beak color, eye ring, eye line,
eyebrow, crest, wing angle, foot color, leg color, tail length, tail shape, tail
markings, flight pattern, song pattern, perching location, perching posture,
regional distribution, migration pattern, song, mating call, or something else.
A novice like Francis will not know which factors distinguish genuine cardi-
nals from cardinal look-alikes.
13
Zagzebski, “Replies to Christoph Jäger,” 190.
14
As a matter of fact, there are several cardinal look-alikes in North America.
Cardinal look-alikes include desert cardinals (pyrrhuloxia), scarlet tanagers, vermil-
lion flycatchers, cedar waxwings, vermillion cardinals, and hybrids of northern car-
dinals and desert cardinals.
Second, she will have a much better understanding of how strongly these
factors are correlated with a bird’s being a cardinal. For example, she will
know how the cardinal’s distinctive color varies with age, sex, geographic
region, and so on.
Third, she will be more practiced at recognizing when general relations
of evidential support are instantiated in concrete particular situations. For
example, Sophia will know that, in general, eastern cardinals are a duller red
than western varieties. But she will also know how to recognize the particular
bright red of a western cardinal in the field and to use that color to distin-
guish it from, say, the desert cardinal (a cardinal look-alike with an overlap-
ping territory).
These aspects of Sophia’s superiority are common to most cases of epis-
temic authority. That is, in general, epistemic authorities are (a) better able
to identify, enumerate, and distinguish the factors on which truths in the
domain depend; (b) more sensitive to the kind of dependencies that exist
between these factors and those truths; and (c) better able to recognize when
these factors are instantiated in concrete situations and in what ways.
5. Calibration
Recall that, according to the general strategy, the preemptionist seeks to

show that authoritative testimony causes a recipient’s own reasons to come
to have a certain feature, one that renders them an unsuitable basis for belief.
If the earlier characterization of the differences between authoritative and
lay belief is roughly correct, however, then it suggests a certain feature that
might be useful to the preemptionist. This feature is a recognized lack of
what we might call “calibration.”
5.1. Calibration Itself
In order to get a better grip on what calibration is and why it might be use-
ful to the preemptionist, consider the following example. Suppose that you
have a meterstick that you use to measure things around you. Then suppose
that, one day, a platinum Standard Meter arrives in your town from Paris
while on a world tour. The platinum Standard Meter is a means by which
you might come to know whether your own meterstick is accurate. In such a
case, you find yourself in what we might call a “calibration situation”—that
is, a situation in which you have the opportunity to verify the accuracy of
your measure of the extended world. Insofar as you are in such a calibration
situation, it makes sense for you to suspend judgment about whether your
donald J. Bungum 45
own meterstick is accurate prior to comparing it to the Standard Meter (i.e.,

by a process of physical manipulation, observation, and comparison). For
until you have actually gone through the physical steps to compare your own
meterstick to the Standard Meter, you should regard your own meterstick as
“uncalibrated,” and thus you should not regard yourself as in a state to judge
reliably whether your meterstick is in fact “well calibrated.”
The notion of calibration is relevant to our target case because it seems
plausible that, because of the epistemic differences between authoritative
and lay belief, Sophia’s authoritative testimony reveals Francis’s perceptual
evidence as uncalibrated with respect to the ornithological world. Consider
that, insofar as Francis is aware of Sophia as an epistemic authority, he is
aware of her possessing evidence and understanding that is more reliable,
more relevant, more easily and expertly interpreted, and more closely associ-
ated with fundamental causes than any other evidence he possesses himself.
Sophia’s understanding and evidence, therefore, is analogous to a Standard
Meterstick for Francis’s own means of identifying cardinals. As a result,
Sophia’s evidence and understanding is precisely the sort of thing against
which Francis can and should calibrate his perceptual evidence. Nevertheless,
Sophia’s superior evidence and understanding does not “come along for the
ride” when Sophia testifies to Francis that the bird in the tree is a cardinal.
Consequently, Sophia’s testimony simultaneously gives Francis a reason to
believe that there is a cardinal in the tree and reveals his own reasons to be
uncalibrated reasons.
5.2.The Epistemological Significance of Calibration
Why does it matter whether Sophia’s testimony reveals Francis’s own reasons
to be uncalibrated? To answer this, let’s return to the meterstick example.
Suppose that you know that the Standard Meter is in town, and suppose that
you have not yet calibrated your own meterstick against the Standard Meter.
Suppose further that, while in this state, you use your meterstick to measure
an object, which your stick suggests is 57 cm wide. Now consider the fol-
lowing proposition: the object is exactly 57 cm wide.15 Intuitively, until you
have calibrated your meterstick against the Standard Meter, your measure-
ment using your uncalibrated meterstick does not provide a suitable basis for
you to believe the proposition. For it seems that you have a suitable basis
15
There are worries here about vagueness, but they do not undermine the
central point, since the same vagueness remains whether a meterstick is calibrated
or uncalibrated, but it remains possible to distinguish calibrated from uncalibrated
metersticks.
for believing the proposition about exact length only if you have a suitable
basis for believing that your own meterstick accurately measures length, but
the only way to know whether your own meterstick accurately measures
length is to compare it to a suitable standard for length, which you have not
yet done.
This example is instructive, since it suggests that, at least when a stan-
dard of calibration is readily accessible, uncalibrated reasons do not provide
suitable bases for belief. In the case of Sophia and Francis, however, Sophia’s
own understanding and evidence is simultaneously what establishes her
belief as authoritative and provides a standard for calibrating Francis’s own
means of identifying cardinals. It seems to follow, then, that until Francis has
somehow compared his method of identifying cardinals to Sophia’s superior
evidence and understanding, his perceptual evidence is uncalibrated, and, as
uncalibrated, his perceptual evidence is not a suitable basis for believing that
there is a cardinal in the tree.
Perhaps, then, the notion of calibration provides the key feature needed
by the preemptionist. For first, authoritative testimony seems to reveal any
nonauthoritative reasons possessed by a subject to be uncalibrated. Second,
reasons recognized as uncalibrated do not seem to provide a suitable basis
for belief. And third, a subject’s nonauthoritative reasons are revealed as
uncalibrated by virtue of the testifier’s very authoritativeness.
5.3. Failure of Calibration
Nevertheless, while the notion of calibration is instructive, it fails to provide

the preemptionist with the key feature he needs. We can grasp the difficulty
for calibration by asking the following question: when, exactly, is Francis’s
perceptual evidence revealed as uncalibrated? If the preemptionist wants to
tie the replacing power of authoritative testimony to the testimony itself,
then the preemptionist needs to say that Francis’s perceptual evidence is not
revealed as uncalibrated until he actually receives Sophia’s testimony about
the cardinal. But given the previous argument, Francis has all that he needs
to recognize his perceptual evidence as uncalibrated even before he learns of
Sophia’s judgment concerning the cardinal. For consider the analogous case
of the meterstick. Is it crucial that the Standard Meter arrives in your town for
you to recognize that your own meterstick is uncalibrated? No. Indeed, all that
is necessary for you to recognize your stick as uncalibrated is your recogni-
tion that (a) a standard of measurement exists and (b) you have not calibrated
your own measure against that standard. But then, in the case of Francis and
Sophia, Francis’s mere recognition of Sophia as an authority in ornithology
will be sufficient for him to recognize her evidence and understanding as a
donald J. Bungum 47
standard against which he has not yet calibrated his own means of identifi-
cation. So Francis’s mere recognition of Sophia’s epistemological superiority
will be sufficient to reveal his own perceptual evidence as uncalibrated.
We must therefore distinguish between the authoritativeness of the
testifier and the authoritativeness of the testimony itself. Preemptionism is a
view about the replacing power of authoritative testimony, not testifiers. Cali-
bration is therefore not the key feature needed by the preemptionist, since
calibration, if it presents a problem for basing at all, depends on the author-
itativeness of the testifier, not the testimony.
The failure of calibration is instructive, however, since it points out
that, in order to defend his view, the preemptionist needs to identify an epis-
temically significant feature of a subject’s own reasons that differs before
and after the subject receives authoritative testimony. We can capture this
lesson by returning to the preemptionist’s causal claim and adding a temporal
qualification:
(Causal Claim**) By virtue of the authoritativeness of authoritative tes-

timony that p received at t, if a recipient of that testimony has any
of his own reasons for believing p, those reasons come to possess
feature X at t.
Causal Claim** makes clear that a subject’s reasons must come to possess the
key feature simultaneously with the subject receiving the authoritative testi-
mony. The notion of calibration fails to provide the feature that the preemp-
tionist needs, since a subject’s reasons can be revealed as uncalibrated before
the subject receives the authority’s testimony.
6. Precision
I will now describe a different way for the preemptionist to fill out the general
strategy. I will argue that this way of filling out the general strategy is much
more promising for the preemptionist. This approach turns on the idea of pre-
cision. I will develop this approach in three steps. First, I will use two examples
to describe a kind of inference that I will call a “precision-dependent infer-
ence.” Second, I will examine the relationship between precision-dependent
inferences and proper basing. I will argue that, for a given subject, reasons
that cannot support precision-dependent inferences from p are not suitable
bases for the subject’s belief that p. Third, I will examine the relationship
between precision-dependent inferences and authoritative testimony. I will
claim that authoritative testimony that p renders a subject’s own reasons
unable to support precision-dependent inferences from p. I will conclude by
showing how the notion of precision-dependent inferences provides what

the preemptionist needs in order to maintain his characteristic claim that a
subject who believes on epistemic authority should not base his belief on his
own reasons but on authoritative testimony alone.
6.1. Precision-Dependent Inferences
In some epistemic endeavors, precision does not matter very much. For
example, suppose the sky is blackened with migrating geese. In order to know
whether there are more than ten geese migrating, a subject would not have
to count each goose individually but could simply see that the sky-blackening
number of geese is more than ten.
In other epistemic endeavors, however, precision is extremely important.
For example, suppose that you are trying to figure out what the sun is like.
One way to do this is to measure the spectrum of electromagnetic radiation
that radiates from the sun. If one measures the energy of radiation very
precisely at each wavelength, a pattern emerges that can be used to infer the
sun’s chemical composition, temperature, density, mass, distance, luminosity,
and other properties. Such patterns, however, do not emerge when one uses
an imprecise measure of the sun’s radiation, such as the feeling of the sun’s
heat on one’s face.
The point of this section is to show that the reasonability of precision-
dependent inferences depends on how precise the evidence is that supports
them. To illustrate, compare two stories. First:
naSa kitchen: Stephen is very peculiar about his steak. His

favorite method is to cook steak sous vide using a temperature-
controlled water bath. While preparing a steak, Stephen checks
the temperature of the water using his digital thermometer,
which reads 128°F. Stephen’s friend Thomas is intrigued.
Thomas pulls a thermometer out of his pocket and explains that
it is the very thermometer he uses in his work at NASA. Thom-
as asks if he can test the water using his own thermometer.
Stephen agrees. Thomas’s thermometer reads 128°F.
Now here is the second story, which is a continuation of naSa kitchen:
not ruined: While Stephen and Thomas are cooking, Nora

walks into the kitchen, holding a magazine. She tells them that,
according to America’s trendiest test kitchen, medium rare steaks
cooked sous vide are ruined once they reach 128.5°F or above.
donald J. Bungum 49
She asks Stephen whether he has he has ruined the steaks. Step-
ping in, Thomas replies, “Let me check.” Pulling out his NASA
thermometer, Thomas sets the precision setting to “maximum”
and measures the water temperature again.16 Thomas’s ther-
mometer reads 128.022°F.
Nora raises a question for Stephen that he cannot immediately answer—namely,

whether he has ruined the steaks by allowing them to reach 128.5°F. I now
want to explore whether—and why—it is reasonable for Stephen to infer
that he has not ruined the steaks.
Consider the first pattern of inference that is open to Stephen, a pattern
of inference based on the reading from his home thermometer:
Inference 1
(HT) The home thermometer reads 128°F
(WT) The water temperature is below 128.5°F
(NR) The steaks are not ruined
Intuitively, the justification that HT provides to WT does not seem sufficient

for Stephen to assure Nora that the steaks are not ruined. This is because,
on the one hand, Stephen will likely be uncertain about whether his home
thermometer changes units on the half degree or on the whole degree. And
if his home thermometer changes units on the whole degree, then a read-
ing of 128°F is consistent with the water temperature being the range of
128.5–128.9°F, which will ruin the steaks. On the other hand, Stephen will
likely be uncertain about the precise calibration of his home thermometer.
If his home thermometer is off by just a tenth of a degree, then even if
the thermometer changes on the half degree, the thermometer could read
128°F when the water temperature is 128.55°F, which will ruin the steaks.
Since Stephen will likely be uncertain about the mechanism and calibration
of his home thermometer, and since these uncertainties render him unable
to affirm that the steaks will not be ruined, it does not seem reasonable for
Stephen to assure Nora that the steaks will not be ruined on the basis of his
home thermometer readings.
Now let’s consider the second pattern of reasoning open to Stephen, the
pattern based on the precise reading from Thomas’s NASA thermometer:
16
This example is based on currently available technology. For example, adjust-
able precision settings are featured on the Traceable Platinum Ultra-Accurate Digital
Thermometer sold by Thomas Scientific out of Swedesboro, NJ.
Inference 2
(NT) The NASA thermometer reads 128.022°F
(WT) The water temperature is below 128.5°F
(NR) The steaks are not ruined
Inference 2 is the same as Inference 1 except for its first step, NT. Clearly, NT
provides justification for WT that is stronger than the justification provided
by HT, since, if we adopt the same conventions about rounding and cali-
bration as earlier, the NASA thermometer’s more precise reading provides
even better evidence that the water temperature is actually below 128.5°F.
Unlike Inference 1, however, Inference 2 does seem to be a sufficient basis
for Stephen to assure Nora that the steaks are not ruined. This can be seen in
three ways. First, the NASA thermometer provides evidence that is more
precise than is necessary to establish the proposition in question, WT. As a
result, the inference from NT to WT does not depend on conventions about
rounding, whereas the inference from HT to WT does.
Second, the inference from NT to WT is not affected by uncertainties
concerning the way in which the NASA thermometer changes units. At its
maximum precision, the NASA thermometer measures temperature to the
nearest thousandth of a degree. At this level of precision, however, it does
not matter whether the thermometer changes its reading on the whole thou-
sandth or the half thousandth, since, on either possibility, the maximum vari-
ance will be no more than two-thousandths of a degree. Such a variance,
however, is well within the safe zone for the steaks.
Third, the inference from NT to WT is not affected (or is affected far
less) by worries concerning calibration. Suppose, for the sake of argument,
that the NASA thermometer is miscalibrated by 0.3°F. On this supposition,
Stephen needs to allow that the temperature could be anywhere in the range
of 127.7°F to 128.3°F. Nevertheless, all of these temperatures are well within
the safe zone for the steaks, so, even if the NASA thermometer isn’t perfectly
calibrated, its greater precision still enables Stephen reasonably to affirm
that the steaks will not be ruined by the temperature.
The differences between these inferences show that, for a given subject,
there can be certain inferences that might appropriately be called “precision-
dependent inferences.” Specifically, in the case of not ruined, Stephen’s
precision-dependent inference is his inference from WT to NR—that is, his
inference from the water temperature’s being below 128.5°F to the steak’s
not being ruined. This inference is “precision-dependent” for Stephen, since
whether the inference is reasonable for him depends on the precision of
the evidence that he has in support of the water’s temperature: making the
same assumptions about calibration and rounding conventions, the inference
donald J. Bungum 51
is reasonable for Stephen when he has precise evidence and unreasonable for
Stephen when he lacks precise evidence.
6.2. Precision-Dependent Inferences and

the Generalized Preemptionist Strategy
I have now described and defined what a precision-dependent inference is,

and I have analyzed a case in which I claim a precision-dependent infer-
ence exists. My next task is to show that the notion of precision-dependent
inferences provides what the preemptionist needs in order to defend his key
claim that the recipient of authoritative testimony that p should not base his
belief that p on his own reasons for p. In order to defend the preemptionist’s
key claim, my method will be to fit the notion of a preemption-dependent
inference into the Generalized Preemptionist Strategy as described and
refined earlier. To review, the two elements of the strategy are as follows:
(Causal Claim**) By virtue of the authoritativeness of authoritative tes-

timony that p received at t, if a recipient of that testimony has any
of his own reasons for believing p, those reasons come to possess
feature X at t.
(Basing Claim) Any reason for believing p marked by feature X is not a
suitable basis for believing p.
With these two claims in place, the burden on the preemptionist is to show
that precision-dependent inferences somehow constitute a “feature X” that
renders true both the causal claim and the basing claim.
In a nutshell, here is how I will fill out the “feature X” in the Generalized
Preemptionist Strategy: I will argue that, in cases of authoritative testimony
that p, a subject’s own reasons become unable to support the precision-
dependent inferences that the subject may reasonably make from his belief
that p. In other words, I will argue that the preemptionist should fill out the
Generalized Strategy in the following way:
(Causal Claim**—PDI) By virtue of the authoritativeness of authori-

tative testimony that p received at t, if a recipient of that testimony
has any of his own reasons for believing p, those reasons at t become
unable to support precision-dependent inferences that the recipient
may reasonably make from p.
(Basing Claim—PDI) Any reason for believing p that cannot support
precision-dependent inferences that a subject may reasonably make
from p is not a suitable basis for believing p.
In the following sections, I will argue for each of these claims, starting with
the basing claim. If my arguments are successful, then preemptionism can
be defended on the basis of the relationships between epistemic authority,
basing, and precision-dependent inferences.
6.3. Precision-Dependent Inferences and Basing
In order to defend the PDI version of the basing claim, it will be useful to
reflect on how Stephen should base his beliefs in naSa kitchen and not
ruined.
As I argued, Stephen’s inference from the water’s being below 128.5°F
(WT) to the steaks being unharmed (NR) is a precision-dependent inference,
and this inference is reasonable for him to make because of the precise read-
ing provided by the NASA thermometer at its maximum precision. More-
over, I argued that the home thermometer reading (HT) cannot support
Stephen’s inference from WT to NR, so it follows by the PDI version of
the basing claim that HT is not a suitable basis for Stephen’s belief that WT.
In other words, according to the basing claim, Stephen should not base his
belief that WT on his belief that HT.
Critics of preemptionism will find this latter consequence implausible:
is it really true that the home thermometer reading contributes nothing to the
basis for Stephen’s belief about the water temperature? After all, we have
conceded at the outset that it is right for Stephen to regard his home ther-
mometer as reasonably reliable, and the NASA thermometer reading (NT)
seems to confirm that Stephen’s home thermometer is accurate. Why, then,
should we think that HT cannot be counted as part of the suitable basis
for Stephen’s belief that WT?
I will answer this question in two parts. First, I will argue that HT by
itself is not part of the suitable basis for Stephen’s belief that WT. Second, I
will argue that HT is not part of the suitable basis for WT even in conjunction
with NT. I take it that these possibilities exhaust the ways in which HT might
be part of the suitable basis for Stephen’s belief that WT. I also take it that
Stephen’s situation is representative of other situations in which a subject is
able to make reasonable precision-dependent inferences from p but has other
reasons for p that do not support those precision-dependent inferences.
Consequently, if my arguments are correct, then the critic of preemptionism
is wrong to reject the PDI version of the basing claim.
donald J. Bungum 53
6.3.1. HT by Itself
Why should we believe that the home thermometer reading (HT) by itself
is not a suitable basis for Stephen’s belief that that the water temperature
is below 128.5°F (WT)? One way to argue for this claim is to point to the
concerns about the home thermometer’s mechanism and calibration that I
pointed out earlier: Stephen is likely uncertain about the mechanism and cal-
ibration of his home thermometer, so he probably is not justified in making
the assumptions needed to regard HT as a guarantee that WT.
A better way to argue for this claim, however, is to reflect on some of
the necessary conditions for a suitable basis for belief. Plausibly, a basis for
believing that p is suitable only if that basis favors p over other propositions
on which p is false. For example, suppose you have a coin, and suppose that
you know the coin is fair. Is your knowledge that the coin is fair a suitable
basis for believing that the coin will turn up heads on the next toss? No. And
the reason is clear: the fairness of the coin makes it equally likely that the coin
will turn up heads or tails, so the knowledge that the coin is fair does not
make it any more likely that the coin will turn up heads than that it will
not turn up heads. The coin’s fairness does not favor a toss of heads, so the
coin’s fairness is not a suitable basis for believing that the coin will come up
heads. In contrast, suppose that you have a different coin, and suppose that
you know that this coin is significantly biased toward heads. Is your knowl-
edge of the coin’s bias a suitable basis for believing that the coin will turn up
heads on the next toss? Well, it will not license absolute confidence, but this
knowledge will at least make it more reasonable for you to predict heads than
in the case of the fair coin—your knowledge of the coin’s bias is at least the
right kind of thing to make it reasonable for you to predict heads, since it
favors the coin turning up heads over it not turning up heads.
Consider now the relationship between HT and WT (recall that HT
is the proposition that “The home thermometer reads 128°F” and WT is
the proposition that “The water temperature is below 128.5°F”). If HT
is a suitable basis for Stephen to believe WT, then HT needs to favor WT in
something like the way in which your knowledge of your coin’s bias favors
its coming up heads. But HT does not favor WT in this way. For clearly, HT
does not entail WT: the home thermometer could read 128°F even if the
water temperature is above 128.5°F. And even on certain reasonable assump-
tions about the home thermometer’s mechanism and calibration (e.g., rounds
on the half integer; perfectly calibrated), HT makes it equally likely that WT
is true as that WT is false. By itself, therefore, HT does not favor WT in the
way that it needs to in order to provide a suitable basis for believing WT. By
itself, therefore, HT is not a suitable basis for believing WT.
6.3.2. HT with NT
At this point, however, an objector might respond that, even if HT is unsuit-

able by itself to form the basis for Stephen’s belief that WT, it does not
follow that HT cannot be part of the basis for Stephen’s belief that WT at
all. Consider the case of the fair coin. Even if your knowledge that the coin
is fair is not a suitable basis for believing that the coin will turn up heads,
you might learn that a mad scientist lurks nearby with a device that makes all
fair coins turn up heads. Clearly, in such a case, your knowledge of the coin’s
fairness can be part of the suitable basis for believing that the coin will land
heads on the next toss. So something similar might happen in the case of HT.
Moreover, in the case of not ruined, there is something unusual nearby that
seems to increase or confirm the significance of the home thermometer’s
reading—namely, the precise reading from the NASA thermometer (NT).
Perhaps, then, HT can be part of the suitable basis for Stephen’s belief that
WT even if it is not a suitable basis by itself.
In the face of this possibility, the proponent of the PDI version of the
basing claim has two options. On the one hand, he can concede the point to
the objector and maintain that a subject’s own reasons are unsuitable bases
for belief only when they are considered in themselves. On the other hand, he
can double down and argue that a subject’s own reasons are unsuitable bases
for belief even when their relations to other reasons are considered. I will
argue for the latter, stronger view. Accordingly, I will consider the two main
ways in which the home thermometer’s reading might stand in a support rela-
tion to the NASA thermometer’s reading and thereby enter into the proper
basis for Stephen’s belief that the steaks are not ruined. According to the first
way, the NASA thermometer’s reading might somehow support the home
thermometer’s reading. According to the second way, the home thermometer
reading might somehow support the NASA thermometer’s reading. I will
argue that, on either of these ways of viewing the support relations between
HT and NT, HT is not a suitable part of the basis for Stephen’s belief
that WT.
Let us first consider the possibility—that is, that the NASA thermometer
reading confirms the goodness of the home thermometer reading as a reason
for believing that the steaks are not ruined.17 Now it is obvious that, once
17
Thanks to Alex Plato for this suggestion. Jäger stresses a similar point in
his criticism of preemptionism. He argues that, in cardinal, Sophia’s authoritative
testimony confirms the goodness of Francis’s perceptual reasons for believing that
there is a cardinal in the tree, making it even less reasonable for Francis to unhinge
his belief about the cardinal from his own perceptual reasons. See Jäger, “Epistemic
Authority,” 177.
donald J. Bungum 55
the precise reading from the NASA thermometer is revealed, Stephen gains
a strong reason for thinking that his home thermometer is not wildly miscal-
ibrated. This is because it is reasonable to expect Thomas’s NASA ther-
mometer not to be wildly miscalibrated, and Stephen’s home thermometer
matches its reading quite closely. For the same reason, it is also obvious that
the NASA thermometer confirms that the home thermometer is quite accu-
rately calibrated. So it seems quite correct to say that the reading from the
NASA thermometer should somehow strengthen Stephen’s confidence in his
own home thermometer.
Nevertheless, when we are considering HT as part of the basis for NR,
we have to keep in mind the specific deficiencies of the home thermometer
that I mentioned earlier. First, there is the worry about the home thermom-
eter’s mechanism of changing units, whether on the whole degree, the half
degree, or something else entirely. This worry concerns the specific design
plan of Stephen’s home thermometer, and Thomas’s NASA thermometer
reading clearly reveals nothing about this design plan. Second, there is the
worry about the home thermometer’s precise degree of calibration. For
example, if the home thermometer is miscalibrated by just 1 percent of its
face value (i.e., roughly 1°F), then this will lead to a significant number of
cases in which Stephen will be in error about whether the steaks are ruined.18
This worry concerns the way in which changes in the world are reflected by
changes in the thermometer’s display, and yet again, the NASA thermometer
reveals nothing about how these changes are related to or correlated with
each other. We can make an analogous point in the case of the fair coin. If
you strongly and reasonably suspect that your coin is fair, then it will not
help you predict your coin’s next toss if the angel Gabriel assures you that
your coin is fair. Just as Gabriel’s confirmation of your coin’s fairness does
not favor a toss of heads, so the NASA thermometer’s confirmation of the
home thermometer’s accuracy and reliability does not favor a temperature
below 128.5°F. Consequently, even if the NASA thermometer confirms that
Stephen’s home thermometer is quite good, it does not support the home
thermometer’s readings in the specific ways necessary for it to figure in the
proper basis for Stephen’s belief that the water is below 128.5°F.
Let us now consider the converse possibility—namely, that the home
thermometer confirms the NASA thermometer reading as a reason to
believe WT.19 Consider eyewitness testimony of a crime. Suppose that several
18
For example, if the home thermometer is high by 1°F, then the steaks will
often not be ruined when the thermometer reads 129°F. Or if the home thermome-
ter is low by 1°F, then the steaks will often be ruined when the thermometer reads
128°F.
19
Thanks to Logan Gage for this suggestion.
eyewitnesses report seeing a crime (e.g., “a short man killed a tall man”),
but some eyewitnesses report additional details due to their closer prox-
imity to the crime scene (e.g., “the short man had blue eyes and a diamond
earring”). Now divide these reports into less detailed “general reports” and
more detailed “specific reports.” In deliberating about the crime, it seems
perfectly legitimate for a subject (e.g., a juror) to include both the general and
specific reports within the basis for his belief that the crime was committed
by a short man with blue eyes. This is because the general reports, even
though they are less precise than the specific reports, still lend credibility
to the specific reports, insofar as all of the reports attest to the event
having the same general features. Furthermore, the general reports can be
included in the basis of a subject’s belief even if he is uncertain about exactly
how the general reports were generated (e.g., where exactly were the eyewit-
nesses standing?) or exactly how accurate they are (e.g., what exactly are the
eyesight measurements for each eyewitness?). But if these points are correct,
then it seems that, by analogy, the critic of preemptionism can say that HT
lends credibility to NT. For if Stephen’s home thermometer is reasonably well
calibrated, then HT should make Stephen more confident that the NASA
thermometer is not wildly miscalibrated (a remote but genuine possibility),
and so HT seems to increase the confidence that Stephen can place in the
NASA thermometer’s precise reading. But if HT strengthens NT as a reason
to believe NR, then it seems reasonable to contend that HT is a suitable part
of the basis for NR, at least insofar as it supports NT. On this possibility,
therefore, the preemptionist is mistaken to contend that HT is not a suitable
part of the basis for Stephen’s belief that WT.
In my view, this is the stronger of the two objections to the PDI version
of the basing claim. Nevertheless, I think that the preemptionist can answer
this objection. The key point is to show that HT cannot provide the right kind
of support to NT in order to be a suitable part of the basis for Stephen’s
belief that WT. In order to see this, we will need to develop a distinc-
tion between two different kinds of evidence within the basis for a belief.
Consider a bomb dog. Suppose that you know that this dog has a strong
track record of correctly identifying bombs. You then see the dog indicate
a bomb. While the dog’s track record by itself does not favor there being a
bomb on any particular occasion, your knowledge of the dog’s track record
does make it more reasonable for you to trust the dog’s indication when
he does signal a bomb. Since it is obvious that your knowledge of the dog’s
track record should be counted as part of the basis of your belief that there is
a bomb, let us distinguish, then, between “indicator evidence” and “trustwor-
thiness evidence” within the basis for a belief. “Indicator evidence” will be
evidence that favors the truth of the belief, while “trustworthiness evidence”
is evidence that the source of the indicator evidence is in fact reliable.
donald J. Bungum 57
Now with this distinction between types of evidence in hand, we can

now see that the objector wants to view HT as trustworthiness evidence
supporting NT. Indeed, this is the point of the objector’s analogy with the
crime scene, since, in the crime scene situation, the general reports serve
as evidence that the specific reports are reliable. In fact, however, HT
cannot provide trustworthiness evidence for NT. Or, more precisely,
HT cannot provide trustworthiness evidence of the right sort. In order to see
this, it is important to recognize that sources of evidence can differ not only
in their degree of trustworthiness but also in their kind of trustworthiness.
For example, a mechanical metronome might have a higher degree of reli-
ability in determining tempo than a metronome app on a smartphone, but
that same mechanical metronome will have no reliability in determining your
geophysical coordinates, which the smartphone might do quite reliably. We
must therefore distinguish between the domain of a source’s trustworthiness
(i.e., what it can be trusted to indicate) and the degree of a source’s trust-
worthiness (i.e., how reliable it is).
Now in the case of the crime scene scenario, the general reports are
evidence that the specific reports will have a higher degree of reliability in the
relevant domain, which is determining the gross physical qualities and actions
of human beings. This is because eyewitnesses to a crime scene are all human
beings using their perceptual faculties, and the fact that more distant eyewit-
nesses can observe a certain event is good evidence that more proximate
eyewitnesses will be able to observe the same event along with further details.
In the case of HT and NT, however, the analogy does not hold. The
home thermometer is designed to measure temperature to the nearest whole
degree. At its best, therefore, the domain of the home thermometer’s trust-
worthiness is measuring temperature to the nearest whole degree. Never-
theless, even if the home thermometer is perfectly reliable in this task, it is
something very different to measure temperature to the nearest thousandth
of a degree. (Indeed, it is at least logically possible that a certain thermometer
should be simultaneously perfectly reliable in indicating temperature to the
nearest whole degree and perfectly unreliable in indicating temperature to
the nearest thousandth of a degree.) Consequently, even if we grant that the
home thermometer is perfectly reliable according to its design plan, and even
if we observe that HT and NT agree about the temperature to the nearest
whole degree, it does not follow that HT provides the right kind of trust-
worthiness evidence to strengthen NT. The trustworthiness of the NASA
thermometer is confirmed only by something that touches on its specific
domain of operation (i.e., indicating temperature to the nearest thousandth
of a degree), and the home thermometer does not come close to confirm-
ing the reliability of the NASA thermometer at this degree of specificity.
And more to the point, the home thermometer does not even confirm the
reliability of the NASA thermometer in indicating temperature to the near-

est tenth of a degree, which is precisely the sort of reliability at stake when it
comes to determining whether the water temperature is below 128.5°F.
The lessons for Stephen’s situation in not ruined can now be drawn
out. The basis for a subject’s belief that p can rightly include trustworthiness
evidence, even if that trustworthiness evidence does not directly favor the
truth of p. Moreover, HT does indeed provide a certain degree of trust-
worthiness evidence for NT, since the home thermometer’s reading suggests
that the NASA thermometer is trustworthy for indicating temperatures down
to the nearest whole number. Nevertheless, HT is a suitable part of the basis
for Stephen’s belief that WT only if HT provides the right kind of trust-
worthiness evidence for NT. For example, it would be irrelevant to Stephen’s
belief that WT if he were to receive confirmation that the NASA thermome-
ter were also a highly reliable metronome. But HT does not provide the
right kind of trustworthiness evidence for NT—Stephen’s belief that WT
depends on the NASA thermometer reliably indicating temperature at least
down to the nearest tenth of a degree, but his home thermometer provides
no evidence concerning whether other thermometers are reliable to the near-
est tenth. It follows that HT is not a suitable part of the basis for Stephen’s
belief that WT, even when HT is considered in relation to NT, and even
when we concede that HT provides a certain kind and degree of confirma-
tion for NT.
The lessons from Stephen’s situation also apply to all other situations
covered by the PDI version of the basing claim, which is as follows:
(Basing Claim—PDI) Any reason for believing p that cannot support

precision-dependent inferences that a subject may reasonably make
from p is not a suitable basis for believing p.
If a reason R for believing p cannot support precision-dependent infer-

ences that a subject may reasonably make from p, then, like Stephen’s belief
HT, R is not a suitable basis by itself for the subject to believe p. In that
case, R may be a suitable part of the subject’s belief that p only if it stands
in the right sort of relationship to reasons for believing p that do support the
subject’s precision-dependent inferences from p. But R either supports or is
supported by those inference-supporting reasons, and in either case, it is still
not a suitable part of the subject’s basis for p. For the inference-supporting
reasons can provide only trustworthiness evidence for the source of R, but
such trustworthiness evidence does not fix the fundamental problem with R,
which is its lack of precision. And R can provide only trustworthiness evi-
dence for the inference-supporting reasons, but the trustworthiness evidence
that it can provide is of the wrong kind, since it is not precise enough to
donald J. Bungum 59
establish the inference-supporting reasons as reliable down to the required

degree of precision to support the inference. Consequently, R cannot be a
suitable basis for the subject’s belief that p, even when R is considered in
relation to the subject’s inference-supporting reason. And this is just what the
PDI version of the basing claim asserts.
6.4. Precision-Dependent Inferences and Authoritative Testimony
It is now time to examine the relationship between precision-dependent

inferences and authoritative testimony. In this section, I will be defending the
PDI version of the preemptionist’s causal claim, which says that authoritative
testimony, by virtue of its authoritativeness, renders a subject’s own reasons
incapable of supporting precision-dependent inferences that the subject may
reasonably make.
Here is a sketch of my argument for the preemptionist’s causal claim.
My key point is that authoritative testimony, as such, contains “implicit preci-
sion” that enables its recipients reasonably to make precision-dependent
inferences that they could not have made before receiving the testimony.
But since authoritative testimony increases the number of precision depen-
dent inferences that a recipient can make without affecting the ability of
a subject’s own reasons to license those inferences, authoritative testimony
“renders” the subject’s own reasons unable to support certain precision-
dependent inferences. This “rendering” has nothing to do with undermining
or contradicting the subject’s own reasons. Rather, authoritative testimony
causes the subject’s own reasons to be unfit only by extending the scope
of the subject’s legitimate cognitive activities beyond what the subject’s own
reasons can sustain.
Following on this sketch, there are three questions that I want to treat in
defense of the preemptionist’s causal claim. First, what is implicit precision?
Second, why should we think that authoritative testimony as such contains
implicit precision? And third, why should we think that the implicit precision
contained in authoritative testimony licenses precision-dependent inferences
that a recipient’s own reasons cannot support? I will address these questions
in order.
Let us turn to the idea of implicit precision. Using the cases of naSa
kitchen and not ruined, we can approach the idea of implicit precision
negatively by comparison with the idea of explicit precision. Recall that, in
naSa kitchen, both the home thermometer and the NASA thermometer
give the same reading of 128°F. In contrast, in not ruined, the NASA ther-
mometer provides an additional measurement of 128.022°F. Now a reading
of 128.022°F rules out all the alternative temperatures ruled out by a reading
of 128°F, but not only those. Additionally, it rules out many other alter-
native temperatures that differ from each other by one one-thousandth of a
degree. Therefore, the explicit determination of the NASA thermometer in
not ruined is more precise than the explicit determination of either ther-
mometer in naSa kitchen. Let us summarize this idea by saying that the
determination of the NASA thermometer in not ruined possesses greater
explicit precision than the determinations of either thermometer in naSa
kitchen.
Now it might seem that the only way for two thermometers to differ
from one another with respect to precision is for them to differ with respect
to explicit precision. But this is not true. Consider this slightly modified
continuation of the original naSa kitchen case:
not ruined*: While Stephen and Thomas are cooking, Nora

walks into the kitchen, holding a magazine. She tells them that,
according to America’s trendiest test kitchen, medium rare steaks
cooked sous vide are ruined once they reach 128.5°F or above.
She asks Stephen whether he has he has ruined the steaks.
Stepping in, Thomas replies, “My first thermometer reading
was 128°F, so I can guarantee that the water is below 128.5°F.
But let me prove it to you.” Pulling out his NASA thermometer,
Thomas sets the precision setting to “maximum” and measures
the water temperature again. Thomas’s thermometer reads
128.022°F.
The only difference between not ruined and not ruined* is what Thomas
says when he steps in to help Stephen. In not ruined*, Thomas cites his
NASA thermometer’s earlier reading of 128°F, and he uses this reading to
guarantee that the water temperature is below 128.5°F.
The modified case of not ruined* is useful for getting at the idea of
implicit precision. In not ruined*, Thomas makes a certain guarantee,
and this guarantee in a sense “goes beyond” the degree of precision that
is contained in the NASA thermometer’s original reading of 128°F. Never-
theless, intuitively, there is a way in which Thomas’s guarantee, based on the
imprecise reading, might still be reasonable. For if we suppose that Thomas
is familiar with how his NASA thermometer works, we can then suppose
that Thomas knows that the thermometer’s reading of 128°F is grounded in
a determination of temperature that is more precise than the determination
explicitly displayed. If Thomas knows that his NASA thermometer’s reading
of 128°F is grounded in a more precise determination of temperature, then
it can be reasonable for him to guarantee something that goes beyond the
explicit precision contained in that reading.
donald J. Bungum 61
We are now up against the idea of a determination having implicit preci-

sion, and we can characterize implicit precision by thinking about what would
make it reasonable for Thomas to offer his guarantee in not ruined*. I
submit that Thomas’s guarantee is reasonable if he fulfills the following
conditions:
(Capability) He knows that his thermometer is capable of measuring

temperature with a degree of precision greater than the one mani-
fested in its first, imprecise reading.
(Activity) He knows that his thermometer’s capacity for measuring
temperature more precisely is active in and responsible for its first,
imprecise reading.
(Connection) He knows how his thermometer’s less precise readings are
connected in lawlike ways to the more precise determinations that it
is capable of making but is presently not manifesting.
It is not hard to imagine Thomas meeting these conditions, and his meeting
these conditions seems to explain why it is reasonable for him to offer his
guarantee in not ruined*. In contrast, it seems unreasonable for Stephen to
offer the same guarantee based on the reading from his home thermometer,
since his home thermometer (we may suppose) is not capable of measuring
temperature more precisely than to the nearest whole degree, and so it is
impossible for Stephen to meet the Capability condition.20
Insofar as the three conditions pick out what it is for the NASA ther-
mometer’s implicit precision to be known, they also help describe what it is
for the NASA thermometer’s implicit precision to exist. Following the condi-
tions, then, a determination or measurement can be said to possess implicit
precision when (a) what is responsible for the determination is capable of
measuring the reality in question more precisely than its explicit determina-
tion suggests, (b) the thing’s ability to measure the reality more precisely is
20
We should note, in passing, that one of the reasons calibration is important is
that it can enable us to use a measuring device beyond its degree of explicit precision.
For example, if Stephen were to calibrate his home thermometer against the NASA
thermometer, he could come to realize that his home thermometer comes to read
128°F precisely when the NASA thermometer reads 128.12°F. Stephen could then
use his knowledge about the calibration of his home thermometer with other knowl-
edge (e.g., the time of his thermometer’s transition from 127°F to 128°F, the rate
of increase in temperature of the item being measured, etc.) to support precision-
dependent inferences that readings from his uncalibrated home thermometer could
not support. Stephen’s using his thermometer in this way, however, will never pro-
vide the same support for precision-dependent inferences that comes from a precise
reading from the NASA thermometer.
active in and responsible for its less precise determination, and (c) the thing’s
less precise determination is connected in lawlike ways to its more precise
determinations. Stephen’s home thermometer fails condition (a), so these
conditions do not enable us to say that its readings possess implicit precision.
To this point, I have characterized implicit precision and described a
case, not ruined*, in which implicit precision seems to be present. Let us
now turn to the second question of this section—namely, why should we
think that authoritative testimony, as such, contains implicit precision? In
order to answer this question, it will be useful to review the earlier asymme-
tries between authoritative and lay belief. Recall the various ways that Sophia
the ornithologist is more sensitive to the ornithological environment than a
layperson like Francis. First, she is superior with respect to remembering and
distinguishing observable properties such as head shape, beak shape, beak
color, eye ring, eye line, crest, wing angle, foot color, leg color, tail length, tail
shape, tail markings, flight pattern, song pattern, perching location, perching
posture, regional distribution, migration pattern, song, mating call, and many
others. Second, she will have a much better understanding of how strongly
these factors are correlated with a bird’s being a certain species. Third, she
will be more practiced at recognizing when these properties are instantiated
in concrete particular situations.
I take it that these ornithological asymmetries are representative of the
types of asymmetries that exist in other domains. But if such asymmetries
are representative, then it is easy to show that authoritative testimony meets
the three conditions for implicit precision given earlier. Consider first the
Capability condition. If the ornithologist Sophia tells Francis that the bird in
the tree is a cardinal, it would be reasonable for Francis to go on to ask her
whether any of the bird’s features were typical for cardinals, whether there
was anything unusual about the bird, whether it manifests any regional vari-
ations, or other similar questions. In other words, it would be reasonable for
Francis to expect Sophia to be tracking not only what species the bird is but
also the ways in which it is more and other than its species. To be capable of
such determinations, however, is to be capable of “measuring” the ornitho-
logical world at a higher degree of precision than that of species alone, so
Sophia easily meets the Capability condition for implicit precision.
Consider next the Activity condition. There are many reasons to think
that human authorities can employ a level of sensitivity to phenomena that
does not find representation in their testimony. First, ordinary language
contains many phrases that suggest this, such as, “There is much more that
I could say on this point,” and so on. Second, human cognitive processing
appears to employ many elements working holistically and in parallel. This
makes it seem unlikely that an epistemic authority could in any way select a
“less precise” mode of processing when more precise modes are available.
donald J. Bungum 63
Third, for an ornithologist like Sophia, the hundreds of distinguishable

aspects of a given cardinal will each seem fitting or “cardinalish” to her, and
each of these aspects will be present and available to her in her perception
of the cardinal. Nevertheless, when Francis asks Sophia what kind of bird is
before them, it would not make sense for Sophia to follow her reply with a
list of hundreds of the bird’s features and why they are typical of cardinals.
So it seems clear that an epistemic authority like Sophia will satisfy the Activ-
ity condition, since her capacity for more precise determinations seems to be
active in her less precise recognition of the bird as a cardinal.
Consider now the connection condition. In essence, the connection
condition says that there are intelligible, lawlike connections between the
source’s less precise explicit determinations and the more precise determi-
nations that the source is able to make but does not make in a particular
instance. The connection condition is important, since, if there were not
intelligible connections between a source’s more and less precise determina-
tions, then the source’s less precise determinations could not provide other
persons with access, reasonable or otherwise, to the greater degree of preci-
sion of which it is capable. It seems, however, that epistemic authorities are
the kinds of persons for whom there are intelligible, lawlike connections
between their more and less specific determinations. Perhaps the clearest sign
of this is that we reasonably expect epistemic authorities to be able to teach
what they know.21 For example, if Sophia were to tell Francis, “That’s a cardi-
nal,” then it would be reasonable for Francis to reply by saying, “Oh. And
how does one tell a cardinal from a noncardinal?” If Sophia were to respond
to Francis by saying, “You know, I have no idea,” then Francis’s confidence
in Sophia as an authority in ornithology would reasonably be shaken. The
fact that we expect epistemic authorities to be able to teach, however, means
that we expect them to be able to distinguish the relevant from the irrelevant,
the essential from the accidental, which is just to say that we expect epistemic
authorities to measure and judge the world more precisely than nonauthor-
ities and then to connect their precise judgments to other judgments that
we laypersons can understand. So it makes sense to say that the less specific
determinations of epistemic authorities are intelligibly connected to more
specific determinations that ground them.
I have now argued that authoritative testimony satisfies the conditions
for containing implicit precision. It now remains to be shown that the implicit
precision contained in authoritative testimony licenses precision-dependent
inferences that a subject’s own reasons cannot support.
21
For more on the relationship between testimony and teaching, see Will Small,
“Teaching and Telling,” Philosophical Explorations 17, no. 3 (2014): 372–87.
Since this whole essay has been framed as a response to Jäger’s unhinging
objection, it seems fitting to argue for this last point through the cardinal
case, which undoubtedly provides the strongest source of support for his
criticism of preemptionism. Returning to the case of cardinal, then, let
us suppose that sometime after his walk with Sophia, Francis is questioned
about the birds he saw. Francis says that he saw a northern cardinal. The ques-
tioner then asks whether he is sure the bird was not a vermillion flycatcher,
scarlet tanager, cedar waxwing, desert cardinal, vermillion cardinal, or north-
ern cardinal hybridized with a desert cardinal. Now if Francis does not know
very much about birds, then this barrage of questions may seem overwhelm-
ing to him. But for present purposes, the question is whether and how Fran-
cis has sufficient resources rationally to maintain confidence that he saw a
northern cardinal during his walk with Sophia, even though he is himself
unable to distinguish northern cardinals from these look-alike species and
even though he maybe has never heard of some of these species.
As before, there are two patterns of inference open for Francis to use in
response to the question. The first is this:
Cardinal Inference 1
(S) The bird seemed like a northern cardinal to me.
(C) The bird was a northern cardinal.
(NCLA) The bird was not any close cardinal look-alike.
The second pattern of inference is this:
Cardinal Inference 2
(OT) An ornithologist testified that the bird was a northern cardinal.
(C) The bird was a northern cardinal.
(NCLA) The bird was not any close cardinal look-alike.
Now the critic of preemptionism will point out that northern cardinals are
fairly distinctive and that Francis is fairly reliable in identifying northern car-
dinals. The critic will therefore contend that S provides nonnegligible justifi-
cation for C, so, since it seems wrong to fail to base one’s belief on justifiers
for that belief, it would be wrong for Francis to fail to base his belief in C on
his own observation.
I have argued, however, that the situation is more complicated than
the critic’s argument would suggest. For once the two inferences are laid
out, it should be clear that, for Francis, the inference from C to NCLA is a
precision-dependent inference. This is because whether or not it is reasonable
donald J. Bungum 65
for Francis to come to hold NCLA on the basis of an inference from C is

dependent on the degree of precision present in his perception of the bird
as in S. For instance, at the one extreme, if every red bird seems like a cardi-
nal to Francis, then it will not be reasonable for him to infer NCLA from C,
whereas, at the other extreme, if Francis knows each bird on Earth by name
and can identify them individually, then it is perfectly reasonable for him to
infer NCLA from C.
Jäger’s cardinal case does not provide enough detail for us to determine
whether the inference from C to NCLA is reasonable for Francis in the story.
Nevertheless, it is easy to imagine the sort of person (I myself am one) who
would be confused by the differences between a northern cardinal and a
vermillion cardinal. Moreover, authoritative testimony is most useful for the
very people who cannot distinguish close cases themselves. It is legitimate
for us to suppose, then, that the appearance of the cardinal to Francis is not
enough for him to rule out the various cardinal look-alikes proposed to him.
On this supposition, Cardinal Inference 1 is not strong enough for Francis to
maintain his confidence.
The real question, therefore, is whether Cardinal Inference 2 is strong
enough for Francis to maintain his belief that he saw a northern cardinal
and to dismiss the cardinal look-alikes. I think that it is indeed strong
enough, and I think that the implicit precision in Sophia’s testimony is what
accounts for its strength. In the face of challenges to his viewing a northern
cardinal, Francis could reason to himself in the following way, “I asked Sophia
whether it was a cardinal, and she took a good look at it from a close distance
in good lighting and told me that it was. She seemed relaxed, interested, atten-
tive, and engaged. There was nothing at all unusual about the situation, and
she is a competent ornithologist.” Now of course, in this scenario, Sophia
does not explicitly mention and dismiss all the possible cardinal look-alikes,
so her testimony does not feature explicit precision that Francis can utilize
in ruling out cardinal look-alikes. Nevertheless, explicit precision is not the
only type of precision that is epistemologically relevant, and Francis’s situa-
tion seems to be similar to that of Thomas in not ruined*, when Thomas
guarantees that the steaks are not ruined on the basis of the less precise
reading by his NASA thermometer. Like Thomas in not ruined*, Francis
can know that Sophia is capable of distinguishing northern cardinals from
cardinal look-alikes, he can know that she employed her powers of precise
determination when she identified the bird on her walk, and he can know
that there are intelligible, lawlike connections between the testimony that she
gives to laypersons and the precise judgments that she makes about a bird’s
individual features. Francis’s knowledge of Sophia as an ornithologist, then,
functions like Thomas’s familiarity with the different precision settings on
his NASA thermometer: Francis’s knowledge of Sophia’s as an ornithologist
gives him access to some of the precision implicit in her less precise deter-
mination, and it permits Francis reasonably to make an inference that only
significant precision in judgment can license.
7. Conclusion
I have now completed my defense of preemptionism. On my view, authorita-

tive testimony possesses implicit precision, and this implicit precision enables
recipients reasonably to carry out precision-dependent inferences that their
own reasons cannot support. I have also argued, however, that if a subject
can reasonably make precision-dependent inferences from p, then other rea-
sons that subject may possess for p are not suitable parts of the subject’s basis
for believing p. It follows that preemptionism is true: recipients of authori-
tative testimony should base their beliefs on that testimony alone and not on
their own reasons.
In her response to Jäger’s unhinging objection, Zagzebski says this:
“Unlike me, the ornithologist would not be fooled by a cardinal look-alike,
if there is such a thing. The ornithologist identifies much more about the
bird than I can, and in fact, the ornithologist with her trained eye sees more
than I see. She has already included whatever in my experience indicates
a cardinal, but she has included more.”22 Zagzebski does not spell out what
the “more” is that the ornithologist sees, nor does she explain why it is
epistemologically significant that the ornithologist’s experience includes more
than that of an ordinary layperson. Nevertheless, in my view, Zagzebski is
on the right track, and this essay has argued that the way to develop preemp-
tionism is through the relationships among authoritative testimony, proper
basing, and precision-dependent inferences. On the preemptionism devel-
oped in this essay, to believe on authoritative testimony is to believe on the
word of another who sees “more” because she sees “more precisely.” This
point is significant because, through authoritative testimony, laypersons can
reasonably make precision-dependent inferences that their own reasons are
not precise enough to support. Authoritative testimony, on my view, enables
laypersons to live and reason in some sense as authorities, even though their
own abilities do not enable them to live or reason this way. Insofar as such an
“elevated” cognitive life requires a proper basis, therefore, this proper basis
is found in the authoritative testimony alone and not in the subject’s own
reasons.
Zagzebski, “Replies to Christoph Jäger,” 4. Pronouns changed for the sake

22
of consistency.
donald J. Bungum 67
In conclusion, while Jäger is correct to point out that two reasons are
generally better than one, he is wrong to suppose that the subject’s own
reasons are always sufficient for the tasks to which they are put. In reply to
Jäger’s unhinging objection, therefore, the preemptionist can reply that the
recipient of authoritative testimony does well to set his own reasons aside
until he can see precisely enough to reason through them. According to my
version of preemptionism, authoritative testimony involves not “epistemic
regression” but rather “epistemic elevation,” and my view points out that it
just might take time for a subject’s own reasons to catch up to his new and
higher form of epistemic life.23
23
Thanks to Eleonore Stump, John Greco, Jonathan Jacobs, Christoph Jäger,
Beth Rath, Logan Gage, and Alex Plato for helpful comments on earlier drafts of
this essay.
The Unassertability of Contextualism
Martijn Blaauw
Delft University of Technology
Jeroen de Ridder
Vrije Universiteit Amsterdam
1. Introduction
There are many versions of epistemological contextualism. They all share the
idea that the truth-conditions of knowledge sentences depend on the con-
versational context of the attributor of knowledge. But they differ in what
shifts with context and how the shift operates. On Keith DeRose’s version of epis-
temological contextualism, what shifts with the conversational context of
the attributor of knowledge is how strong an epistemic position a subject
must be in to satisfy “knows” (where a subject’s strength of epistemic posi-
tion is determined by how far out into logical space she can track the truth),
and how the shift operates is dictated by the “Rule of Sensitivity.” DeRose
originally defended his version of contextualism by arguing that it provides
a convincing answer to the problem of radical skepticism and, furthermore,
reconciles seemingly conflicting intuitions about certain puzzle cases. He has
subsequently added a third argument for contextualism: it is powerfully moti-
vated by the knowledge account of assertion.1
In what follows, we argue that the contextualist, in basing contextualism
on the knowledge account of assertion, violates the knowledge account of
assertion whenever she asserts contextualism in the context in which it is most
natural to do so: that of a philosophical debate.2 This essay is structured as
1
Keith DeRose, “Assertion, Knowledge, and Context,” Philosophical Review 111,
no. 2 (2002): 167–205.
2
In this essay, “contextualism” will refer to DeRose-style contextualism as de-
veloped in Keith DeRose, “Contextualism and Knowledge Attributions,” Philosophy
and Phenomenological Research 52, no. 4 (1992): 913–29; “Solving the Skeptical Prob-
lem,” Philosophical Review 104, no. 1 (1995): 1–52; “Contextualism: An Explanation
and Defense,” in The Blackwell Guide to Epistemology, ed. John Greco and Ernest Sosa
(Malden, MA: Blackwell, 1999), 187–205; “Now You Know It Now You Don’t,” in
Proceedings of the Twentieth World Congress of Philosophy Vol. V, Epistemology (Bowling
© Martijn Blaauw and Jeroen de Ridder, Quaestiones Disputatae, Vol. 8, No. 2 (Spring 2018)
martiJn Blaauw and Jeroen de ridder 69
follows. In §§2–3, we clarify contextualism, the knowledge account of asser-

tion, and the relation between them. In §4, we present the argument against
the assertability of contextualism in the context of philosophy. In §§5–6,
we consider four objections to our argument and find them wanting. In our
argument against the assertability of contextualism in the context of philoso-
phy, the notion of sensitivity—central to DeRose’s contextualism—plays a
crucial role. We conclude in §7, however, by considering a modified version
of our argument that is phrased not in terms of sensitivity but in terms of
another key DeRosean notion: strength of epistemic position.
2. Contextualism
What is the relation between knowledge and context? According to contextu-

alism, “the truth-conditions of knowledge-ascribing and knowledge-denying
sentences . . . fluctuate in certain ways according to the context in which they
are uttered.”3 So on this view, a particular sentence of the form “S knows that
p” can be true when uttered or entertained in a nonskeptical context but false
when uttered or entertained in a skeptical context, even if the values of S and
p are kept constant across these contexts. What is crucial is that this variation
in truth-conditions of knowledge sentences depends on the use of the verb
“to know” in that knowledge sentence. Thus the main thesis of contextual-
ism can be stated as follows:
C: Due to a characteristic of “knows,” whether “S knows that p” is

true depends on features of the conversational context in which “S
knows that p” is uttered.4
C is the denial of what has long been the standard approach to the relation
between knowledge and features of the conversational context: invariant-
ism.5 According to this theory, features of the conversational context have
Green, OH: Philosophy Documentation Center), 91–106; “Assertion,” 167–205; The

Case for Contextualism: Knowledge, Skepticism, and Context, Vol. 1 (New York: Oxford
University Press, 2009).
3
DeRose, “Assertion,” 168.
4
Cf. also DeRose, Case for Contextualism, ch. 1.
5
For a formulation of invariantism, see Peter Unger, Ignorance: A Case for Scep-
ticism (Oxford: Oxford University Press, 1975). For versions of what is known as
“subject sensitive invariantism,” see John Hawthorne, Knowledge and Lotteries (Oxford:
Oxford University Press, 2004); Jason Stanley, Knowledge and Practical Interests (New
York: Oxford University Press, 2005); and Jeremy Fantl and Matthew McGrath,
Knowledge in an Uncertain World (Oxford: Oxford University Press, 2009). Jonathan
70 the unaSSertaBility of contextualiSm
no influence whatsoever on whether a subject can be truthfully attributed

knowledge. Now C raises many questions, two of the most pressing of which
are what shifts with changes in the conversational context and what brings about
a shift in conversational context.
Starting with the first question, DeRose has argued that what shifts with
changes in the conversational context is the strength of epistemic position that
is required for knowledge.6 Here strength of epistemic position is defined
in terms of the tracking condition for knowledge, where the strength of a
subject S’s epistemic position depends on how far out into logical space S
can track the truth.7 The farther out into logical space S can track the truth,
the stronger her epistemic position will be. Now what shifts in accordance
with the conversational context is how far out into logical space S should be
able to track the truth in order for her to count as knowing that p. In contexts
with low epistemic standards, for S’s belief that p to count as knowledge, a
fairly weak epistemic position would suffice: S only has to track the truth in
those worlds closest to the actual world in which p is true (S would continue
to believe p in worlds in which p is true). In contexts with high standards,
however, for S’s belief that p to count as knowledge, a strong epistemic posi-
tion is necessary: S has to track the truth far out into logical space, including
worlds in which p is false. Thus in high-standard contexts, one’s strength of
epistemic position with respect to p must be such that one’s belief that p is
sensitive, where a belief is sensitive just in case the following conditional holds:
SC: If p were not the case, S would not have believed that p.
As to the second question, DeRose captures the mechanism that brings about
a shift in the strength of epistemic position required for S to know p in what
we call the “Rule of Sensitivity”:
Schaffer’s contrastivism is a version of invariantism, albeit one where contextual

features do play a central role. See his “Contrastive Knowledge,” in Oxford Studies in
Epistemology, Vol. I, ed. Tamar Szabó Gendler and John Hawthorne (Oxford: Oxford
University Press, 2005), 235–71; “What Shifts? Thresholds, Standards, or Alterna-
tives?,” in Contextualism in Philosophy, ed. Gerhard Preyer and Georg Peter (Oxford:
Oxford University Press, 2005), 115–30; and “Knowledge in the Image of Asser-
tion,” Philosophical Issues 18 (2008): 1–19.
6
DeRose, “Skeptical Problem.”
7
The locus classicus for the notion of tracking is Robert Nozick, Philosophical Ex-
planations (Cambridge: Harvard University Press, 1981). Sherrilyn Roush, Tracking
Truth: Knowledge, Evidence, and Science (New York: Oxford University Press, 2005), of-
fers an updated treatment.
RS: When it is asserted that some subject S knows (or does not know)
some proposition P, the standards for knowledge (the standards for
how good an epistemic position one must be in to count as knowing)
tend to be raised, if need be, to such a level as to require S’s belief in
that particular P to be sensitive for it to count as knowledge.8
Attributing knowledge to a subject and denying a subject knowledge will

thus tend to raise the standards for knowledge so as to require the belief
in the target proposition to be sensitive to count as knowledge.9 Given RS,
any particular utterance of “S knows that p” will be true if and only if (i) S
believes that p, (ii) p, and (iii) S’s epistemic position is such that S’s belief that
p is sensitive.
Of course, many details are left unspecified in this characterization of
contextualism. But for assembling our argument for the unassertability
of contextualism in the context of philosophy (§4), this is all we need.
3. The Knowledge Account of Assertion
What is the relation between assertion and knowledge? According to the

knowledge account of assertion, knowledge plays a normative role in gov-
erning what one may assert.10 Specifically, the knowledge account of asser-
tion poses the following norm:
KAA: One ought to assert that p only if one knows that p.11
8
DeRose, “Skeptical Problem,” 36; Case for Contextualism, 14.
9
It remains moot what will tend to lower the standards again. Duncan Pritchard,
“Contextualism, Scepticism, and the Problem of Epistemic Descent,” Dialectica 55,
no. 4 (2002): 327–49, contains a useful discussion of this issue.
10
The knowledge account of assertion is defended in, for example, Unger,
Ignorance; Timothy Williamson, Knowledge and Its Limits (Oxford: Oxford University
Press, 2000); DeRose, “Assertion”; Hawthorne, Knowledge; John Turri, “Prompting
Challenges,” Analysis 70, no. 3 (2010): 456–62; Matthew Benton, “Two More for
the Knowledge Account of Assertion,” Analysis 71, no. 4 (2011): 684–87; Marti-
jn Blaauw, “Reinforcing the Knowledge Account of Assertion,” Analysis 72, no. 1
(2012): 105–8; Martijn Blaauw and Jeroen de Ridder, “Unsafe Assertions,” Austral-
asian Journal of Philosophy 90, no. 4 (2012): 797–801.
11
KAA is also accepted by Schaffer, “Knowledge in the Image,” where it is used
in an argument in favor of contrastivism. It is, however, by no means without critics.
See, for instance, Matthew Weiner, “Must We Know What We Say?,” Philosophical
Review 114, no. 2 (2005): 227–51; Igor Douven, “Assertion, Knowledge, and Rational
Credibility,” Philosophical Review 115, no. 4 (2006): 449–85; Jennifer Lackey, “Norms
As with most norms, there is a distinction to be made between primary and

secondary propriety. Acknowledging this distinction, DeRose writes, “As
happens with other rules, a kind of secondary propriety/impropriety will
arise with respect to [the knowledge account of assertion]. While those who
assert appropriately (with respect to this rule) in a primary sense will be those
who actually obey it, a speaker who broke this rule in a blameless fashion
(one who asserted something she didn’t know, but reasonably thought she
did know) would in some secondary sense be asserting properly.”12 Accord-
ing to DeRose, then, even if you did not comply with KAA because what
you asserted was not known by you, you could still have complied with KAA
in a weaker sense if you reasonably believed that you knew the proposition
you asserted. Subjects who assert falsehoods that they reasonably believe to
be true can behave appropriately in one sense and inappropriately in another.
DeRose accepts KAA but also argues that what epistemic position we
must be in so that we can be warranted in asserting that p is a contextual
matter. In some contexts, one must be in a very good epistemic position to
be warranted in asserting that p, whereas in other contexts, a lesser epistemic
standing suffices. So even though the norm for assertion is knowledge, there
is a contextual variation in when a subject S is warranted in asserting some-
thing. This leads DeRose to accept the following thesis:
CV: What epistemic position one must be in so that one can be war-
ranted in asserting that p is a context-variable matter.
DeRose now continues to argue that CV and KAA imply C. If the norm for
warranted assertion is knowledge and if our ordinary practices display a flexi-
bility with respect to what epistemic position one ought to be in so that one
can warrantedly assert a proposition, then there must be a flexibility in what
knowledge is as well. Here is DeRose himself on this point: “The knowl-
edge account of assertion provides a powerful argument for contextualism:
If the standards for when one is in a position to warrantedly assert that P
are the same as those that comprise a truth-condition for ‘I know that
P,’ then if the former vary with context, so do the latter. In short: The knowl-
edge account of assertion together with the context-sensitivity of assertabil-
ity yields contextualism about knowledge.”13 And so KAA combined with
of Assertion,” Noûs 41, no. 4 (2007): 594–626; Sanford C. Goldberg, Assertion: On the
Philosophical Significance of Assertoric Speech (New York: Oxford University Press, 2015).
12
13
Ibid., 187.
CV—that is, the recognition that there is a contextual variation in when it is

appropriate to assert something—support contextualism about knowledge.14
A complication here is that, given CV, DeRose cannot accept KAA as
it stands. For DeRose, “knows” has many different senses and does not
denote one property in all contexts. What it takes for “S knows that p” to be
true depends on the conversational context of the attributor of knowledge.
Accordingly, DeRose must hold that KAA actually expresses different norms
in different contexts. Here is how DeRose contextualizes the knowledge rule:
“To be positioned to assert that p, one must know that p according to the
standards for knowledge that are in place as one makes one’s assertion.”15
And so we arrive at the knowledge account of assertion contextualized in
DeRose’s own formulation (2009, 99):
KAAC: A speaker, S, is well-enough positioned with respect to p to be

able to properly assert that p if and only if S knows that p according to
the standards for knowledge that are in place as S makes her assertion.16
This, in a nutshell, is DeRose’s argument for contextualism. How could

anyone rest a “madly swaying distinction [i.e., the conditions for warranted
assertability, MB & JdR] upon a stubbornly fixed foundation [i.e., the truth-
conditions for ‘S knows that p,’ which invariantism holds to be fixed, MB &
JdR]?”17
Again, many details are left unspecified in this characterization of KAA.
But for assembling the argument for the unassertability of contextualism in
the context of philosophy, this is all we need. It is to this argument that we
now turn.
4. The Unassertability of Contextualism
The argument for the unassertability of contextualism in the context of phi-

losophy runs as follows:
14
Adam Leite, “How to Link Assertion and Knowledge without Going Con-
textualist: A Reply to DeRose’s ‘Assertion, Knowledge, and Context,’” Philosophical
Studies 134, no. 2 (2007): 111–29, contains a useful discussion of DeRose’s argument
from KAA to contextualism.
15
16
DeRose, Case for Contextualism, 99. See Schaffer, “Knowledge in the Image,”
for a similar rule geared to fit his contrastivist treatment of the norm of assertion.
17
DeRose, Case for Contextualism, 101.
(1) Contextualism (C)

(2) Knowledge account of assertion contextualized (KAAC)
(3) A subject S should assert contextualism in the context of philoso-
phy only if “I know that contextualism is true” is true for S in that
context. [from 2]
(4) In the context of philosophy, the standards for knowledge require
S’s belief that contextualism is true to be sensitive for it to count as
knowledge.
(5) S’s belief that contextualism is true is not sensitive.
(6) In the context of philosophy, “I know that contextualism is true” is
false for S. [from 4 and 5, modus tollens]
(7) Hence S ought not to assert contextualism in the context of phi-
losophy. [from 3 and 6]
The argument is valid. (1) and (2) are just the assumptions that contextualism
and the knowledge account of assertion contextualized are true, and (3) is
the application of KAAC to our particular assertion. Hence DeRose should
find (1) through (3) uncontroversial. Only (4) and (5) stand in need of further
support.
Starting with premise (4), it maintains that the standards for knowledge
in a philosophical discussion are at least as demanding as when the word
“know” is uttered, thus requiring belief in a philosophical position to be
sensitive for it to count as knowledge. But why would that be? We offer two
motivations.
First, it is predicted by the nature of philosophical inquiry. Philosophy
is an activity that is characterized by (among other things) precision, rigorous
argumentation, and the continuous search for new error-possibilities, including remote ones.
Like other academic disciplines, it is a systematic search for knowledge with
the highest attainable degree of justification by means of the best methods
available to us. The level of scrutiny is immensely high in the context of
philosophy. Not so high that it is required that the philosopher eliminate all
skeptical scenarios in order to know her philosophical position—indeed, this
isn’t a requirement for scientific knowledge in general. But high enough that
it is required for knowledge that the philosopher be able to eliminate error-
possibilities for her philosophical position and thus track the truth in a broad
range of possible worlds, including some in which her philosophical position
is false. What’s more, in epistemology in particular, the nature of knowledge is
investigated. So arguably the word “knowledge” lingers over any epistemo-
logical debate. That brings to mind RS, albeit implicitly.
Second, contextualists have argued that the truth conditions of knowl-
edge attributions are affected not only by RS but also by the stakes (the costs
of being wrong). In the well-known Bank Cases, for instance, the subject is
correctly attributed knowledge in the first Bank Case but is correctly denied
knowledge (of the very same proposition) in the second Bank Case only
because the stakes have changed from low to high.18 We maintain that the
stakes are indeed high in the context of philosophy. Not in the sense that
the costs of being wrong are bankruptcy, divorce, or suicide, but in the sense
that the costs of being wrong are (partially) incorrect views about issues of
fundamental importance: the fundamental nature of reality and ourselves.
The reason we do philosophy is to understand ourselves and the world.
Clearly this matters a lot to humans, or we wouldn’t have stayed at it for
millennia. So we conclude that the standards for knowledge in a philosoph-
ical discussion are high or at least that they are as high as in contexts where
the word “know” is explicitly uttered.19 In philosophy, our beliefs need to be
sensitive to be instances of knowledge.
The question then is the following: Is S’s belief that contextualism is
true sensitive? In defense of premise (5), we offer two motivations. First,
philosophy is extremely difficult. Many questions and issues that philoso-
phy deals with are close to the limits of what our cognitive capacities have
a grip on. The history of philosophy shows this. Most substantive positive
philosophical theses have both historical and contemporary proponents and
opponents of formidable intelligence, who offer powerful arguments for
and against those theses. A very natural explanation for this is that we aren’t
very receptive to philosophical facts. They are elusive. Nothing guarantees
that our philosophical beliefs generally track the truth to any significant
degree, like, for instance, our perceptual beliefs. Because of this, the contex-
tualist’s belief is very unlikely to be sensitive.
Second, consider the notion of an epistemic peer that has become famil-
iar from discussions about the epistemic significance of disagreement.20
18
The Bank Cases were introduced in DeRose, “Contextualism and Knowledge
Attributions.”
19
Arguably, the fact that philosophy means “love of wisdom” already puts seri-
ous philosophical discussions under the spell of that most demanding of epistemic
states.
20
See, for instance, Thomas Kelly, “The Epistemic Significance of Disagree-
ment,” in Oxford Studies in Epistemology, Vol. I, ed. Tamar Szabó Gendler and John
Hawthorne (Oxford: Oxford University Press, 2005), 167–96; David Christensen,
“Epistemology of Disagreement: The Good News,” Philosophical Review 116, no. 2
(2007): 187–217; Richard Feldman, “Reasonable Religious Disagreement,” in Philos-
ophers without Gods, ed. Louise Antony (New York: Oxford University Press, 2007),
194–214; Richard Feldman and Ted Warfield, eds., Disagreement (New York: Oxford
University Press, 2010); and David Christensen and Jennifer Lackey, eds., The Epis-
temology of Disagreement: New Essays (New York: Oxford University Press, 2013), for
influential contributions to this debate.
The notion of an epistemic peer can be characterized by the following two

principles:21
Evidential equality: A and B are evidential equals relative to the question

whether p when A and B are equally familiar with the evidence and
arguments that bear on the question whether p.
Cognitive equality: A and B are cognitive equals relative to the question
whether p when A and B are equally competent, intelligent, and fair-
minded in their abilities to assess the evidence and arguments that
bear on the question whether p.
With respect to the contextualism/invariantism debate, some philosophers

defend (forms of) contextualism, while others defend (forms of) invariant-
ism. What is crucial to our defense of (5) is that the philosophers engaged
in this debate—or at least some of them on either side—are epistemic peers
and ought to recognize each other as such. They share the same evidence and
they are cognitive equals.
Now if one evaluates sensitivity’s counterfactual condition, one consid-
ers the nearest possible world in which p is false—the world in which p is false
but that is otherwise as similar to the actual world as possible—and then tries
to determine whether, in that world, S believes that p. If S does believe that p
in that world, then her belief that p is insensitive. If S does not believe that
p in that world, her belief is sensitive.
How does the contextualist’s belief that contextualism is true fare with
respect to this evaluation? Suppose, for the sake of argument, that this belief
is indeed true in the actual world. Now consider the nearest possible world
in which contextualism is false but that is otherwise as similar to the actual
world as possible—call this world w. We will now have to try to determine
whether the contextualist believes that contextualism is true in w.
Given the fact that some of the contextualist’s epistemic peers do not
believe that contextualism is true in the actual world in which contextual-
ism is indeed true, it seems plausible that the contextualist wouldn’t believe
that contextualism is false in w. The situation is entirely symmetrical. So if
invariantists are liable to make the mistake of believing their position to be
true in a world in which it is false, then contextualists will be just as liable
to make the same mistake with respect to their position. The reason for this
is that the actual world and w will be highly similar from an evidential point
of view. After all, a number of the contextualist’s peers take the evidence
in the actual world to support invariantism. They are aware of contextual-
ism and the arguments in its support, but they have objections to them and
21
See Kelly, “Epistemic Significance,” 174–75.
offer their own rival arguments in favor of invariantism. A world in which

contextualism is true can thus look very similar to one in which invariantism
is true, so the nearest world in which contextualism is false, w, will certainly
look very similar to one in which it is true. Hence it is easy, even for highly
trained and capable epistemologists, to mistake a world in which the evidence
supports contextualism for one in which it supports invariantism. But then
something analogous will be true for the nearest world in which contextual-
ism is false—that is, world w. In that world, the contextualist is likely to make
the mistake of taking the evidence to support contextualism when it in fact
supports invariantism. The evidence for and against philosophical positions
is extremely subtle. Why would the contextualist have special access to the
philosophical facts and be responsive to these subtle differences in the avail-
able evidence in ways that her epistemic peers are not? We conclude that the
contextualist’s belief that contextualism is true is not a sensitive belief.22
With premises (4) and (5) thus supported, the argument looks solid.23
5. Replies Rejected
How might the contextualist reply? We initially anticipate the following

three objections, the first of which is to say that even though it would be
inappropriate in the primary sense to assert contextualism, it could still
be appropriate in a secondary sense to do so. We would respond, first and
foremost, that even if asserting contextualism could be appropriate in a sec-
ondary sense, this only eases the pain without taking it away. Strictly speaking,
we are still not allowed to assert contextualism in philosophical contexts. But
more important, we submit that asserting contextualism is inappropriate in
both senses. It is inappropriate in the primary sense because the act of assert-
ing contextualism while contextualism isn’t known to be true violates the
knowledge account of assertion; it is inappropriate in the secondary sense
because it seems unlikely that DeRose can reasonably believe that he knows
22
See Sanford C. Goldberg, “Reliabilism in Philosophy,” Philosophical Studies 142,
no. 1 (2009): 105–17, for similar arguments to the effect that our belief-forming
methods in philosophical matters are not reliable.
23
After having written this essay, it came to our attention that Christoph Jäger,
“Contextualism and the Knowledge Norm of Assertion,” Analysis 72, no. 3 (2012):
491–98, presents a different argument for a similar conclusion, to wit that the posi-
tion that combines contextualism and KAA (as DeRose’s contextualism does) can-
not coherently be stated. His argument, however, crucially relies on two principles
that we do not assume: one is a contextualized principle about the factivity of knowl-
edge, the other an epistemic closure principle. Hence his and our arguments are
complementary.
that contextualism is true in light of the many problems that his epistemic
peers have identified for contextualism.
The second reply the contextualist might give is an indifferent shrug: the
problem we have mounted against contextualism is a problem for any philo-
sophical view. Given KAA, no philosophical view can ever be asserted in the
context of a philosophical discussion. What’s so special about contextualism?
We grant that contextualism isn’t special in being the only philosophical posi-
tion that is vulnerable to the objection we have introduced in this essay. But
it is special in that it is allegedly powerfully motivated by KAA. If our objection
is correct, contextualism is powerfully motivated by an account of assertion
that at the same time forbids it from being asserted in that context in which
it most naturally and relevantly should be asserted.
The third objection is to deny that our argument establishes any signifi-
cant result because no philosopher ever wants to assert her philosophical
position in the context of a philosophical discussion. Instead, philosophers
merely hypothesize their favored positions to be true, hold them up for
consideration, tentatively entertain them, provisionally accept them, or take
some other noncommittal cognitive attitude toward them.24
We reply that this construal of philosophy is too deflationary. It is at
odds with how philosophers conceive of what they are doing as expressed
through their typical behavior in speech and writing. They write papers to
argue for philosophical theses, and some of them even develop book-length
arguments to defend their favored positions. They sometimes develop a
strong commitment to a philosophical claim and get excited when they see
a novel argument for that claim or irritated when they spot a problem for
it. Some of them spend entire careers, or substantial parts of it, developing
and defending a view. These are clear indications that philosophers do take
committal cognitive attitudes toward those philosophical theses they take to
be true and that their utterances of them are most naturally understood as
assertions.25 The characteristic phraseology one finds in philosophical papers
and books provides further evidence. Philosophers defend positions, establish
claims, argue that so-and-so is the best solution to a philosophical problem,
give compelling arguments, offer strong reasons, and conclude that such-and-such
is correct, the best explanation, a superior account, and so on. Indeed, consider
what DeRose himself says about contextualism: it is “the correct solution
24
Goldberg, Assertion, thinks that these are indeed the typical attitudes philos-
ophers take toward their views. Since he also maintains that philosophers do assert
their views, he rejects KAA and proposes replacing it by a context-sensitive norm.
25
Of course, this is not to deny that philosophers also frequently take various
noncommittal cognitive attitudes to certain philosophical theses.
to the puzzle [the Argument from Ignorance] confronts us with”;26 it “can

finally solve this perennially thorny philosophical problem.”27 Other candi-
date solutions to the skeptical puzzle result from “a failure to see the truth
of contextualism”;28 the contextualist solution is “the best resolution of our
puzzling conflict of intuitions.”29 To interpret all of these and similar locu-
tions as something weaker than assertions of a philosophical thesis surely
goes against the plain sense of words. So we conclude that the third reply is
unconvincing as well. Philosophers, or at least some of them, do assert philo-
sophical positions. If a position cannot coherently be asserted in the context
of philosophy, that is a significant blow against it.30
6. One More Reply Rejected
A fourth and final objection is to deny premise (5) and argue that a contextual-
ist’s belief in the truth of contextualism is sensitive, since contextualism—like
many other philosophical theses—is, if true, necessarily true. Beliefs in neces-
sary truths trivially satisfy SC because its antecedent is never actualized. If p is
a necessary truth, then there are no possible worlds in which p is false. Hence
the sensitivity condition (“If p were not the case, S would not have believed
p”) is always satisfied. We have several lines of response to this objection.
The first and most important thing to note is that if belief in necessary
truths is indeed trivially sensitive, then this is an embarrassment for SC: it
means that knowledge of necessary truths comes too easy.31 Hence there
26
DeRose, “Skeptical Problem,” 2.
27
Ibid., 3.
28
Ibid., 41.
29
Ibid., 49.
30
Logan Paul Gage called our attention to the fact that DeRose himself explores
the issue of assertions in philosophical contexts in his new book The Appearance of
Ignorance: Knowledge, Skepticism, and Context, Volume 2 (New York: Oxford University
Press, 2018), briefly in chapter 4 and more fully in appendix C. Although he doesn’t
fully commit, he appears to favor a view on which (a) there isn’t much knowledge
to be had in philosophy, (b) we nonetheless make assertions, but (c) operate under a
“pretense of knowledge” when doing so. Sometimes, however, we need to “get real”
and lift the pretense, for instance in response to challenges questioning whether a
speaker really knows what she asserts. A fuller evaluation of this proposal must await
another occasion, but for now it suffices to note that—pretense aside—our conclu-
sion stands: asserting contextualism in philosophy violates the norm for assertion
that is supposed to be an important motivation for the view.
31
This problem led Nozick himself to abandon the sensitivity requirement for
knowledge of necessary truths. See Ernest Sosa, “How to Defeat Opposition to
is initial reason to doubt that this objection has much force. Defending
the assertability of contextualism in philosophical contexts by exploiting a
problematic feature of the sensitivity condition doesn’t seem like an attrac-
tive dialectical strategy.32 But suppose, for the sake of argument, that some
suitable modification of sensitivity can be devised on which belief in neces-
sary truths comes out as sensitive but not trivially so.33 Is it indeed plausible
to think of contextualism as a necessary truth?
To the extent that contextualism is a linguistic thesis about the word “know”
and its cognates, its truth cannot be necessary. Facts about how words are
used in a language are highly contingent due to the conventional nature of
language. Just as the word “know” could have meant something entirely
different, it also seems possible that the truth conditions for knowledge-
attributing sentences would not have shifted with context.
Alternatively, contextualism can be construed as a thesis about the concept
of knowledge. This is what DeRose himself alludes to when he empha-
sizes that contextualism is not just a piece of philosophy of language but
has profound importance for epistemology34 and explicates contextualism as
claiming that “knowledge” has many different senses.35 On such a construal,
the contextualist could maintain with some plausibility that contextualism
expresses a conceptual and hence necessary truth about the concept of
knowledge.
We don’t believe, however, that this suggestion has much going for it.
First, as we have noted before, the current epistemological scene features
knowledgeable and intelligent philosophers who deny that contextualism
is true. They defend invariantism of various sorts. This state of affairs is
a strong indication that, given what we know about the world and about
knowledge attributions specifically, contextualism is not the only possibility.
It looks as if invariantism represents a possible way knowledge attribu-
tions could have worked. Given that this is the way things look and absent
independent reasons to deny it, we have good reason to suppose that there
are possible worlds in which the concept of knowledge behaves as invarian-
tists say.
Moore,” Philosophical Perspectives 13 (1999): 141–53, 146.

32
DeRose is aware of this problem for Sensitivity and appeals to the notion of
“strength of epistemic position” to handle knowledge of necessary truths. See his
“Sosa, Safety, Sensitivity, and Skeptical Hypotheses,” in Sosa and His Critics, ed. John
Greco (Oxford: Blackwell, 2004), 22–41. We return to this in the next section.
33
Cf. Roush, Tracking Truth, 134–47, for a sophisticated attempt.
34
DeRose, “Contextualism: Explanation and Defense,” 188–89.
35
Ibid., 191–92.
Second, whether the truth conditions for knowledge attributions shift

with conversational contexts would seem to depend on what the world is
like naturally and socially. There is a relationship between the way the world
is and the way the concept of knowledge and the word “knowledge” func-
tion. Therefore, the truth of contextualism cannot be necessary. In possible
worlds with relevantly different physical or social characteristics, contextual-
ism could be false and invariantism true.36 To see this, consider the following
examples of possible worlds in which contextualism appears to be false.37
Consider a possible world in which we have vastly superior cognitive
abilities—we’re as close to being omniscient as is possible for beings who are
spatiotemporally limited. As a result, we know the things we know with high
degrees of confidence and tend to have maximal justification for what we
know. We hardly make mistakes and believe virtually no falsehoods. In such
epistemically optimal conditions, it seems implausible that the truth condi-
tions for knowledge ascriptions would shift with conversational context. After
all, no matter what the conversational context, we always satisfy demanding
epistemic standards. There is no reason why the truth conditions for knowl-
edge sentences would be sensitive to context.
As a second example, imagine a possible world physically very similar
to ours but with different sociolinguistic conventions. There may be worlds
in which people are extremely risk-averse in epistemic matters. They only
ascribe knowledge when people have indubitable beliefs of the kind Descartes
thought was required for knowledge. In these worlds, even in low-standards
contexts where one is in a relatively strong epistemic position with regard to
a proposition, linguistic conventions dictate that one not ascribe knowledge
to oneself. There may also be worlds where the conventions vis-à-vis “know”
license what seem to us extremely liberal and unwarranted knowledge ascrip-
tions. Minimally justified beliefs all get to be called knowledge, no matter
what the conversational context. People in these worlds have no problem
at all saying that we can know the denials of skeptical hypotheses, even in
a philosophical discussion. Being raised on their lax linguistic conventions,
they don’t share our intuitions with regard to the appropriateness of the use
36
See Stephen Hetherington, “Is This a World Where Knowledge Has to In-
clude Justification?,” Philosophy and Phenomenological Research 75, no. 1 (2007): 41–69,
for an analogous argument to the effect that knowledge doesn’t necessarily include
justification.
37
Note, however, that nothing in particular hangs on the details of these ex-
amples; they only serve to make plausible the general claim that the truth of contex-
tualism (or other claims about knowledge attributions) depends on the contingencies
of the natural and social world.
of “know.” In such worlds, contextualism would be false. Since such worlds

seem to be genuinely possible, contextualism isn’t necessarily true.38
As a final resort, the contextualist might propose that contextualism
is not a necessary truth of the conceptual variety but an a posteriori neces-
sary truth of the sort Kripke familiarized us with. She might propose that
“knowledge” is a rigid designator for the concept of knowledge that figures
in contextualism. Just as it is necessary that water is H2O, it is necessary
that knowledge is what contextualism says it is, or so this proposal main-
tains. Although there may be possible worlds in which people use “knowl-
edge” to refer to an invariant relation, these worlds aren’t worlds in which
contextualism is false. Rather, they are worlds in which something other than
the contextualist’s notion of knowledge occupies the role played by knowl-
edge in our world (just as XYZ occupies the role of watery stuff in possible
worlds in which watery stuff is not H2O). We could call this knowledge-y
stuff schmowledge. According to this proposal, then, the reason that contextu-
alism is necessarily true is that in possible worlds where schmowledge occu-
pies the knowledge-role and is the referent of “knowledge,” knowledge is still
what the contextualist says it is in the actual world, even though schmowledge
is not. Hence there are no possible worlds in which knowledge and knowl-
edge attributions do not behave as contextualism has it.
In response to this proposal, we reply, first, that, to the best of our
knowledge, contextualists have neither claimed nor defended that “knowl-
edge” is a rigid designator. So it would at least be a surprising result that
contextualism requires this to be so. Moreover, the view that knowledge is a
natural kind is not a popular view.39 Because rigid designators typically latch
onto natural kinds, this speaks against “knowledge” being a rigid designator.
Furthermore, the core idea of contextualism—that “know” has many differ-
ent senses—is at odds with “knowledge” being a rigid designator. A rigid
designator serves to pick out one and the same thing in all possible worlds
in which that thing exists. That is something a word with many different
senses cannot do. Suppose “water” has many different senses that shift with
context. In one context it means H2O; in another it might mean hydrogen
peroxide, H2O2; and in yet another heavy water, D2O. Such facts would seem
38
It is no objection to point out that in the worlds described, the contextualist
would not believe contextualism. The point of these examples was not to estab-
lish that there are nearby worlds in which contextualism is false but still believed to
be true by contextualists (for this, see section 3), but to show that the truth of con-
textualism is not necessary.
39
A notable dissenter is Hilary Kornblith, Knowledge and Its Place in Nature (Ox-
ford: Oxford University Press, 2003).
to preclude “water” from being a rigid designator, precisely because it no

longer picks out one and the same thing across possible worlds.40
Fortunately, however, it doesn’t matter where we come down on this
issue because even if “knowledge” were a rigid designator and the truth
of contextualism a posteriori necessary, a version of our argument still goes
through. To see this, we must consider how contextualism could be a neces-
sary a posteriori truth.41 If “knowledge” were a rigid designator, a sentence
describing the central contextualist claim, such as “The truth conditions
for knowledge attributions are context-sensitive,” would express a neces-
sary truth because “knowledge” picks out our actual notion of knowledge in
every possible world in which this notion exists. But now note that this is just
one of two possible ways to think about how the term “knowledge” applies to
the world. The first, and presumably most natural, way to think about it is
to take our world to be the actual world. If we do so, and we assume for the
sake of argument that contextualism is true in our world, the result is what
we just described: “knowledge” applies to the contextualist notion of knowl-
edge that occupies the role of knowledge in our world because this contextu-
alist notion is “the knowledge-y stuff of our acquaintance.” Another way to
think about it, however, is to consider what the word “knowledge” would apply
to under the hypothesis that another world than our world is actual. Had a world where
schmowledge occupies the knowledge-role been actual, the term “knowledge”
would have applied to the invariantist schmowledge because in such a world,
schmowledge would have been the knowledge-y stuff of our acquaintance,
which we would have baptized “knowledge.”
This shows that there is an important sense in which the truth of
contextualism is not necessary, even if “knowledge” is supposed to be a rigid
designator. If we go by the second way of thinking about the intension of
“knowledge,” the truth of contextualism is contingent. If another possible
world than our world had been actual, “knowledge” could easily have applied
to an invariantist notion of knowledge. Hence on this understanding, the
sentence “The truth conditions for knowledge attributions are context-
sensitive” may well have been false and is therefore not a necessary truth.
40
Alvin Goldman, “Philosophical Intuitions: Their Target, Their Source, and
Their Epistemic Status,” Grazer Philosophische Studien 74, no. 1 (2007): 1–26, 17, ex-
presses a similar worry.
41
What follows is inspired by the two-dimensional modal logic treatment of
necessary a posteriori truths found in David Chalmers, The Conscious Mind: In Search of
a Fundamental Theory (New York: Oxford University Press, 1996), 56–65; and Frank
Jackson, From Metaphysics to Ethics: A Defence of Conceptual Analysis (Oxford: Clarendon
Press, 1998), ch. 3.
The upshot of the previous discussion is that the truth of contextualism is

not a necessary truth. Hence the fourth objection fails too. Contextualism
cannot plausibly be construed as a conceptual truth, nor is it plausible to
think of it as an a posteriori necessary truth in the Kripkean sense. And even
if its truth were a posteriori necessary, it would still be contingent in another
important sense.
7. A Modified Argument: Sensitivity vs.

Strength of Epistemic Position
In replying to the fourth objection, we started out by noting that necessary

truths are a problem for SC because belief in a necessary truth trivially sat-
isfies it. The contextualist would do well, therefore, not to insist on SC for
the case of necessary truths. In this final section, we consider what happens
if the contextualist adopts a different criterion for appropriate knowledge
ascriptions in high-standards contexts. Since our argument in its present for-
mulation hinges on the contextualist’s belief being insensitive, this is a press-
ing matter.
In fact, DeRose himself proposes a further criterion, next to sensitivity.
In a paper in which he compares and contrasts his sensitivity condition with
Ernest Sosa’s safety condition, he emphasizes that, for him, it is ultimately
strength of epistemic position—and not sensitivity—that matters for knowledge,
although a failure to satisfy Sensitivity does often provide a correct explana-
tion of why people lack knowledge.42 Strength of epistemic position is to
be understood in terms of one’s beliefs matching the facts of the matter or
tracking the truth in the relevant sphere of (nearby) possible worlds.43 This
is importantly different from sensitivity. Whereas sensitivity only concerns
one’s belief about p in possible worlds where p is false, strength also takes into
account one’s belief in worlds in which p is true. Unlike the sensitivity of one’s
belief in a certain truth, then, the strength of one’s epistemic position with
regard to a proposition can be upset by the presence of nearby possible
worlds in which that proposition is true but one fails to believe it.
What this means is that our whole argument can also be put in terms of
strength of epistemic position instead of sensitivity. As follows:
(1) Contextualism (C)

(2) Knowledge Account of Assertion Contextualized (KAAC)
42
DeRose, “Sosa, Safety.”
43
DeRose, “Skeptical Problem,” 34; “Sosa, Safety,” 33–35.
(3) A subject S should assert contextualism in the context of a philo-

sophical discussion only if “I know that contextualism is true” is
true for S in that context. [from 2]
(4′) In the context of a philosophical discussion, in order for S’s belief
that contextualism is true to count as knowledge, S needs a strong
epistemic position.
(5′) S isn’t in a strong epistemic position vis-á-vis the belief that contex-
tualism is true.
(6) In the context of a philosophical discussion, “I know that
contextualism is true” is false for S. [from 4 and 5, modus tollens]
(7) Hence S ought not to assert contextualism in the context of a phil-
osophical discussion. [from 3 and 6]
The defense of (4′) and (5′) would mostly run along the same lines as that of
(4) and (5) because insensitivity and strength are closely connected. Insensi-
tivity of someone’s belief detracts from the strength of that person’s epis-
temic position. So by showing someone’s belief to be insensitive, one also
shows her epistemic position to be compromised.
In defense of (5′), however, an additional consideration can be added.
If p is true in the actual world, then the strength of S’s epistemic position
with regard to p is upset by the presence of nearby worlds in which p is still
true but S fails to believe that p. This is particularly pertinent to the present
discussion. If we assume, again, that contextualism is true in the actual world,
it seems that the contextualist could have easily failed to believe that contex-
tualism is true in a nearby world in which it is true. Given that there are well-
informed and competent epistemologists who believe that invariantism is
true (and hence that contextualism is false) in the actual world, it is extremely
plausible that someone who believes that contextualism is true in the actual
world could believe that invariantism is true in a nearby world in which, say,
she received a somewhat different training in philosophy or had followed
a different research trajectory. After all, these worlds would be extremely
similar as far as the relevant evidence is concerned, perhaps even indistin-
guishable. It follows that the contextualist’s epistemic position with regard to
contextualism isn’t very strong—not strong enough to count as knowing in
the context of a philosophical discussion.
Putting the argument in terms of strength of epistemic position also
evades the fourth objection, which turned on the idea that the truth of
contextualism is necessary and that belief in it is therefore automatically
sensitive. Since SC only looks at worlds in which a proposition is false, neces-
sary truths come out as trivially sensitive because the antecedent of the crite-
rion (if p were not the case) is never actualized. No such shortcut is available
for strength of epistemic position, since nothing whatsoever guarantees
that one will automatically be in a very strong position with regard to neces-
sary truths. If one fails to believe a necessary truth in nearby worlds, one’s
epistemic position with regard to that truth will be weak. So even if our reply
to the fourth objection would turn out to be unsuccessful, our revised argu-
ment in terms of strength of epistemic position still goes through. Even if
contextualism is a necessary truth after all and the contextualist is prepared
to bite the bullet on the trivial sensitivity of belief in necessary truths, she
still wouldn’t be off the hook. Her epistemic position vis-à-vis contextualism
is too weak to count as knowing in the context of a philosophical discussion
and, therefore, in asserting contextualism she would violate KAAC.
If it has these advantages, one might wonder why we didn’t formulate
our argument in terms of strength of epistemic position in the first place.
The reason is that, in our opinion, using that notion would have introduced
unnecessary vagueness into the discussion because it invites questions
about the exact strength that is required throughout various contexts, the size
of the relevant sphere of possible worlds, and the right ordering of worlds in
that sphere. In contrast, focusing on sensitivity provided a clear-cut crite-
rion and effectively sidestepped these difficult questions. In addition, of
course, DeRose himself insists that in most cases, the insensitivity of a belief
provides the correct explanation of why that belief fails to constitute knowl-
edge. Hence showing that the contextualist’s belief that contextualism is true
is insensitive is a good argument for the conclusion that “I know that contex-
tualism is true” is false for the contextualist in a philosophical discussion,
even if it is admitted that ultimately, strength of epistemic position is the
crucial notion.
8. Envoi
Bringing the preceding points together, what moral emerges from the discus-
sion in this essay? Not that contextualism is false. Not that the knowledge
account of assertion is false. Not even that contextualism cannot be based on
the knowledge account of assertion. What emerges is that if contextualists
base their theory on the knowledge account of assertion, they can no longer
coherently assert their own theory in the most natural and appropriate con-
text: that of philosophy.44
44
Thanks to audiences at the Vrije Universiteit Amsterdam, the University of
Geneva, and the University of Aberdeen. We are particularly grateful for comments
from Peter Baumann, Jessica Brown, Igor Douven, Logan Paul Gage, Hilary Korn-
blith, Duncan Pritchard, Ram Neta, Jonathan Schaffer, and Matt Weiner.
Can Experience Fulfill the
Many Roles of Evidence?
Logan Paul Gage

Franciscan University of Steubenville
Introduction
It is still a live question in epistemology and philosophy of science as to

what exactly evidence is. In my view, evidence consists in experiences called
“seemings.” This view is a version of the phenomenal conception of evidence, the
position that evidence consists in nonfactive mental states with propositional
content.1 This conception is opposed by sense-data theorists, disjunctivists,
and those who think evidence consists in physical objects or publicly observ-
able states of affairs (what I call the courtroom conception of evidence). Thomas
Kelly has recently argued that the phenomenal conception cannot play all
the roles evidence plays and is thus inadequate.2 Having first explained the
nature of seemings, in this essay I utilize Kelly’s own understanding of the
four major roles of evidence and argue that the phenomenal conception can
play each one. Experience is a good candidate for evidence.
The Nature of Seeming States
We all have seemings (or seeming states), experiences in which something

seems to be the case. Not everything we might call an “experience” is evi-
dence. For instance, we might say that Samantha underwent the experience
of surgery even though she was anesthetized. But this sort of unconscious
“experience” would be a poor candidate for epistemic evidence. So would the
experience of the periphery of your visual field; you might be conscious of
1
I hold that the mental state or experience itself is the evidence, not merely the
propositional content of the experience or extramental facts. See especially John
Turri, “The Ontology of Epistemic Reasons,” Noûs 43, no. 3 (2009): 490–512. How-
ever, propositionalism and factualism are not necessarily at odds with the phenome-
nal conception or a commonsense, experience-first epistemology.
2
Thomas Kelly, “Evidence,” in Edward N. Zalta, ed., The Stanford Encyclopedia of
Philosophy (Winter 2016 edition).
© Logan Paul Gage, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)
88 can experience fulfill the many roleS of evidence
its contents but not explicitly aware of them. Seemings are, then, a subset of
experiences taken in this broad sense—a subset in which one is aware of the
experience’s contents.
Regarding this content, William Tolhurst argues that seemings are inten-
tional states—states that are about something.3 They are not mere noninten-
tional states like an afterimage from a camera flash. Hence it is natural to
see them as conscious experiences with propositional content (or content
about the world). The content of seemings aims at having a “world-to-mind
direction of fit.”4 Seemings aim at capturing the way the world really is. A
desire might have propositional content but not aim at capturing the truth
about the world. Seemings have a distinctive phenomenological character.
Tolhurst describes it as “felt veridicality”; they “have the feel of truth, the
feel of a state whose content reveals how things really are.”5 Tucker writes
of seemings’ “assertiveness,” while Huemer refers to the “forcefulness” of
seemings that “represent their contents as actualized.”6 Seemings, then, differ
from other mental states that do not recommend themselves as representing
the way that the world really is. The seeming that there is a person in front of
me is distinguished from merely imagining that there is a person in front
of me, not by the propositional content but by this distinctive phenomenology.
It is crucial to distinguish seemings not only from nonconscious states
but also from mere sensory states or sensations. Seemings are “thick” expe-
riences; they involve not just seeing but a construal or seeing-as. Think of the
buzz of a bee or a solid green visual impression. Typically, when we have
sensory impressions, they are accompanied by seeming states (the seeming
that there is a bee nearby or that there is a green object before me), but they
need not be. They are conceptually distinct. Think of a baby receiving visual
sensations for the first time. Without concepts under which those sensations
might be organized, classified, and made intelligible, it is difficult to see how
anything in particular would seem to the baby to be the case. She sees colors
but does not see them as a bird. It is reasonable to think that for the baby
these are sensations without seemings, and hence the two are distinct.7
We should not, of course, posit distinct mental states like seemings with-
out need. Perhaps for this reason many have thought that seemings might
3
Tolhurst, “Seemings,” American Philosophical Quarterly 35, no. 3 (1998): 293.
4
John Searle, Intentionality (New York: Cambridge University Press, 1983), 8.
5
Tolhurst, “Seemings,” 298–99.
6
Chris Tucker, “Why Open-Minded People Should Endorse Dogmatism,” Phil-
osophical Perspectives 24, no. 1 (2010): 530. Michael Huemer, Skepticism and the Veil of
Perception (Lanham, MD: Rowman & Littlefield, 2001), 77.
7
Tucker often distinguishes sensations from seemings with reference to blind-
sight and associative agnosia.
logan paul gage 89
be reducible to more familiar states like beliefs or inclinations/dispositions

to believe.8 Advocates of seemings tend to resist this reduction, however.9
Seemings do not appear to be beliefs because, even if seemings and incli-
nations typically go hand-in-hand, it is possible for it to seem that p without
believing (or being inclined to believe) that p.10 The standard example comes
from known illusions. If I know I am standing in front of a carnival mirror, I
will not believe (or even be tempted or inclined to believe) that I am as tall as
I appear, or so the reasoning goes. Seemings advocates get the right conclu-
sion here, but I think the reasoning is mistaken. If I am aware of the illusion,
then the content of my seeming will be “I appear really tall” rather than “I
am tall.” And I either believe or am inclined to believe the former. More
persuasive in my view is Huemer’s argument that seemings often “provide
non-trivial explanations for what we are disposed to believe. I am disposed
to accept that there is a white cat on the couch because that is the way things
appear to me, and this is not just to say that I am disposed to accept that there
is a white cat because I am so disposed.”11 Viewing seemings as nondoxas-
tic evidential states makes sense of the typical phenomenology of belief-
formation (where belief tends to be based on the way things seem to the
subject), and it stops the regress problem in an obvious way (by pointing to
something other than a belief to ground basic beliefs).
In summary, a seeming state is a nondoxastic, conscious experience of
which we are aware, with propositional content, distinct from belief and
mere sensation, which has the “feel” of revealing the way the world is. We
appear to have a variety of seemings that might be divided along the lines of
our basic sources of knowledge: perception, memory, introspection, rational
intuition, and perhaps others. What remains to be seen is whether seeming
states can play the four major roles of evidence.
8
See D. M. Armstrong, Perception and the Physical World (London: Routledge,
1961), 84–87; William G. Lycan, Judgement and Justification (New York: Cambridge
University Press, 1988), 165–66; Richard Swinburne, Epistemic Justification (Oxford:
Oxford University Press, 2001), 135–51; and Jason Rogers and Jonathan Matheson,
“Bergmann’s Dilemma: Exit Strategies for Internalists,” Philosophical Studies 152, no. 1
(2011): 55–80.
9
Advocates of seemings are not alone in holding that evidence is nondoxastic.
For example, John L. Pollock, Contemporary Theories of Knowledge (Totowa, NJ: Row-
man & Littlefield, 1986), 87–92; and Paul K. Moser, Knowledge and Evidence (New
York: Cambridge University Press, 1989), 88.
10
George Bealer, “A Theory of the A Priori,” Philosophical Perspectives 13 (1999):
31; Huemer, Skepticism and the Veil of Perception, 99–100, and “Compassionate Phe-
nomenal Conservatism,” Philosophy and Phenomenological Research 74, no. 1 (2007):
30–31.
11
Huemer, “Compassionate Phenomenal Conservatism,” 31.
The First Role: Evidence as That Which Justifies Belief
Reconciling the many philosophical accounts of evidence with how evidence

is conceived and spoken of in other disciplines, or even by lay folk, is exceed-
ingly difficult. Evidence just seems to play numerous, distinct roles. The first
such role that Kelly draws our attention to is that of justifying beliefs. Many
philosophers are convinced that the concept of evidence “is inseparable
from that of justification” or that which makes belief reasonable.12 Typical
evidentialists, of course, believe that evidence is the only thing that epistemi-
cally justifies belief. The phenomenal conception of evidence, I argue in this
section, accords well with this role of evidence.
The connection between seemings and epistemic justification appears
intuitive. If I ask you why you believe that p, it is very natural for you to
explain that it appears to you that p is true. You believe there is a tree outside
because it seems like there is one. Hence many philosophers have suggested
something along the following lines: “If it seems to S that p, then, in the
absence of defeaters, S thereby has at least some degree of justification
for believing that p.”13 Because internalists like Huemer have developed the
notion of seemings alongside their theories of justification, one might worry
that only internalists care about seemings, and perhaps only internalists care
about this first role of evidence as justifier. But this would be a mistake.
Recent work has begun to incorporate seemings in an externalist-friendly
manner.14 True enough, first-person evidence (like seemings) features
more prominently in internalist theories than in externalist ones. Yet even
Plantinga affirms that having evidence for one’s beliefs, being internally
justified in one’s beliefs, and being internally rational are epistemic virtues.15
Additionally, on externalist accounts, evidence is often necessary to warrant
or justification because one will often need defeater-defeaters (i.e., evidence
12
Jaegwon Kim, “What Is ‘Naturalized Epistemology’?,” Philosophical Perspectives
2 (1988): 390–91.
13
Huemer, “Compassionate Phenomenal Conservatism,” 30. For a catalog of
such principles, see Logan Paul Gage, “Objectivity and Subjectivity in Epistemology:
A Defense of the Phenomenal Conception of Evidence” (PhD diss., Baylor Univer-
sity, 2014, ch. 4).
14
For example, Michael Bergmann, “Externalist Justification and the Role of
Seemings,” Philosophical Studies 166, no. 1 (2013): 163–84. For further interaction
with Bergmann on externalism, internalism, and seemings, see Logan Paul Gage,
“Phenomenal Conservatism and the Subject’s Perspective Objection,” Acta Analytica
31, no. 1 (2016): 43–58.
15
Alvin Plantinga, Warrant and Proper Function (New York: Oxford University
Press, 1993), 3; and Warranted Christian Belief (New York: Oxford University Press,
2000), 203–4, 241.
logan paul gage 91
against potential defeaters).16 Moreover, as Kelly notes, many reliabilists have

felt the weight of the clairvoyance challenge to reliabilism and have modified
their externalist models accordingly.17 In attempting to account for why one
must be at least somewhat responsive to evidence, such externalists bolster
the commonsense connection between evidence and justified belief.
While evidence is that which justifies propositions or beliefs, evidence is
typically (if not always) defeasible. That is, for any piece of evidence e1 had
by S for some proposition p, there may be future evidence e2 that S could
gain that would either (i) undercut the support e1 was thought to give p or
(ii) simply outweigh the strength of the support e1 continues to give to p such
that the conjunction of e1 and e2 does not support the proposition that p.18
The phenomenal conception of evidence makes sense of the defeasibility
of evidence. In conscious experience it seems to us that certain proposi-
tions are true and that they bear on the truth of other propositions. But it is
possible that further experience may make it seem as though certain propo-
sitions were never really well-supported or that previous experience is now
outweighed by further experience.
But if evidence consisted of physical objects or publicly observable states
of affairs, in what sense would evidence be defeasible? How is a knife defea-
sible evidence that Smith committed the murder? Say that Detective Reagan
finds what appears to be Smith’s knife in a victim’s back. Later on, however,
Reagan finds Smith’s actual knife. On the phenomenal account, it is quite
easy to understand the first piece of evidence as misleading evidence. Misleading
evidence is not to be confused with apparent or fake evidence. Mislead-
ing evidence is genuinely evidence for a conclusion that turns out to be false.
Apparent or fake evidence, however, never really supports the conclusion in
the first place. As Kelly explains, “The fact that misleading evidence is genu-
ine evidence is why beliefs based on misleading evidence can be reasonable,
given that what it is reasonable to believe depends on one’s evidence.”19 It
appeared to Reagan that this was Smith’s knife, so he then in fact had evidence
that Smith was the murderer. But upon finding the second knife, he had a
second seeming state in which Smith did not appear to be the murderer.
The phenomenal account makes sense of Reagan’s initial seeming state or
evidence. The appearance of Smith’s knife at the crime scene is objectively
16
Plantinga, Warranted Christian Belief, 357–73.
17
For example, Alvin Goldman, Epistemology and Cognition (Cambridge: Harvard
18
Cf. John L. Pollock, Knowledge and Justification (Princeton: Princeton University
Press, 1974), 42–43.
19
Thomas Kelly, “Evidence: Fundamental Concepts and the Phenomenal Con-
ception,” Philosophy Compass 3, no. 5 (2008): 937. Cf. Kelly, “Evidence,” 55n9.
good evidence that Smith is the murderer (i.e., it raises the probability that
Smith is the murderer). This evidence can be outweighed, however, by learn-
ing that the appearance of Smith’s knife was only an appearance and nothing
more. So Smith’s total evidence does not support the proposition that Smith
is the murderer. The courtroom conception of evidence, however, has the
awkward entailment that Reagan never really had evidence for Smith’s guilt
in the first place. If evidence does not consist in the appearance of things
but in the objects themselves, then finding someone else’s knife at the crime
scene is in no way evidence for Smith’s guilt. So the courtroom conception
seems to collapse the intuitive distinction between misleading evidence and
apparent or fake evidence.
The defeasible nature of evidence also cuts against a Williamsonian view
of evidence in which one’s total evidence is simply the collective body of
propositions one knows—E = K, as the formula has it.20 On such a concep-
tion, evidence is factive; to know a proposition p, p must be true. On William-
son’s view one must say, as on the courtroom conception, that Detective
Reagan never had evidence that Smith was guilty if Reagan believed the false
proposition that “Smith’s knife was found at the crime scene.” Williamson
would likely protest that Reagan still had as evidence the known proposition
that “I [Reagan] was in such a state that it appeared to me that Smith’s knife
was at the crime scene.” There are two vexing problems with this approach.
First, a proposition must be believed to be known. And despite Williamson’s
assertion to the contrary,21 it seems psychologically implausible that we typi-
cally have beliefs about our appearance states rather than about the world
itself. It seems much more likely that Reagan believed that Smith’s knife was
found at the scene rather than that he believed he was appeared to Smith-
knifely at the scene. Second, even if Reagan did believe this higher-order
proposition about his experience, what justifies this belief ? For Williamson,
only the other propositions Reagan knows can constitute Reagan’s evidence.
So given that the phenomenal character of the experience cannot justify, and
given that it would be absurd for the believed higher-order proposition to
justify itself, which of Reagan’s other known propositions could possibly
justify this higher-order belief that he is having such-and-such an experience?22
If the courtroom and Williamsonian conceptions of evidence end
up implying that Reagan had no evidence for Smith’s guilt—and thus,
20
Timothy Williamson, Knowledge and Its Limits (New York: Oxford University
Press, 2000), 184–208.
21
Williamson, Knowledge and Its Limits, 198–99.
22
Cf. Anthony Brueckner, “E = K and Perceptual Knowledge,” in Patrick Gree-
nough and Duncan Pritchard, ed., Williamson on Knowledge (New York: Oxford Uni-
versity Press, 2009), 8.
logan paul gage 93
presumably, that he was unjustified in believing Smith guilty—then the

phenomenal conception clearly has the advantage in allowing that Reagan
was justified in his false belief because he possessed good, if misleading,
evidence. Indeed, it is precisely this consideration in regard to the mislead-
ing evidence of illusions and hallucinations that led prominent empiricists of
the early twentieth century like Russell to adopt the phenomenal conception
of evidence in the first place.23 However, recent “disjunctivists”24 maintain
that my counterpart and I in the New Evil Demon thought experiment25—
where we have all the same phenomenal experiences (qualitatively speaking)
but my counterpart has no veridical experiences of an external world—do not
share the same evidence. But if a fundamental role of evidence is to justify,
and my counterpart in the evil demon scenario appears to be justified in
holding his mistaken beliefs,26 then there is some pressure on the disjunc-
tivists to also revise their conception of justification and rationality. For this
reason Williamson writes, “Rational thinkers are not always in a position to
know what their evidence is; they are not always in a position to know what
rationality requires of them.”27 Following in Williamson’s footsteps, Clayton
Littlejohn has been forced to the radical conclusion that justification itself
must be factive.28 After all, if false beliefs are not evidence, then what kind of
guide for belief and action could they be? And if false beliefs cannot appro-
priately guide belief and action, then false beliefs cannot justify belief and
action. Hence justification is factive (i.e., one can only be justified in believing
true propositions). So the Williamsonian view of evidence not only offers
a revisionist view of evidence but also faces pressure to offer a revisionist
view of rationality and justification. We should not, then, follow Williamson
and the disjunctivists in their initial revision of evidence. The phenomenal
conception has the edge.
23
Bertrand Russell, The Problems of Philosophy, 2nd ed. (Oxford: Oxford Univer-
sity Press, 1998 [1912]); and Our Knowledge of the External World (Ithaca, NY: Cornell
University Library, 2009 [1914]).
24
For example, John McDowell, “Criteria, Defeasibility, and Knowledge,” Pro-
ceedings of the British Academy 68 (1982): 455–79; Williamson, Knowledge and Its Limits;
and Duncan Pritchard, Epistemological Disjunctivism (Oxford: Oxford University Press,
2012).
25
Keith Lehrer and Stewart Cohen, “Justification, Truth, and Coherence,” Syn-
these 55, no. 2 (1983): 191–207.
26
See B. J. C. Madison, “Epistemic Value and the New Evil Demon,” Pacific
Philosophical Quarterly 98, no. 1 (2017): 91–98, for three reasons to think that victims
of the New Evil Demon are justified.
27
Williamson, Knowledge and Its Limits, 164.
28
Clayton Littlejohn, Justification and the Truth-Connection (New York: Cambridge
The phenomenal conception of evidence also accords better with the

common notion that subjects must possess evidence for p if subjects are to
be justified in believing that p. If evidence consists in seeming states, this
makes perfect sense. We “possess” our seemings in the sense that they are our
conscious states; we have direct, privileged, first-person access to them. On
the courtroom conception, however, this seems impossible. In what sense
does the jury have the evidence of the knife or the fingerprint in a murder
trial? Perhaps this makes some sense in that they were once in a room with
those physical objects. But think of our evidence that Caesar crossed the
Rubicon in 49 BC. Surely we possess evidence for this proposition without
ever having access to the relevant archaeological material or the initial state
of affairs.
Defenders of the courtroom conception might reply that, while all
evidence consists in physical objects or public states of affairs, in order to
have a justified belief, a subject must have access to the relevant physical
object or public state of affairs and have some sort of experience with the
object or public state of affairs. So properly speaking, evidence consists
in objective things or states, while justification requires the addition of
subjective experience. But first, note the cost: evidence itself, on this view,
does not justify beliefs—not without an experiential state. In other words,
strictly speaking, this view does not account for the role of evidence as that
which justifies belief. Furthermore, consider again the case in which Detec-
tive Reagan believes he has seen Smith’s knife at the crime scene but, in
fact, it was not Smith’s knife. On the view under consideration, it is still
difficult to see how Reagan was justified in his false belief that Smith is the
murderer, even though Reagan surely seems justified. After all, on the view
under consideration, Reagan needs evidence and an appropriate experien-
tial state for justification. But if the knife was not Smith’s—or even worse,
say there was no physical knife but Reagan was the victim of an ingenious
illusion—then, because he had no evidence, Reagan’s belief was unjustified
(despite all appearances to the contrary). So on this view too there is pres-
sure to heavily revise our conception of justification. Better, then, to admit
that appearance states can justify, even in the absence of the relevant physi-
cal objects or states of affairs. We conclude not only that the phenomenal
conception of evidence can play the first role of evidence but that it does so
better than its chief rivals.
logan paul gage 95
The Second Role: Evidence as That

Which Rational Thinkers Respect
The second major role played by evidence that Kelly identifies is similar to,
yet distinct from, the first. It has long been thought that a hallmark of ratio-
nality consists in responsiveness to one’s evidence—whether this respon-
siveness consists in heeding higher-order evidence, conditionalizing on
new evidence, proportioning the strength of one’s belief to the strength
of one’s evidence, and so on. As I will argue, the phenomenal conception
accords better with recent developments in epistemology than its rivals.
Kelly draws our attention to recent literature that notes that higher-order
evidence (evidence about our evidence) is itself evidence.29 Say you observe
a strange, bright light in the night sky. The next morning, your neighbor says
she saw it too. You would naturally take this as evidence that your evidence
was legitimate (you weren’t hallucinating) and as further confirmation of the
strange light. Some have worried that this natural reaction might double-
count your evidence.30 This would be correct if the evidence was the physical
object or state of affairs itself. But if the evidence is your experience and your
neighbor’s testimony is a further experience in which it seems to you there was
a strange light, then your new and stronger seeming is itself evidence. Even
in the case of experts using the same first-order evidence, learning that the
other expert took the first-order evidence in the same way is a further expe-
rience that would seem to legitimately raise the expert’s credence. There is no
double-counting; there were two separate experiences. Thus the phenomenal
conception allows for the common practice of respecting one’s higher-order
evidence in a clear way.
With regard to conditionalization, Kelly himself notes that several
epistemologists—most prominently Putnam—have argued that the intro-
duction of a rival alternative hypothesis, even in the absence of any indepen-
dent evidence for it, should lower our credence in our original hypothesis.31
This would make little sense on the courtroom conception of evidence. After
all, by hypothesis, we have no independent evidence for the rival hypothesis,
so we have no physical evidence for it. Why then would it be rational to lower
one’s credence in her original hypothesis? On the phenomenal conception,
29
For example, Richard Feldman, “Respecting the Evidence,” in John Haw-
thorne, ed., Philosophical Perspectives vol. 19 (Oxford: Blackwell, 2005), 95–119.
30
Thomas Kelly, “The Epistemic Significance of Disagreement,” in Tamar
Szabo Gendler and John Hawthorne, ed., Oxford Studies in Epistemology 1 (Oxford:
Oxford University Press, 2005), 167–96.
31
Kelly, “Evidence,” 7–8. Hilary Putnam, Mind, Language, and Reality (Cam-
bridge: Cambridge University Press, 1975).
the introduction of the rival hypothesis will tend to induce some doubt and
decrease the strength of the seeming that the original hypothesis is true.
The strength of the evidence for the original hypothesis, in other words, has
decreased. So the phenomenal conception makes sense of why one’s confi-
dence should generally reflect the space of relevant alternative hypotheses of
which one is aware.
Notice too that our duty is to respect our evidence rather than to believe
the truth. Consider again my twin in the New Evil Demon scenario. Many
have suggested that because we appear to have the same justification for
believing the same propositions, we both ought to believe the same set of
propositions.32 There is something fundamentally good and correct about
heeding our evidence, even if it all turns out to be misleading. My deceived
twin is no less epistemically virtuous for having been systematically deceived.
Rational creatures are to do their epistemic best vis-à-vis their evidence,
regardless of access to the truth.
Recall, now, that on disjunctivism my demon-deceived counterpart and
I do not share the same evidence (since only my mental states are factive).
Because my counterpart believes numerous propositions about the material
world without any evidence, he would seem far and away less rational than
me despite our shared phenomenological and doxastic life. By sheer luck I
became more rational than my counterpart.33 The way to reject the disjunc-
tivist claim that my counterpart is irrational (or even less rational) is to accept
a view of evidence according to which we share the exact same evidence.
Yet we do not share the same knowledge, since my counterpart is deceived.
We do not share access to the same physical objects, as my counterpart may
not have access to a material world at all. What we share are our phenom-
enal appearances. Given that this is all we share, if we are to retain (i) the
commonsense intuition that we are equally rational and (ii) the commonsense
view that responsiveness to evidence is a hallmark of rationality, it is difficult
to see our evidence as consisting in anything other than our phenomenal
appearances. As Kelly writes, “Intuitively, the Demon misleads his victims
by exploiting their rationality, inasmuch as he trades on the sensitivity of their
beliefs to misleading evidence. . . . But the Demon misleads by providing his
victims with misleading experiences. Hence the temptation to simply identify
32
Stewart Cohen, “Justification and Truth,” Philosophical Studies 46, no. 3 (1984):
279–95.
33
Silins also argues that there is a problem lurking for evidential externalists (like
disjunctivists) in the other direction. On evidential externalism, it is possible for my
demon-deceived counterpart to sometimes be more justified than me depending on
how closely we align our credences with our total evidence. Nicholas Silins, “Decep-
tion and Evidence,” Philosophical Perspectives 19, no. 1 (2005): 375–404.
logan paul gage 97
one’s evidence with one’s experiences: once again, the phenomenal concep-
tion of evidence looms.”34 The phenomenal conception is at its strongest in
fulfilling the second role of evidence.
Still, critics of the phenomenal conception might object: if my completely
deceived counterpart is justified, then what good is evidence? Critics might
claim that the value of being a rational agent lies in getting at the truth (or in
increasing the likelihood of arriving at the truth) rather than in responsive-
ness to evidence.35 But this would be a mistake. As far back as Plato’s Meno,
people have wondered what makes knowledge more valuable than true belief.
The most plausible candidate is a first-person responsiveness to evidence or
reasons. But if the critic were right, then respecting one’s evidence would
seem to add nothing to the value of true belief. This cannot be a proper
account of rationality, for it implies that one wholly unresponsive to (or even
disdainful of) evidence, but who still believes the truth, could be rational. The
phenomenal conception, then, also has the advantage regarding the second
role of evidence as that which rational thinkers respect.
The Third Role: Evidence as a Guide to Truth
While the phenomenal conception easily fulfills the first two roles of evi-
dence and squares with common intuitions about justification and rationality,
it is less than clear that it can fulfill the final two roles. In its third role, Kelly
observes that evidence appears to function as a sign or mark of the truth.
Evidence is often thought of as that which indicates the truth of something
else. On the standard and most straightforward model, evidence e is thought
of as that which confirms (or disconfirms) a hypothesis h by making the
truth of h more (or less) probable.36 With this in mind, some might con-
cede that reasons-responsiveness or justification adds value to true belief but
still worry about totally severing the link between evidence and truth. One’s
phenomenal evidence in a demon world seems of little value: “one might
worry that a view according to which perfectly following one’s evidence is
compatible with a more or less completely mistaken view of one’s situation
threatens to render obscure why following one’s evidence would be a good
thing to do relative to the goal of having true rather than false beliefs.”37
34
Kelly, “Evidence,” 14.
35
Ibid., 15.
36
More technically, e is evidence for h iff the probability of h given e (in conjunc-
tion with background information i) is greater than the unconditional the probability
of h.
37
But why concede that the phenomenal conception severs the connection
to truth? My demon-deceived counterpart’s phenomenal evidence that “the
Sun is in the sky” indicates the truth of that false proposition. My counter-
part’s phenomenal evidence makes the evidenced propositions epistemically
probable for him. This epistemic probability is no less objective for being
epistemic rather than statistical. As far back as the beginning of modern
probability theory in the seventeenth century, we can distinguish two quite
distinct kinds of probability, the first dealing with statistical frequencies and
the second dealing with the degree to which an evidence set confirms a given
proposition.38 The latter can be given an objective or subjective interpreta-
tion,39 and I see no reason why the phenomenalist cannot take the objective
interpretation. Phenomenal states objectively evidence the truth of beliefs
with the same propositional content (even if the beliefs are false).
Some have thought that there are counterexamples to the necessity of
such evidential relationships. Ted Poston helpfully summarizes Bergmann’s
purported counterexample40 to the thesis that “the fittingness of doxas-
tic response B to evidence E is an essential property of that response to
that evidence” this way: “The counterexample Bergmann presents involves
possible cognizers who experience olfactory sensations of the type [normal humans]
experience when [normal humans] smell a meadow full of flowers whenever they
pick up a billiard ball and [naturally and non-inferentially] form the belief
that there is a smallish hard round object in my hand. . . . This belief is fitting, so
Bergmann claims. . . . However, the same belief is an unfitting response to
the same evidence in actual cognizers.”41 The phenomenal conception can
account for Bergmann’s intuition and yet uphold the necessity of evidential
relationships. While sensations often trigger seemings (and hence evidence),
they are not themselves evidence; they are not the bearers of propositional
content that stand in these objective evidential relationships.42 Given that
design plans can vary, it certainly seems possible for different sensations to
trigger different seemings. The phenomenal view, however, is committed to
38
Ian Hacking, The Emergence of Probability: A Philosophical Study of Early Ideas
about Probability, Induction and Statistical Inference, 2nd ed. (New York: Cambridge Uni-
versity Press, 2006), 11–17.
39
Swinburne, Epistemic Justification, ch. 3.
40
Michael Bergmann, Justification without Awareness: A Defense of Epistemic Exter-
nalism (New York: Oxford University Press, 2006), 118–21.
41
Ted Poston, “Justification without Awareness,” Philosophy and Phenomenological
Research 77, no. 2 (2008): 572.
42
Seemings rather than sensations are capable of providing justification. Sensa-
tions are evidentially relevant, but this is only because sensations “often affect what
I am justified in believing by affecting the way things seem.” Tucker, “Open-Minded
People,” 530–31.
logan paul gage 99
an objective, necessary relation between (i) seemings and what they evidence
but not between (ii) sensations and seemings or sensations and beliefs. This
counterexample only affects the latter.
The critic of the phenomenal conception vis-à-vis the third role follows
BonJour and others in holding that evidence and justification are only instru-
mentally valuable in attaining the truth.43 Even if this were the case, because
experience can objectively evidence certain propositions, even the demon-
deceived’s only hope of attaining truth is to heed his experiential evidence.
Heeding experience makes possible valuable states like knowledge and under-
standing that go beyond true belief. Understanding goes beyond true belief
to see conceptual and/or explanatory connections. True belief is not our
ultimate goal; true belief alone cannot explain the value of rational inquiry.
Rather, true belief where we possess reasons and see connections between
propositions and states of affairs is the true telos of the rational animal. The
value is in seeing for one’s self.
However, we should question whether having evidentially grounded or
justified belief is only instrumentally valuable. The view that justification is
only instrumentally valuable relative to the goal of true belief has unsavory
implications. Consider the following: A and B are both demon-deceived,
yet A is an excellent reasoner given how things seem to him, while B is an
extremely poor reasoner given the way things seem to her. Do the beliefs
of A and B have equal epistemic value? Madison argues44 that if following
evidence or holding justified beliefs is only valuable relative to attaining the
truth, and neither A nor B has any hope of attaining the truth because of
the evil demon, then neither of their beliefs have epistemic value. But intu-
itively this is not the case. Madison thinks that this is because A’s beliefs,
while false, can still be appropriately based on her seemings, are not overly
hasty nor overly confident, can cohere with each other, can display sensi-
tivity to defeaters, and so on. If this is correct, then justified belief (belief
appropriately based on evidence) is intrinsically valuable/excellent apart from
true belief.
Even if the foregoing is mistaken, we should not assume that factive
evidence would secure the truth connection better than phenomenalism.
First, recall that Williamson thinks that one can know the proposition that
“it seems to me that p,” and hence this proposition can be evidence for the
demon-deceived even if p is false.45 As I argued, we don’t normally have such
metalevel beliefs, and thus the demon-deceived would not typically have this
43
Laurence BonJour, The Structure of Empirical Knowledge (Cambridge: Harvard
44
Madison, “Epistemic Value,” 91–98.
45
Williamson, Knowledge and Its Limits, 198–99.
proposition as evidence. But if I am wrong and Williamson is correct, then

his factive view of evidence is at no advantage over the phenomenal concep-
tion regarding the truth connection. The demon-deceived with this known
proposition about experience will be led to innocently believe numerous false
propositions. So the factivity of evidence does not secure the truth connection.
Likewise, Williamson thinks that we are not always in a position to know
which propositions we really know and which we only think we know.46 If
correct, then what must guide our thoughts and actions is the way things
seem. Seemings—even for disjunctivists—are the very guide of life.
Setting aside those who think of evidence as factive, what about the
claim that “evidence for h must be a generally reliable indicator of the truth of
h”? Well, seeing as we have no access to God’s book of objective statistical
correlations, our only guide to reality is the way things seem. While some
might want to treat “e is evidence for h” as synonymous with “e is a reliable
indicator of h” so as to secure the connection between evidence and truth,
this is a mistake. Should our well-established statistical correlations fail to
hold in distant lands or even in the future (just think of Goodman’s grue
paradox), surely we should not say that we had no evidence for our current
scientific beliefs. The phenomenal conception and the third (signifying) role
of evidence, then, far from being incompatible, intertwine nicely.
The Fourth Role: Evidence as Neutral Arbiter
When it comes to justifying individuals from their own first-person point

of view, the phenomenal conception is more than up to the task. The more
difficult question is whether phenomenal evidence can serve the role it often
does in science—the fourth role of neutral, public, intersubjective arbiter.47
Listen to common usages and you will notice that “evidence” is often a con-
trastive notion. The person who has evidence for her belief is contrasted
in literature, film, and everyday discourse with the person who blindly fol-
lows tradition, untested prejudice, ancient texts, or ideological (and especially
theological) dogma.48 One key feature of this role of evidence is its public or
intersubjective nature, which is thought to lead to converging opinion over
Ibid., 174.
46
Achinstein, for one, argues that epistemologists’ conception of evidence is at

47
odds with how scientists conceive of evidence. Peter Achinstein, The Book of Evidence
(New York: Oxford University Press, 2001).
48
For recent work on phenomenal evidence and religious belief, see Logan Paul
Gage and Blake McAllister, “The Phenomenal Conservative Approach to Religious
Epistemology,” in John DePoe and Tyler Dalton McNabb, ed., Debating Religious
logan paul gage 101
time. Especially in science we have come to expect consensus opinion that

outstrips the consensus-forging powers of, for instance, religion. For this
reason, the word “evidence” itself is most readily identified in the public
imagination with science.
Kelly mentions at least three potential problems for the phenomenal
conception of evidence lurking in this fourth role: objectivity, peer disagree-
ment, and publicity. I will treat each in turn. It should be kept in mind that
these challenges have been thought to be the most serious—the very reason
cited by the logical positivists as to why they rejected the phenomenal concep-
tion, and one of the key reasons that prominent philosophers like Kelly and
Williamson reject it today.
Objectivity
The first problem Kelly raises is the objectivity of evidence. We already began
to treat this issue in the previous section and saw that the phenomenal con-
ception can easily maintain that evidential relationships are objective. So we
will only briefly address the issue here with reference to a common example
in the evidence literature—the example of Koplik’s spots—in order to gain
further insight into the phenomenal conception’s ability to handle this role
of evidence.
Kelly asks us to consider two individuals A and B. A sees
(i) the patient has Koplik’s spots on her skin
as evidence for
(ii) the patient has measles
while B does not.49 B is ignorant of the fact that Koplik’s spots typically
indicate the presence of measles—that is, that (i) statistically correlates with
(ii). The potential problem is that (i) seems to be evidence for (ii), objec-
tively speaking. That is, (i) is evidence for (ii) regardless of whether B is igno-
rant of the correlation. In this sense of evidence, we might say that smoke
was evidence for fire before anyone noticed it. The phenomenal conception
of evidence, it might be alleged, has trouble here since it thinks of evidence
as being possessed by a given individual. So, while it sure seems that by virtue
Epistemology: An Introduction to Five Views on the Knowledge of God (New York: Blooms-
bury Academic, forthcoming).
49
Kelly, “Evidence,” 35–36.
of being aware of (i) B has evidence for (ii), on the phenomenal conception it
would be natural to say that B lacks evidence for (ii) since (i) does not prompt
the seeming that (ii) is the case.
In response, the advocate of the phenomenal conception should note
that “evidence” is simply used in more than one way. There is a scientific
use of “evidence” in which “we are adopting an idealized third-person
perspective.”50 The epistemological sense of “evidence” in which we are
interested, however, is one in which a person’s evidence (her first-person
perspective) supports some proposition or propositions. Kelly himself holds
that “evidence” in the latter sense is basically synonymous with “reasons
for belief.”51 Given this, the phenomenal conception gets the correct answer
in this case: B does not possess reasons to believe the patient has measles.
Surely it is unreasonable for B to believe that the patient has measles given
what he knows (and does not know)—that is, given the way things seem from
his perspective. In fact, the phenomenalist should not agree that B has (i) as
evidence at all. B is not in a mental state with (i) as its content. He knows
nothing about Koplik’s spots. He only knows the patient has red spots on her
skin. This being the case, neither (i) nor (ii) is justified for B given the evidence
he possesses. What epistemologists are interested in is what propositions are
justified for A and B. Unless A and B know of the statistical correlation
between (i) and (ii), that correlation is not currently part of their perspective/
evidence—even if we can see that it is potentially part of their evidence from a
third-person perspective. The phenomenal conception thus renders the right
verdict and has little to fear from this first challenge regarding the objectivity
of evidence.
Peer Disagreement52
A second challenge for the phenomenal conception vis-à-vis this fourth

role of evidence arises from peer disagreement. A theory of evidence
must account for how it is that two seemingly rational and well-informed
people—what philosophers call “epistemic peers”—can come to widely
divergent positions. An important condition of peerhood is that peers share
the same (relevant) evidence. As Catherine Elgin writes, “Disagreement per se
50
Trent Dougherty and Patrick Rysiew, “Experience First,” in Matthias Steup,
John Turri, and Ernest Sosa, ed., Contemporary Debates in Epistemology, 2nd ed. (Mal-
den, MA: Wiley Blackwell, 2014), 19.
51
Kelly, “Evidence: Fundamental Concepts,” 953n4.
52
This section draws on Logan Paul Gage, “Evidence and What We Make of
It,” Southwest Philosophy Review 30. no. 2 (2014): 89–99.
logan paul gage 103
does not jeopardize epistemic standing. More problematic are cases in which
opponents are, and consider themselves to be, epistemic peers. Then they
have the same evidence.”53 One might be tempted to think that at least one of
the peers must be having a lapse of rationality if they disagree. But as Kelly
rightly remarks, since at least Kuhn there has been an increasing realization
of just how often fully informed scientists have disagreed.54 So either one has
to say that scientists are not engaged in the rational enterprise we thought
they were, or else one must account for rational peer disagreement. A variety
of solutions have been offered. Kuhn argued that rationality is relative to a
paradigm, and so two scientists coming to divergent opinions could be fully
rational if working within different frameworks.55 Carnap held that such sci-
entists might both be rational if they employed different inductive methods.56
Bayesians often appeal to the fact that the two scientists may have very dif-
ferent prior probability distributions. While there are limitations to disagree-
ment in that one’s priors can simply be swamped by overwhelming evidence,
these distributions can lead to reasonable disagreement.
These approaches have a similar structure: “What it is reasonable for one
to believe depends not only on one’s total evidence but also on some further
feature F (one’s prior probability distribution, paradigm, inductive method).
Because this further feature F can vary between different individuals, even
quite different responses to a given body of evidence might be equally reasonable. On such
views, the bearing of a given body of evidence on a given theory becomes a
highly relativized matter. For this reason, the capacity of evidence to gener-
ate agreement among even impeccably rational individuals is in principle
subject to significant limitations.”57 In other words, any view of evidence
that appeals to an extraevidential factor to explain rational peer disagreement
seems unable to fulfill this fourth role of evidence as neutral-arbiter; the
extraevidential factor does the explanatory work. I would like to suggest that
a serious advantage of the phenomenal view is its ability to account for ratio-
nal disagreement without appealing to extraevidential factors.
In order to explain rational peer disagreement, rival conceptions of
evidence tend to appeal to the view that evidence is one thing and what we
53
Catherine Z. Elgin, “Persistent Disagreement,” in Richard Feldman and
Ted A. Warfield, ed., Disagreement (New York: Oxford University Press, 2010), 53
(emphasis added). Cf. Kelly, “Epistemic Significance,” 174–75.
54
55
Thomas S. Kuhn, The Structure of Scientific Revolutions, 3rd ed. (Chicago: Univer-
sity of Chicago Press, 1996 [1962]).
56
Rudolf Carnap, The Continuum of Inductive Methods (Chicago: University of
Chicago Press, 1952).
57
Kelly, “Evidence,” 37 (emphasis added).
make of it is quite another.58 Many Bayesians, for instance, implicitly affirm

this distinction when they distinguish between conditional probabilities on
observations and unconditional probabilities of observations. If we want to
affirm the courtroom conception of evidence and yet deny that irrationality
in science is rampant, we might appeal to this distinction between evidence
and what we make of it. We might say that while two scientists share the same
evidence, they make different things of the evidence—perhaps they concep-
tualize or construe it in different ways, make different connections to other
pieces of literature, and so on.
To illustrate this view of evidence, say two equally well-informed and
talented arborists, Amy and Adam, walk through an unfamiliar forest.
Though the same evidence is available to both, Amy and Adam disagree over
the proper classification of a new tree species they discover. In the common
construal of evidence, Amy and Adam share the same evidence but use their
individual knowledge and skill (each of which is comparable to the other’s
knowledge and skill) to form the common evidence into support for incom-
patible propositions.
This distinction between evidence and what we make of it, I argue, rests
on a faulty notion of what evidence is in the first place. On the phenomenal
conception, evidence does not exist in the form of objects or events that are
“out there” in the world but in individuals’ mental states. We do not have
access to raw (i.e., unperceived) data. We have, for better or worse, our own
view of the world. It makes little sense, on the phenomenal conception, to
speak of evidence and what we make of it. The phenomenalist need not
think this distinction too far off the mark, however. There is still a great role
that each one of us, along with our background beliefs, plays in arranging
or construing data coming at us from the external world. Yet the phenome-
nalist should insist that this role is preevidential—it comes before the mental
states that constitute one’s evidence. Background beliefs, for instance, are
not merely static evidence but shape further evidence for a subject S by affecting
which mental states S has.
If this is correct, while our two arborists have very similar visual fields,
Amy, due to her particular background knowledge, actually has different mental
states than Adam in the forest. Amy and Adam amass different evidence (in
the form of seemings) because of their different background beliefs, concep-
tual frameworks, prior experiences, and so on. Their background beliefs and
experiences did not affect what Amy and Adam did with their common
evidence; rather, given their background beliefs, they simply have different
Cf. Jonathan L. Kvanvig, “The Rational Significance of Reflective Ascent,” in

58
Trent Dougherty, ed., Evidentialism and Its Discontents (New York: Oxford University
Press, 2011), 53.
logan paul gage 105
evidence despite their shared visual field and sensations. It seems to Amy that
the tree is of one species, while to Adam it genuinely seems another. Even
when it appears that two subjects have similar evidence for a given proposi-
tion p, they often do not—for they may have very different evidence in the
form of different seemings. In affirming this, the phenomenalist has a plau-
sible way to handle Kelly’s quandary regarding peer disagreement without
appealing to extraevidential factors to explain the disagreement.59 Hence the
phenomenal conception of evidence holds up well against this aspect of
the fourth role of evidence.
Publicity
Third, and finally, one might see a problem for the phenomenal conception
of evidence as regards the role of neutral arbiter because the phenomenal
conception posits that evidence is essentially private and subjective. Mental
states are only accessible to a single individual. How then can the phenom-
enal conception allow for neutral arbitration between individuals? Public
objects or events may seem better candidates for evidence when it comes to
science and intersubjective arbitration.
By way of reply, it is true that two individuals A and B cannot share the
same token mental states; each must have his own. But there seems no reason
to think that two individuals cannot have the same propositional content
within their two, admittedly different, token mental states. Kelly rightly notes
that the publicity of evidence has been a defining feature of science since
its earliest days. Robert Boyle, for one, was adamant that the witnessing
of experimental results was to be a communal affair.60 The public nature of
evidence in science has also been championed by Hempel, Popper, and more
recently Williamson.61 But surely Boyle and company are not insisting that
there must be no difference in perspective among witnesses to an experiment.
Surely they are not suggesting that science is subjective if these observers
59
The phenomenal conception’s recognition that two subjects may not share
the same evidence even when at first glance it might appear otherwise can also lead
toward a satisfactory resolution of other recent philosophical quandaries like Keith
DeRose’s Bank Cases. See Logan Paul Gage, “Against Contextualism: Belief, Evi-
dence, & the Bank Cases,” Principia 17, no. 1 (2013): 57–70.
60
Steven Shapin, “Pump and Circumstance: Robert Boyle’s Literary Technol-
ogy,” Social Studies of Science 14, no. 4 (1984): 481–520.
61
Carl G. Hempel, Fundamentals of Concept Formation in Empirical Science (Chicago:
University of Chicago Press, 1952), 22; Karl Popper, The Logic of Scientific Discovery
(London: Routledge, 2002 [1935]); and Williamson, Knowledge and Its Limits, 193.
possess different token mental states. If the concern is that nothing is shared
between multiple observers on the phenomenal conception, phenomenalists
should simply deny the charge: observers are likely to share the same type of
mental state with of the same (or similar) propositional content. Note too
that even if the courtroom conception of evidence is correct, in order for
an individual to possess the evidence, to gain a reason for belief, the evidence
must be appropriated by individual experience.
While Kelly worries about our inability to share token mental states,62
the logical positivists worried about our ability to argue about evidence
on the phenomenal conception. As Carnap explained his early view of
evidence, “Since the most certain knowledge is that of the immediately given,
whereas knowledge of material things is derivative and less certain, it seemed
that the philosopher must employ a language which uses sense-data as a basis.
In the Vienna discussions my attitude changed gradually toward a preference
for the physicalistic language. . . . In my view, one of the most important
advantages of the physicalistic language is its intersubjectivity, i.e., the fact
that the events described in this language are in principle observable by all
users of the language.”63 Ayer also describes the positivists’ shift away from
the phenomenal conception of the early sense-data theorists. He writes,
It was held . . . that perceiving physical objects was to be analyzed

in terms of having sensations, or as Russell put it, of sensing
sense data. Though physical objects might be publicly accessible,
sense data were taken to be private. There could be no question
of our literally sharing one another’s sense data, any more than we
can literally share one another’s thoughts or images or feelings. . . .
But the most serious difficulty [with this view] lay in the privacy
of the objects to which the elementary statements were supposed
to refer. . . . Because of such difficulties, Neurath, and subse-
quently Carnap . . . argued that if elementary statements were to
serve as the basis for the intersubjective statements of science, they
must themselves be intersubjective. They must refer, not to private
incommunicable experiences, but to public physical events.64
The worry isn’t just that our token mental states are private in that we alone
host them. The worry is about our ability to share evidence. Kelly develops
this worry nicely. The phenomenal conception, he writes,
62
Kelly, “Evidence: Fundamental Concepts,” 949–50.
63
Rudolf Carnap, “Intellectual Autobiography,” in Paul Arthur Schilpp, ed., The
Philosophy of Rudolf Carnap (La Salle, IL: Open Court, 1963), 50–52.
64
A. J. Ayer, Logical Positivism (New York: MacMillan, 1959), 17–20.
logan paul gage 107
stands in no small measure of tension with the idea that a cen-

tral function of evidence is to serve as a neutral arbiter among
competing views. For it is natural to think that the ability of ev-
idence to play this latter role depends crucially on its having an
essentially public character, i.e., that it is the sort of thing which
can be grasped and appreciated by multiple individuals. Here, the
most natural contenders would seem to be physical objects and
the states of affairs and events in which they participate, since
it is such entities that are characteristically accessible to multiple
observers. (I ask what evidence there is for your diagnosis that
the patient suffers from measles; in response you might simply
point to or demonstrate the lesions on her skin.) On the other hand,
to the extent that one’s evidence consists of essentially private
states there would seem to be no possibility of sharing one’s
evidence with others. But it is precisely the possibility of sharing
relevant evidence which is naturally thought to secure the objec-
tivity of science.65
So the problem is not just that two subjects cannot share a token seeming but
that to fulfill the fourth, public role of evidence, evidence must be sharable
and communicable.66
Let us begin with the former. Why must the phenomenalist deny
the ability to share evidence? Take Kelly’s own example: “I ask what
evidence there is for your diagnosis that the patient suffers from measles; in
response, you might simply point to or demonstrate the lesions on her skin.”67 This
makes perfect sense on the phenomenal conception: you point to the lesions
because you are attempting to get me to have a type-identical seeming with
the same (or similar) content. Carnap thought it a big advantage of his phys-
icalist conception of evidence that “the events described in this language are
in principle observable by all users of the language.”68 Yet what does this
mean except that if another person were in the same position at the same
time with the same relevant background beliefs and concepts, and so on,
then they could also observe that p? The phenomenalist need not deny this
but only claim that the evidence itself is not the event but the witnessing of
the event. Scientists attempting to replicate one another’s experiments, for
instance, are attempting to witness or observe the same states or events.
65
66
Cf. Kelly, “Evidence: Fundamental Concepts,” 949–50, and Williamson,
Knowledge and Its Limits, 193.
67
68
Carnap, “Intellectual Autobiography,” 52.
The phenomenalist can even explain why the public nature of scientific
evidence is not as public as advertised. That is, the phenomenal concep-
tion accounts for the fact that not just anyone can easily gain the relevant
scientific evidence (recall Carnap’s qualification that evidence is publicly
observable “in principle”). Consider the words of Hanson: “The layman
simply cannot see what the physicist sees . . . when the physicist looks at an
X-ray tube, he sees the instrument in terms of electrical circuit theory, ther-
modynamic theory, the theories of metal and glass structure, thermionic
emission, optical transmission, refraction, diffraction, atomic theory, quan-
tum theory and special relativity.”69 This difference obtains not only between
experts and nonexperts, according to Hanson, but between two experts like
Hooke and Newton (or Amy and Adam).70 The nature of scientific evidence,
then, is not exactly publicly available or “there for anyone to see”—at least
not in the most straightforward sense. Rather, as we saw with peer disagree-
ment, because evidence consists in seemings, a lowly philosopher could
view the same physical objects (e.g., the X-ray tube) and not have reason to
believe the same propositions as the trained scientist. Observers’ evidence
can differ without any physical fact before them differing. The phenomenal
conception, but not the courtroom conception, accounts for this. But this is
not to deny that scientific evidence is uniquely objective in some sense. In order
to have evidence like that of my scientist counterpart standing in the same
situation, I can undertake specialized training, read the same books and jour-
nals, and so on—which is surely possible in principle, unlike the ability to
observe another’s headache.
On the phenomenal conception, one can also in principle communicate the
content of their evidence because it has propositional content. The reason
Carnap, Ayer, and others71 think otherwise is that they think the phenomenal
conception is wedded to sense-data theory. On classical sense-data theory,
one has foundational sensations from which one infers various proposi-
tions. Sense-data theory is, then, an expression of indirect realism. We do
not have foundational evidence for external-world, object-level propositions.
Instead, we have sensations as our evidence. And this sort of evidence does
indeed seem very difficult to communicate. It is difficult, for instance, to
69
Norwood Russell Hanson, Patterns of Discovery: An Inquiry into the Conceptual
Foundations of Science (Cambridge: Cambridge University Press, 1961), 19.
70
Hanson, Patterns of Discovery, 13. N.B., the phenomenal conception need not
claim that observations are theory-laden in a way that prohibits one peer from poten-
tially observing what another observes. Cf. Jerry Fodor, A Theory of Content and Other
Essays (Cambridge: MIT Press, 1992), 251.
71
Kelly also links the phenomenal conception to indirect realism and sense-data
theory. “Evidence: Fundamental Concepts,” 945.
logan paul gage 109
know what sort of content sensations might have; classically sense-data has
been considered nonpropositional.72 Now recall again Kelly’s example: “I
ask what evidence there is for your diagnosis that the patient suffers from
measles; in response, you might simply point to or demonstrate the lesions on
her skin.”73 Perhaps the worry is that on an indirect realist, sense-data view,
one’s evidence is neither the lesions themselves nor the direct perception
of the lesions. Rather, one’s evidence—and that of which one is directly
aware—is the given, nonpropositional content of the sensory experience.
That experience, its content, and how one tacitly inferred propositions from
the sensing of nonpropositional sense-data may prove hard to communicate
indeed.
Yet note that earlier we distinguished (nonpropositional) sensations
from (propositional) seemings. The phenomenalist need only take the
latter as evidence. Moreover, the phenomenalist need not be an indirect
realist. Whereas the sense-data theorist thinks we only have direct access
to sensory experience, the phenomenalist might maintain that we have
access to objects by having mental states. That is to say, the phenomenalist
can hold that normally perception makes us directly aware of external-
world objects. True, as Huemer says, “we cannot perceive external objects
without having perceptual experiences that represent them.” But it is a
mistake to conclude
that we are not really, or not directly, perceiving external

objects at all, but only our representations. In fact, perceptual
experiences are the “tool” with which we perceive external
objects. Their existence no more precludes us from perceiving
those objects than the use of an axe precludes the woodcutter
from chopping his wood. And just as it would be a mistake
to conclude that the man is really chopping his axe, so it is a
mistake to conclude that we are really perceiving (or otherwise
enjoying awareness of) our perceptual experiences. We per-
ceive external objects by having perceptual experiences—in the
sense that those experiences partly constitute our perceiving of
external objects.74
The direct realist can claim that she is directly aware of physical objects and
that (noninferential) evidence for those objects exists in the form of seeming
72
Ibid., 940.
73
74
Huemer, Skepticism and the Veil of Perception, 81. Cf. Mortimer J. Adler, Ten
Philosophical Mistakes (New York: Touchstone, 1985), 5–53.
states. Being directly aware of a blue light, for example, is compatible with
one’s evidence being the seeming that there is a blue light. This evidence
is propositional and hence, in principle, communicable to others. Hence I
conclude that the phenomenal conception can play the fourth role of neutral
arbiter by providing a level of objectivity, by explaining how two peers can
rationally disagree, and by providing for the publicity of evidence as sharable
and communicable.
Conclusion
In Kelly’s final estimation, there is a paradoxical tension between two of

the four roles of evidence.75 On the one hand, evidence is that which
justifies belief. In this regard, introspective, first-person phenomenal evi-
dence fares quite well, while the courtroom conception fares poorly. On
the other hand, evidence is often seen as that which might lead to consen-
sus opinion and can be shared by multiple individuals. Here, Kelly thinks,
the courtroom conception fares well, while the phenomenal conception
fares poorly. If the preceding arguments were successful, this dissatis-
fying paradox is resolvable. What is more, we are now in a position to
think about why the phenomenal conception is able to play both kinds
of roles.
Following the terminology of Silins, we can see why Williamson and the
late-logical positivists might be lumped together despite their unique views
of evidence.76 Both are expressions of “evidential externalism” in thinking
that evidence is a matter not merely of inner mental states but also of the
subject’s external environment. Similarly, we can see why many lump together
the phenomenal conception with sense-data theory: both are expressions
of “evidential internalism,” which sees evidence as consisting in wholly
internal mental states. But as Kelly himself notes, the propositionalist/
nonpropositionalist distinction cuts across the evidential internalist/externalist
divide.77 A propositionalist, here, is someone who thinks that evidence has
propositional content. Hence we have four basic views: propositionalist and
nonpropositionalist evidential internalism and propositionalist and nonprop-
ositionalist evidential externalism.
75
Kelly, “Evidence,” 45–46.
76
Silins, “Deception and Evidence.”
77
Kelly, “Evidence: Fundamental Concepts,” 940.
logan paul gage 111
Matrix of Major Conceptions of Evidence
Propositional Content No Propositional Content

Evidential Seeming State Conception Sense-Data Conception
Internalism
Evidential Disjunctivist Conception Courtroom Conception
Externalism
Considering this logical space reveals why the phenomenal conception

is able to play the role of both individual justifier and intersubjective arbiter.
On the one hand, it must be person-relative and subjective (or internalist) in
order to be the kind of thing that can adequately justify beliefs for particular
subjects. On the other, it must have propositional content that can easily exist
in objective evidential relationships. While the sense-data theory is subjective/
internal, it has difficulty accounting for the objectivity of evidential relation-
ships. While the courtroom conception was thought to yield objectivity, it has
little chance of playing the role of justifier. While the disjunctivist concep-
tion of evidence is propositional (and hence can stand in objective evidential
relationships), it is too objective and thus yields intuitively incorrect conclu-
sions about justification and rationality. The phenomenal conception can
play the various roles of evidence precisely because it is subjective (internal)
and objective (propositional) in the right ways.78
78
The author thanks Trent Dougherty, Blake McAllister, and Alex Plato for
helpful comments on this essay.
The World, the Deceiver, and
The Face in the Frost
Lydia McGrew
Introduction: A Tantalizingly Incomplete

Solution to the Deceiver Challenge
In an appendix to The Foundations of Knowledge, Timothy McGrew provides

the outline of a solution to the problem of the external world.1 McGrew
argues that the probability of the existence of a deceiver who makes it appear
that we live in a real external world must be lower than the probability of a
real external world itself because the ontological commitments of the latter
hypothesis will always necessarily be greater than those of the former. In the
latter hypothesis, we posit a mental state of the deceiver as a cause of each
of the apparently real things that seem to exist in the external world, but the
deceiver himself also exists as an entity who is not merely the sum of all of
these mental states.
McGrew argues, further, that any time we conditionalize on some particu-
lar mental state of our own that we normally take to be caused by real objects
in the external world, the gap in probability between realism and the deceiver
hypothesis grows larger. He bases this argument on the probabilistic fact
that if one theory is strictly simpler than some other theory, the confirma-
tion a given piece of evidence affords to the simpler theory is always greater
than the confirmation it affords to the more complex theory—the difference
between the old probability of the simpler theory and its new probability
is always greater than the comparable difference between the old and new
probabilities of the more complex theory.2 McGrew’s argument thus shows,
if we take it to be successful, that the prior probability of a deceiver scenario
is lower than the prior probability of realism and also that, as we gradually
conditionalize on more and more everyday evidence, the gap in probability
between the two will continue to grow. This set of conclusions would seem
to mean (since we have a great deal of sensory evidence that we normally
1
Timothy McGrew, The Foundations of Knowledge (Lanham, MD: Littlefield Ad-
ams, 1995), 135ff.
2
Ibid., 134–36.
© Lydia McGrew, Quaestiones Disputatae, Vol. 7, No. 2 (Spring 2017)
lydia mcgrew 113
take to be caused by external-world objects) that what seems intuitively right

is actually supported by epistemology and probability theory—namely, that
realism is enormously more probable than a deceiver scenario based on all
of our evidence, despite the fact that (as McGrew has set up the conditions)
the two hypotheses are empirically equivalent in the sense of giving identical
probability to all of our sensory and memorial evidence.
If this explanation of what McGrew has shown is correct, his argument
is of tremendous importance and deserves a good deal more attention than
it has received. In point of fact, everything he has argued probabilistically
is completely correct. The only difficulty arises when we come to gauging
the significance of the results and in particular of the result concerning the
growing gap between realism and an empirically equivalent deceiver scenario.3
McGrew makes no distinction in his 1995 discussion between a growth
in the absolute value of the gap between the two hypotheses and a growth in
the ratio of the probabilities of the two hypotheses. This is significant for the
interpretation of his result. Intuitively, what it seems should happen is that
the ratio of the probabilities of realism and the deceiver hypothesis comes to
favor realism heavily, so that, if we take it that hypotheses other than either
of these are strongly disconfirmed or even flatly ruled out by the evidence,
we are in the end strongly justified in believing realism. Realism, it seems,
should have a very high posterior probability conditional on all evidence.
This, however, does not follow from what McGrew has argued. For suppose
that we grant the point that McGrew has argued concerning the prior proba-
bilities. This point, however, could be satisfied by a very weak inequality.
Let us suppose then that the prior inequality favors realism over deceiverism
by a ratio of 51/49. This could be the case even if both prior probabilities
were very low—for example, if we were to model those prior probabilities
as .00051/.00049. (The prior probabilities of realism and deceiverism do
not have to sum to 1.) As we conditionalize on further evidence for which
the two theories are empirically equivalent, that ratio does not change. This
follows from their empirical equivalence and holds despite the growth in
the absolute gap between them. Repeated conditionalizing upon evidence
for which two theories are empirically equivalent simply causes the ratio of
the posteriors to approach the ratio of the priors. If all other theories are
effectively ruled out by the evidence, the posterior probabilities for realism
and the deceiver hypothesis on this model, when all our evidence is taken
into account, will actually be .51 and .49. The absolute gap between these is,
of course, much larger than the absolute gap between .00051 and .00049, but
since the ratio is identical, the final result is not nearly as exciting as we could
3
I am indebted to Timothy McGrew for the criticism that follows of his own
1995 argument.
114 The World, The deceiver, and The Face in The FrosT
have wished. Speaking colloquially, realism ends up in this model only slightly
more probable than not. The growth of the absolute gap is simply not very
epistemically important by itself, aside from further information about the
ratio of the priors, and McGrew’s argument about the priors does not, by
itself, justify postulating a prior ratio that strongly favors realism as opposed
to a ratio that favors realism merely to some unspecified extent. Hence his
argument as a whole does not justify us in concluding that the posterior ratio
strongly favors realism either.
It is a significant thing in itself to argue successfully that realism must
have a higher prior and hence higher posterior probability than an empiri-
cally equivalent deceiver scenario; we should not minimize the value of that
inequality. But it does not by itself justify the very high posterior proba-
bility we intuitively give to realism in our daily activities and that philosophers
would like to argue for in answer to skepticism. If a probability only slightly
above .5 is considered insufficient for justified belief, this argument would
not show that belief in realism is justified. Hence McGrew’s solution to the
problem of the external world is incomplete.
What makes McGrew’s answer to the problem of the external world
incomplete is his granting the skeptic for the sake of the argument that the
only type of deceiver hypothesis we need to consider is one that is by defi-
nition empirically equivalent to realism. If H′ is defined as being empirically
equivalent to H no matter what evidence comes to light or no matter how much
evidence is involved, then empirical reasoning is brought to a halt by treating
paradigmatic ad hocness as epistemically legitimate. Such a move would not
merely favor the skeptic in the realism/antirealism debate but would also
make it impossible to give strongly preferential rankings to different hypoth-
eses within realism and antirealism.
I will argue that realism is strongly favored over antirealism if rational
empirical reasoning is possible at all—that is, if it is not blocked by treat-
ing empirical equivalence a priori as an insuperable problem.
1. Enshrining Ad Hocness
Elsewhere I have argued that there should be two concepts of ad hocness.4

One, the formal concept, involves the disconfirmation of a hypothesis by
some evidence E because of the necessity to conjoin the hypothesis with an
auxiliary that had (in the old probability distribution) low probability condi-
tional on the hypothesis, while the negation of the hypothesis had no similar
4
Lydia McGrew, “On Not Counting the Cost: Ad Hocness and Disconfirma-
tion,” Acta Analytica 29 (2014): 491–505.
lydia mcgrew 115
problem explaining the evidence. The second concept, which is argumenta-

tive and informal, involves irrationally thinking or trying to induce others to
think that a hypothesis has not been disconfirmed in this way when in fact
it has been. Thus all argumentative ad hocness is formal ad hocness, though not
vice versa; a reasoner might admit the correct degree of disconfirmation of his
preferred theory by the need to conjoin it with a low-probability auxiliary but
might rationally continue to believe the theory if it had enough independent
evidence in its favor.
A paradigmatic example of argumentative ad hocness is the infamous story
of dianetics founder L. Ron Hubbard when he attempted to demonstrate
the success of his methods with a protégé. He told the audience that the
young woman, having gone through his tutelage to become a “clear,” could
remember every detail of her past life. The demonstration was a dismal fail-
ure; the woman could not answer the audience’s questions, including a simple
question about the color of Hubbard’s tie. Desperate to save face, Hubbard
hypothesized that by calling her forward with the phrase “Come out here
now,” he had fixed her in the present and had thus made it impossible for her
to demonstrate her remarkable abilities.5
Hubbard’s epistemic and rhetorical behavior on this occasion was an
attempt to induce irrationality in others by refusing to admit disconfirmation.
He was asking the audience to overlook the serious disconfirmation of his
claims by the failure of the experiment, and he was presenting the auxiliary
hypothesis concerning the phrase “Come out here now” as an excuse for his
followers to agree that disconfirmation had not occurred.
Such paradigmatic argumentative ad hocness is all the worse if applied
repeatedly to multiple items of contrary data. Argumentative ad hocness indi-
cates that the person “accommodating” the contrary evidence is refusing
to admit that the evidence is, indeed, contrary. In that case, he can never
be convinced by evidence to change his mind. From a Bayesian perspec-
tive, this means that he is not properly conditionalizing on the evidence,
since the evidence does in fact significantly disconfirm his preferred theory,
while he is retaining the same or virtually the same probability for the theory
regardless of evidence. The same principle would apply if the subject were
admitting only an extremely minimal and nonthreatening degree of discon-
firmation when, in fact, the theory is significantly disconfirmed. One can
on this analysis admit the importance of some sort of ultimate openness to
falsification—that is, disconfirmation so great that the proposition should no
longer be believed—without adopting the Popperian baggage that can come
with the term falsification.
5
Martin Gardner, Fads and Fallacies in the Name of Science (Mineola, NY: Dover,
1957), 270–71.
The Cartesian skeptic argues that we cannot be justified in believing

in external-world realism because of the definitional empirical equivalence
between a deceiver scenario (be it a brain in a vat scenario or a Cartesian
scenario without the mechanism of brains and vats) and realism. The idea
is that, since the deceiver scenario has been defined to give just as high a
probability to all of our sensory and memorial evidence as that conferred
by realism, we cannot decide between them. Even if the Cartesian skep-
tic were to admit McGrew’s point that deceiverism is to some undefined
degree less probable than realism (which he probably would not), McGrew’s
argument permits the skeptic to retreat only to the point of saying that real-
ism is marginally more probable than not—perhaps as little as something like
.500001, which hardly seems adequate for justified belief in any robust sense.
But it is important to realize the enormous power that the Cartesian
skeptic is tacitly giving to this notion of definitional empirical equivalence.
Empirical equivalence is being made “trumps” in all possible instances. It
seems that, in order to argue that realism can never be justified, the Cartesian
skeptic has to adopt some principle like this.
Trumping Empirical Equivalence Thesis—For any empirical

hypothesis H, we can define a rival hypothesis H′ (or a disjunc-
tion of such rivals) that entails the falsehood of H and that
is by definition empirically equivalent to H for all nondeductive
evidence E, whether presently possessed or not. Since we cannot justify
strongly disparate prior probabilities for H and H′, then, given
the possibility that such an H′ is true, we cannot be justified in
believing H.
The claim that we cannot justify strongly disparate prior probabilities might
be defended on the grounds that absolute prior probabilities (prior to all
empirical evidence) are indefensible or counterintuitive, an issue I will return
to in the penultimate section.
Perhaps the skeptic never thought of himself as making any such sweep-
ing claim for all empirical hypotheses, but it is difficult to see what principled
grounds there could be for making empirical equivalence “trumps” in the
case of realism versus antirealism while not making it “trumps” in other
disputes. If we cannot be justified in believing realism because of the possi-
bility that all of our experiences are caused by a superpowerful deceiver or
deceivers, and because this deceiver hypothesis (or class of hypotheses) has
been declared to be by definition empirically equivalent to realism for all
evidence, why would the same principle not apply to justification in other
contexts? For example, suppose that we consider the conditional probability
that real dogs exist if realism is true—P(D|R & E), where E is the evidence
lydia mcgrew 117
that most ordinary people have concerning the existence of dogs. If we knew
with certainty that realism were true, could we then be justified in believing
that dogs exist, given our usual empirical evidence? If we accept the skep-
tic’s strictures on belief in realism, then by parity of reasoning it seems that
we could not. For it would be possible to construct an in-world skeptical
hypothesis according to which dogs do not exist (there are only nonliving
dog robots) and there is a gigantic conspiracy at a high level to convince
people that real dogs (mammalian animals who breathe, breed, etc.) exist.
One could then declare this hypothesis D′ to be by definition empirically
equivalent to D for all evidence E that might otherwise be relevant, and in
that case one could never be justified in believing D.
What this shows is that the skeptic’s approach to the realism/antirealism
debate enshrines argumentative ad hocness and treats it as though it were not
irrational. For the Trumping Empirical Equivalence Thesis will ultimately
mean that H′ cannot be disconfirmed. One way to see this point is to envis-
age an initial set of evidence as being so entirely inexplicable by anything
other than H and H′ that it rules out everything else—that is, conditionalizing
on that initial evidence makes P(~H & ~H′) = 0. In that case, suppose that
more evidence beyond that initial set comes in. By that time, H and H′ form a
partition, and if the evidence is relevant, it will confirm one or the other and
hence disconfirm one or the other. But H′ has been declared to be by defini-
tion empirically equivalent to H for all possible nondeductive evidence, so H could
never gain probability at the expense of H′ and their probabilities would
not change. Further evidence would be irrelevant. But it seems clear that it
is possible that there could always be more evidence that would be relevant
to, say, deciding between
D: Dogs exist.
and
D′: Dogs do not exist, but dog robots exist along with a conspiracy to
make people believe in dogs.
If D′ can be simply declared by fiat to be empirically equivalent to D for all

possible evidence, then, once everything else is ruled out, it becomes impossible
to disconfirm either of them and hence, inter alia, impossible to disconfirm
D′. But whatever evidence we have so far on a particular empirical issue, it
seems always to be possible that we could obtain additional relevant evidence.
In the case of the existence of dogs, if we have n apparent dog-like expe-
riences thus far, we could always have n+1 apparent dog-like experiences.
Or we could obtain some new type of evidence, such as the opportunity
to do a postmortem examination of what appeared to be a dog and see

that it appears to have mammalian organs and tissues. That evidence, then,
would have to be treated as not disconfirming D′ by an argumentatively ad
hoc move—by, for example, adding the auxiliary of further cleverness on the
part of the deceivers and acting as though this rendered the evidence nondis-
confirmatory. The same is true if the skeptic admits some slight difference
in posterior probability between the skeptical and the realistic scenario but
asserts that we can never be justified in considering the realistic scenario to
be any more probable than that—that is, that the ratio between the pos-
terior probabilities cannot undergo significant change. Once the skeptical
scenario reaches its “floor” posterior probability, further disconfirmation
will be treated as impossible on the grounds of the defined empirical equiv-
alence alone.
Aside from the possibility that all options other than H and H′ are strictly
ruled out by some evidence, it ought to be possible in principle that some
future evidence could both confirm H and disconfirm H′ at the same time.
While this is not necessarily the case if H and H′ do not form a partition, it
may be the case for some particular item of evidence. But Trumping Empir-
ical Equivalence makes this impossible by declaring that H can never be
confirmed at the expense of H′, no matter what the evidence is.
This brings us to another problem with Trumping Empirical Equiv-
alence: the H′ envisaged in the “principle” articulated there is not a well-
defined hypothesis at all and hence cannot have a real probability. Because
the H′ is defined in a purely derivative, piggybacked fashion vis-à-vis H and
vis-à-vis all possible evidence, rather than a definite set of evidence, H′ does
not have well-specified content. It is an indefinite hypothesis. The content
of H′ might well need to be augmented and hence changed by the addi-
tion of further auxiliaries to account for some later evidence that comes to
light. In fact, a modification that would help H′ account for one piece of
evidence as well as H does could in principle be harmful to the ability of H′
to account for some other piece of evidence. Only hypotheses with definite
content can have probabilities, either in a prior or a posterior probability
distribution, so the H′ described by the trumping principle should not be
taken to be a univocal hypothesis with a genuine probability.
It is important not to become confused here by something akin to a
scope shift. It may be correct to say, for some particular stated evidence
E and a hypothesis H, that it is possible to invent a rival H′ that entails the
falsehood of H and gives equal probability to E. But it doesn’t follow that,
for any hypothesis H, there exists some univocal, meaningful rival H′ that is
empirically equivalent to H for all possible E. Even if an H′ is gerrymandered
to account for some particular set of evidence just as well as H does, it does
not follow that that H′, without having its content changed by elaboration
lydia mcgrew 119
(and being disconfirmed because of that ad hoc change), will account for
some other evidence as well as H does nor that H′ can be meaningfully defined
as always doing so while continuing to be the same hypothesis. It is there-
fore possible that an H′ that is to some degree empirically similar to H could
be disconfirmed by further evidence and that H could in the end be highly
justified. Indeed, this possibility must be kept open if the course of inquiry
is not to be blocked a priori.
The Trumping Empirical Equivalence Thesis, then, ultimately makes
ongoing, accurate conditionalization on empirical evidence impossible even
in cases where it seems intuitively obvious that there could be further relevant
empirical evidence that should move our probabilities, disconfirming one
hypothesis and confirming another. This insistence that empirical reasoning
becomes arbitrarily “stuck” at a certain point merely because of empirical
equivalence arises from the move of declaring that a rival hypothesis can
simply be deemed to be empirically equivalent to some H for all evidence. This
insistence is a science-stopper in the strict sense and in general an empirica
l-rationality–stopper, not merely in the realist/antirealist debate but across
the board.
We must, then, reject the Trumping Empirical Equivalence Thesis if
we are to maintain the possibility of ongoing empirical rationality and not
wrongly treat argumentative ad hocness as rational. Either Trumping Empir-
ical Equivalence is false or rational empirical justification by nondeductive
reasoning is not possible, since the skeptic of any hypothesis H (other than
those known to be true a priori or by direct introspection) could always refer
to the possible truth of some definitionally empirically equivalent H′.
This point opens up important resources for answering the external-
world skeptic. Once Trumping Empirical Equivalence is rejected, there is
no principled reason to insist that we must dream up a “packed” hypothesis
that explains even all of our current evidence and to treat that rival as, at most,
slightly less probable than the realist scenario. For if we were required to do
that at every particular moment, the antirational effect would be exactly the
same as treating H′ as by definition empirically equivalent to H. Since consis-
tency would require us to apply this in all other scenarios as well, we would be
obliged to block justification for any hypothesis on the basis of nondeductive
evidence, simply because an empirically equivalent rival could be invented
for the existing evidence. At every moment, for every new piece or type of
evidence that we encountered or brought to mind, we would have to allow
the skeptic to come up with a new version of a rival hypothesis, gerryman-
dered to account for the new evidence, and summarily declare that to be
only (at most) slightly less plausible than realism. This would enshrine ad
hocness and prevent us from properly conditionalizing on new evidence that
confirmed realism at the expense of deceiverism. Instead, as I will suggest
in the next two sections, we should evaluate and compare realism to a more
generic rival.
2. Ad Hoc Irrationality and In-World Deceivers
To see how the concept of additional evidence and ongoing disconfirma-

tion can help in answering deceiver scenarios, consider an in-world deceiver
claim—the case of the clever burglars. Suppose that I have a paranoid friend
visiting me who fears that the CIA follows him and searches the houses
where he stays. One morning he tells me that the CIA has been in the
house in the middle of the night before. Upon questioning him, I learn that
he is not saying this on the basis of anything he heard or saw in the night.
Indeed, he does not expect that he would have seen or heard anything had
the CIA broken in. The “deceiver” hypothesis in this case is as follows:
B Clever and capable CIA burglars who are very good at hiding their
traces have broken into the house in the night and searched it.
This hypothesis is intended to be to some degree empirically similar

to the claim that no one has broken into the house, since the burglars are
supposed to be good at hiding their traces. But does this mean that it is
impossible for it to be disconfirmed by any empirical evidence that is well-
explained by a normal night? If so, something is wrong. It ought, epistemically,
to be possible to disconfirm the hypothesis and, in the process, to confirm
the hypothesis that the night was undisturbed—N (for normal):
N Last night was a normal night in which no one broke into the house
or searched it.
B and N, of course, do not form a probabilistic partition. But virtually any

evidence that confirms N will ipso facto disconfirm not only B but also the
presence of ordinary burglars without special powers. And a great deal of
evidence that disconfirms B would be of a sort that would confirm N.
In such a situation in real life, I would point out to my friend evidence
such as the following:
E1 The dead bolts on the house doors, which lock only from the inside,
are all locked.
My friend, given his paranoia, may then tell me that the CIA must have some
ingenious device, the details of which he does not know, for locking dead
lydia mcgrew 121
bolts from the outside while making it look as though they have been locked
from the inside. But the very need to postulate this auxiliary hypothesis to
deal with a particular piece of contrary evidence, while N has no need for any
such special auxiliary, explains why B is disconfirmed by E1, and my friend
should admit it. After all, merely saying that the CIA burglars are clever does
not in itself mean that they have such a surprisingly ingenious device for
mimicking the locking of dead bolts from the inside. The existence of such
an ingenious device actually has quite low probability conditional on B. E1 is
quite well-explained by N. Arguably, it is better explained by N than by ~N,
so it confirms N.
I might then bring up the following:
E2 None of the burglar alarms went off in the night, and the burglar
alarm is protected by a complicated password, which I never write
down, and it is working properly as confirmed by a test this morning.
Again, the paranoid friend can hypothesize that the CIA has found some
unknown way of figuring out my password for my burglar alarm system
and thus turning it off and back on without a trace, but this too will be an
auxiliary that was fairly improbable on B itself, while ~B has no such hand-
icap. Hence my friend should admit that the postulation of this auxiliary to
account for E2 comes with a cost—B has been disconfirmed once again.
And so it goes for various points. I have (let’s say) extremely sharp hear-
ing, sleep lightly, and got up several times, and I heard and saw nothing out
of the ordinary in the night. The floor is messy and includes my child’s care-
ful arrangement of dinosaur models in front of most of the windows, where
nothing is crushed or appears out of place from its position the previous
night. There is new-fallen snow outside, which shows no traces of footprints,
and so on.
The important point is just this: while B implied a certain degree, perhaps
even a high degree, of cleverness, technological capability, and knowledge on
the part of the CIA agents, it need not and should not be taken to be defini-
tionally empirically equivalent to the hypothesis that last night was a perfectly
normal night at my house in which no one broke in at all. If my friend were
to keep on saying, “But I told you already that these are very clever agents
who know how to hide their traces” in response to all evidence, refusing to
admit that any such evidence could disconfirm B and basing this refusal on
the fact that he had simply defined B in such a way as to be immune to such
disconfirmation, he would be irrational. Whatever the prior probability of B,
its posterior probability certainly should be significantly lower after condi-
tionalizing on a set of evidence such as this.
It would even be possible to elaborate the clever burglar hypothesis so

as to account for evidence concerning the background probability of B. For
example, suppose that we confront the paranoid friend with a CIA agent
who tells him how, in his past experience, the organization actually searches
houses and who asserts (with details) that some trace would have been left
in my house had this occurred the previous night. The paranoid friend can
simply say that the CIA agent doesn’t know about this clandestine program
that is tracking him. Suppose that the paranoid friend admits that he actually
has no prior evidence of such CIA activity but points out that this is what
would be expected if the agents have always in the past been sufficiently clever
at hiding their traces as to make themselves empirically invisible. Should
we at that point, even if not accepting his theory, retreat to saying that it has
an equal or near-equal probability to N because, via elaboration, he has made
it empirically equivalent for all evidence relevant to the prior probability as
well as the specific evidence concerning my house?
The in-world case shows how enshrining ad hocness means not letting
the evidence speak. If empirical equivalence were given as much clout as the
external-world skeptic gives it, the same principle would apply to in-world
cases whenever a sufficiently elaborate in-world skeptical scenario could be
gerrymandered to be empirically equivalent to a nonskeptical hypothesis for
all data up to the present, including data relevant to the prior probability.
3. Generic Deceiverism and Generic Realism
Consider the following statement of realism, which I will call “generic

realism”:
GR: There exists a fairly stable extramental physical world to which I

have fairly reliable sensory and memorial access.
Now consider a statement of the deceiver hypothesis, which I will call

“generic deceiverism”:
GD: Generic realism is false, and there exists a powerful deceiver (or
group of them) who wishes to produce evidence that will cause me
to think that there exists a fairly stable extramental physical world to
which I have fairly reliable sensory and memorial access.
I propose that, if we reject the Trumping Empirical Equivalence Thesis, we

can treat GD and GR as rival hypotheses and see how they do in competition
vis-à-vis the empirical evidence that we possess.
lydia mcgrew 123
GD, of course, is not by definition empirically equivalent to GR. Nor has

it even been gerrymandered to be empirically equivalent for all of the rele-
vant evidence that I can think of right now. (Much of this essay will consist in
arguing that much of the evidence I can think of is better explained by GR.)
But as discussed, it was only the idea that deceiverism is by definition empiri-
cally equivalent to realism that generated a motivation for trying to “pack”
deceiverism so that it can account for all the evidence we can think of just
as well as realism does. If we reject Trumping Empirical Equivalence, we are
free to consider a more generic rival deceiver hypothesis, to call gradually to
mind and attend carefully to various items and classes of our evidence, and
to see in a natural epistemic fashion the effect of this cumulative case.
The resemblance to the clever burglar hypothesis in the last section
should be clear. In both cases, we are using a hypothesis involving deception
that has some explanatory force but is also capable of straightforward discon-
firmation. B would have been empirically equivalent to N for some evidence.
The fact that the house has not been visibly ransacked, for example, will not
distinguish between B and N, since presumably clever CIA stalkers who want
to hide their traces will not go around trashing houses like common burglars
looking for money or saleable goods. But B and N do not account equally
well for the dead-bolted doors.
In point of fact, GD sounds a lot like the deceiver hypothesis we philos-
ophers have thought of ourselves as discussing all along. The deceiver in GD
is said to be powerful, hence to have significant ability to bring his wishes
about as well as a desire to make me believe in a real physical world, and the
broadly stated definition of GR is pasted in as the state of affairs the deceiver
wishes to simulate. GD is defined by the prima facie content of the general
class of antirealist deceiver scenarios. In the nature of the case, since such
hypotheses involve the concept of a deceiver, GD will have a connection to
realism. To some degree, the likelihood of evidence given GD will resemble
that given GR, but not to an indefinite extent.
By “treating GR and GD as rivals,” I mean simply this: I propose to
model the evaluation of GR and GD by treating them as if they have prior
probabilities (prior to all empirical evidence) greater than zero and less than
one (without specifying what these prior probabilities are) and as if those
prior probabilities are at least equal to each other, though retaining the possi-
bility that GR has a higher prior probability than GD due to McGrew’s origi-
nal considerations. I will examine the details of my available phenomenal
evidence to see which hypothesis is confirmed by it and to estimate whether
that confirmation is weak, strong, very strong, and so on.6
6
While I will frequently use plural pronouns such as our and we, obviously the
evidence I am describing is actually my own evidence.
GR and GD do not form a prior partition. Even assuming the existence

of one Cartesian subject, there are possibilities incompatible with either of
these hypotheses. It is possible prior to all specific memorial and sensory
evidence that there is an external world, that there is no deceiver, but that
I do not have even approximately reliable sensory or memorial access to
that world. In that case, the world might have any of an infinite or near-
infinite number of different actual structures, but I would not know much
of anything about it. It is possible, prior to all experience, that there is no
deceiver, that there is a real external world, and that I have relatively reliable
access to it, but that it is so wildly unstable that my experience will be like that
of an ongoing chaotic hallucination. It is possible, prior to all experience,
that there is no deceiver and no world external to my mind and that my mind
is just randomly generating images. And so forth.
There is, however, a reason why generic realism and deceiver scenar-
ios (where the latter include brain-in-vat scenarios) have been so commonly
treated as the only players in the epistemic game, and that is quite simply
because the denial of both has zero or near-zero likelihood vis-à-vis even
a portion of the evidence that we in fact have. For example, if I do not have
even relatively reliable access to some real world and there is no deceiver, and
if the real world is quite different from anything that I seem to experience
(e.g., if the real world is made entirely of pink clouds), then my memories of
apparent interaction with an apparently stable and complex world are entirely
unexplained. If the world were wildly unstable (in the absence of a deceiver)
and if I were to have relatively reliable access to it, my present memory-type
experiences would be quite different from what they actually are.
Since we will be looking gradually at a cumulative case concerning
realism, I shall follow philosophical tradition by treating the conjunction
(~GR & ~GD) as being ruled out early on in such a way that it deserves no
further consideration. In other words, we may take it that in our model, if
we were to conditionalize on even a portion of our normal sensory and
apparent-memory evidence, that conjunction disappears from the picture or
comes so close to doing so as not to merit further thought, causing GR and
GD to form a partition at that point (or as near as makes no difference).
Another relevant point is that, just as in the case of the clever burglars,
evidence that confirms GR can simultaneously disconfirm GD even if they do
not form a partition. Evidence does not necessarily do so, even for mutually
exclusive hypotheses (if they do not form a partition), but it may do so, just as
the evidence of the deadbolts on the door both confirms N and disconfirms
B even though N and B do not form a partition. Hence even if we conceive
of ourselves as conditionalizing on all evidence simultaneously, GR can be
confirmed at the expense of GD or vice versa.
lydia mcgrew 125
There is plausibly some generosity built into any model that treats the
prior probability of GD as equal to rather than less than GR. GD already
involves one of the features used by McGrew to argue for a strictly lower
probability for a deceiver scenario than for realism—namely, the existence
of the deceiver over and above the existence of those of his thoughts and
actions that produce the evidence in my mind. I am inclined not to push
too hard on this point, however, because McGrew’s argument depends on
empirical equivalence. The argument for a strictly richer ontology under
a deceiver scenario involves assuming that there are the same number of
evidence-causing entities according to the two hypotheses and that the
deceiver himself is an entity over and above all of these. Since my version of
the argument turns crucially on not treating GD as, by definition, empirically
equivalent to GR, this move is not available to me. That is to say, I am (as
the following argument will show) allowing for the possibility of deceivers
who do not have all of the thoughts and do not engage in all of the actions
necessary to give us the various types and quantities of evidence we actually
have. It does seem intuitively that treating GR and GD as if they have equal
priors is not taking account of the extra ontological burden of the deceiver
himself in GD, whoever he is (or they are), but I am not attempting to model
this point, merely leaving it open as a possibility.
I will argue that, initial appearances notwithstanding, GD is a much,
much poorer explanation of the evidence that we have than GR. That is,
GD gives much lower probability to the evidence we have than GR does.
Therefore, even if we were to treat GD and GR as if they have equivalent,
intermediate prior probabilities “to begin with,” if we treat the denial of their
disjunction as being effectively zeroed out by the evidence, and if we condi-
tionalize on the evidence we have in all its quantity and variety, GR is strongly
justified and has a very high posterior probability.
4. The Face in the Frost
We can begin to get a sense of the likelihood problems that bedevil GD from
a charming but little-known fantasy novel by John Bellairs called The Face in
the Frost.7 The story, set in an imaginary land, pits the lovable good wizards
Roger Bacon and Prospero against the evil sorcerer Melichus. Melichus has
obtained a book of black magic that gives him the power to make what one
might call pseudothings (though the book doesn’t give them a name). Pseudo-
things are remarkably robust apparitions. They are visible, tangible, smellable,
and three-dimensional. While they last, they are subject to intersubjective
7
John Bellairs, The Face in the Frost (New York: Ace, 1978).
verification. But they lack both the perdurance and the full detail of the real
things they pretend to be.
The reader sees the height of Melichus’s power to make pseudothings
when Prospero, temporarily separated from his friend Roger, puts up for the
night in what he takes to be the real town of Five Dials.8 When Prospero
asks an old man for directions to an inn, we get the first hint that something
is not quite right:
The old man pointed his crooked cane toward a shadowy side
street and worked his jaws a couple of times before speaking . . .
“Well, ye’d have yer best luck at the Card Player. Go down that
alley and turn right. Ye’ll see the sign. Mern crost brig.”
Prospero cupped his ear. “What was that last thing you
said?”
The old man looked flustered and shook his head, mum-
bling.
“’S no matter. G’by. Dirks in cairn.”
The old man’s insertion of nonsense phrases is just the beginning. Prospero
finds the conversation at the inn “curiously vague and listless. . . . Everyone
was . . . saying the same thing in different ways.” When he goes up to bed
carrying a candle, something strikes him as odd about the mirror in the hall.
As uneasiness grows in Prospero, he picks up a small box from the table
in his room:
It didn’t even rattle. The heart-shaped brass lock plate on the

front was smooth to his touch. It had no keyhole. He turned the
box over, looking for hidden locks and spring releases, but there
was nothing . . . Why did that mirror bother him? . . . [In the hall, he]
fished his metal matchbox out of an inside pocket and struck a
light. . . . He lit [a candle] and tiptoed . . . to the place where the
mirror hung. Prospero stared and felt a chill pass through his
body. The mirror showed nothing—not his face, not his candle,
not the wall behind him. All he saw was a black glassy surface.
Fighting down rising fear, Prospero went back upstairs and
began to knock on doors, at first softly, then sharply. He tried
the doors. Locked. Locked. And locked. Like the box, the doors
didn’t even rattle. On an impulse, he opened his pocket knife
and tried to slide the blade into the space between a door and its
8
Ibid., 88ff.
lydia mcgrew 127
jamb. The point struck solid wood, for what looked like a crack
was merely a black line.9
Eventually, in a flourish of drama, the inn and the entire town melt unpleas-
antly away, leaving Prospero alone in a field.
The Five Dials scene shows what happens when a deceiver lacks the
power, the creativity, or the continued motivation to make his deceptions as
detailed, realistic, and perduring as mind-independent things. Though Meli-
chus is quite an impressive sorcerer, the illusion can leave gaps that allow the
initially deceived subject to suspect the truth. This class of possibilities will
come up repeatedly in the analysis that follows.
The Face in the Frost also helps us see how evidence for or against real-
ism could be better or worse, stronger or weaker, in any direction, and that
our approach to the realist/antirealist debate needs to have the flexibility to
handle this epistemic fact. The evidence for realism is (intuitively) stronger
if I have clear vision of detailed objects than if I have only fuzzy vision of
indeterminate objects. The evidence for the existence of a deceiver would be
stronger if the deceiver occasionally revealed himself by letting something
lapse. The same is true in the case of the clever burglars. I would be rationally
more inclined to believe my friend’s scary story if I found a suspicious object
dropped on the floor of the house on the morning in question—say, a small
weapon of a kind that no one in the house possesses. This would be some
positive evidence for his theory and hence better evidence than the complete
appearance of normalcy. Conversely, the appearance of normalcy is better
evidence for normalcy if the house is hard to break into than if it is easy
to break into.
Treating GR and GD as our two candidate hypotheses allows us to
model this sort of flexibility. A deceiver who, accidentally or on purpose,
reveals his existence is compatible with GD rather than being excluded from
GD by definition as a failure. So there could be, in theory, evidence for GD
that is better than what we have. Conversely, the evidence we have for realism
can vary in strength, as discussed in the next section. Such an admission of
variation in the strength of the evidence to decide between the two hypoth-
eses would be impossible if we attempted to make realism and deceiverism
empirically equivalent for all evidence.
9
Ibid., 92–93.
5. What Does the Evidence Say?
The evidence that favors GR over GD is difficult to lay out not because there
is any paucity of it but because there is so much of it, and I make no pre-
tense to be displaying all of it out here; time and space would fail to canvass
it thoroughly. Here I will lay out several categories of evidence that favor GR
over GD while asking the reader to extrapolate in order to get a sense of the
quantity of evidence subsumed under each. In many places, to show that I
am bearing in mind that we are not allowed, when describing evidence, to
assume that extramental objects, persons, and events really exist, I will use the
prefix “apparent” to refer to them (e.g., apparent people, apparent objects,
etc.). Where I do not do so, the prefix should be taken as read.
5.1. Detail, Layers,Variety
If you mentally survey what you spontaneously think you know about the
world, what you think you have seen under a microscope or even a magnify-
ing glass or through a telescope, and all the books and articles you seem to
have read describing what other people have supposedly seen at levels you
have never observed yourself, you will find an amazing wealth of detailed
information going up and down the scale from microcosm to macrocosm.
Even if we leave subatomic particles out of consideration, we can talk about
the structure of the atom, the structure of the cell, and the crystalline struc-
ture of sand grains. We have—or think we have—an entire periodic table list-
ing various elements and the ways in which they are distinct. As we move up
above the medium-sized goods around us, we think of planets, the sun, other
stars, and galaxies. All these things are lodged in our present consciousness as
supposed entities and, more important, entities within entities, microcosmic
structures out of which other structures are built.
At a more mundane level, boxes appear to open, as do doors and books,
and a near-infinite variety of objects and words are found inside. A car’s
hood can (it seems) be opened, exposing an engine, which can in turn be
taken apart to reveal the many parts that fit together to make the car run,
with a logical structure that explains (to the mechanically minded) why and
how it works. Buildings appear to have rooms inside, forests appear to be
filled with trees we have not seen at one time but do seem to see at another
time when we go to check, and the oceans appear to be filled with life-forms
that we discover only when we go there or look at videos and photos or read
reports from others who claim to have been there. Our own (apparent) Earth
appears to be, from our individual perspective, filled with vast numbers of
lydia mcgrew 129
places we have never seen but that we find to be there when we travel and to
meet our expectations based on maps and descriptions.
It is important to realize that all of this is vast overkill from the perspec-
tive of GD. Let us grant that some objects, even many objects, outside
ourselves must appear to exist if the deceiver in GD is to be allowed the
power to effect his goals to any interesting extent at all. Nonetheless, there
is no reason whatsoever for him to go this far. The deceiver need not
make apparent things so complicated, so multifaceted, so many-layered, to
serve his purpose. Why should he bother? He could convince us of the exis-
tence of an external world without this lavish display of creativity. And even
if he wanted to make a set of deceptions as exciting, interlocked, and multi-
layered as what our evidence gives us, why assume that he can? Melichus
apparently either can’t or does not want to. That is why boxes in Five Dials
neither rattle nor open, doors don’t either, conversations are dull, and faux
mirrors do not reflect the light of faux candles. Melichus apparently just can’t
keep that many balls in the air all at once. It is much easier for him to make a
world that is, so to speak, all surface.
Consider the same matter from the perspective of GR. Certainly, GR
could be true while the world itself was much simpler than it presently is. In
that sense, we get no positive prediction of a highly detailed world out of
GR either. But what we are concerned with is comparative likelihood. Extra
detail confirms GR over GD just in case the probability of that degree
of detail is greater given GR than given GD. And here there seems to be no
contest. For if there is a real world outside of ourselves and independent of
what someone is trying to make us think at the moment, we would expect
that it has some features with which we are not presently in contact. The
individual subject is not the “measure of all things” if GR is true. And if
that world is relatively stable and we have relatively stable and reliable access
to it, then we should be able to find out about previously unknown features.
They are, in a real sense, waiting there to be discovered. No one has to trick
us by making them merely appear to be there for our benefit. It makes sense,
moreover, that some parts of the world would be smaller than others and
would exist inside of those others, and that we should be able to find out
which ones are which and how they are organized when we go to look. More-
over, the details that we discover appear to hang together (e.g., in the fact
that the underlying properties of matter give rise to the readily visible prop-
erties) in a way that makes sense if there are real objects outside ourselves
that really have this structure. If there are real objects outside ourselves, we
will expect that their underlying structure—whatever it is—will not be causally
independent of their more obvious structure. The inside of the box will have
a certain shape because it is related to the outside of the box. The normally
invisible structure of matter will explain its visible properties because these
are real, stable things. The various aspects of reality are objectively related
to each other. No one decides by an act of will to make us seem to see a lower
layer, which must be made to appear to relate to a higher layer. The different layers
and facets of reality we see appear to be related to each other because they
are, in an important sense, parts of the same thing. Thus the comparative
likelihood of the appearance of detail, and the logical relations among details,
favors GR over GD.
Though some of these categorizations are a bit arbitrary, one may as
well file under the category of “detail” the apparent fact that man is able to
discover laws governing the behavior of matter. Or so my present memories
appear to indicate. Before me I have what appears to be a science book, and
when I open it I seem to read about Boyle’s law: “If the temperature of a gas
remains constant, its volume and pressure are inversely related.”
GD states that the powerful deceiver wants us to believe that we live in
a relatively stable physical world. But once again, allowing us to think that we
have discovered laws by which the various apparent entities in this apparent
world operate is going much farther than necessary for this purpose. Simpler
regularities would do. Attempted scientific experiments, intended to establish
the underlying laws that give us some sort of “why” for the behavior of
the matter we think surrounds us could be met by a kind of experimental
“white noise,” just as Prospero’s knife encounters nothing but wood when
he attempts to insert it between door and jamb. If we take solipsism fully
seriously (which GD requires that we do, except in a Berkeleyan form), the
deceiver has had to invent for me, at this present moment, all of the scien-
tific information that I think I know and all of the apparent other minds
who have taught me or written books, all of the laws that I right now think I
understand, and all of the beliefs I have about the experimental results that
have established those laws.
In contrast, if an external world really does exist and if we really do have
relatively reliable sensory and memorial access to it, it at least makes sense to
ask why the surface regularities we find do hold. Moreover, any true under-
lying laws that do explain the behavior of matter will be at least in principle
available. That is to say, there is no person who could choose not to make them
up and make us seem to find them. A deceiver could choose to disappoint
inquirers, but Nature must give up her secrets.
The third type of detail evidence that tells in favor of GR over GD is
what I might call the van Leeuwenhoek evidence. At first, this part of the
argument might seem to be a restatement of the initial argument from detail:
van Leeuwenhoek believed (or my apparent memories of what apparent
history books say seem to indicate that he believed) that he was not wasting
his time by making an instrument that would allow him to see nature at the
level of the microcosm, and he apparently found something there. But while
lydia mcgrew 131
the van Leeuwenhoek argument up to that point is just a restatement of the

first point—that the deceiver, to generate our evidence, would have to envis-
age microcosmic worlds and convince us that we have had access to them—
it goes beyond that point. For van Leeuwenhoek (seemed to) make his
investigation by using one (apparent) physical object—the microscope—
to investigate other apparent physical objects (micro-organisms, cells, etc.). So
in order to make us believe that we have access to the microlevel of nature in
the way that we appear to have it, the deceiver also has to engage in yet more
unnecessary creativity concerning the apparent properties of glass, the way
it can be used to magnify things, and the like. And the same goes a fortiori for
the electron microscope, for which the deceiver has to get yet more creative
concerning one set of apparent physical objects and the way in which they
seem to enable us to investigate another. This goes beyond simply making us
think that we have some sort of access to the microworld, which could have
been accomplished, for example, by telling us simpler stories about specially
endowed individuals with unexplained supersight (like Superman’s X-ray
vision) that enables them to see germs. Once again, to give us our actual
evidence about the apparent physical world, the deceiver must be ultracre-
ative and ultramotivated to endow this entirely fake world with interlocking
details quite unnecessary to his general project qua deceiver.
Again, we would not necessarily expect given GR that one type of matter
(glass, for example) would allow us to see another type of matter better. But
given GR, a search for such instruments and an attempt to build them makes
sense. Given GR, for example, we have our own physically mediated sensory
access to the world. We even find that it is sometimes better and sometimes
worse and that we can engage in simple actions using our own bodies to
make it better (e.g., squinting or using Galileo’s trick of making a lens with
one’s fingers). Perhaps we can make for ourselves instruments that mimic
this access at its best and that extend it still further. We can at least examine
real materials around us to see if any of them seem to serve that purpose.
The probability is not particularly high a priori that we will be able to do so,
even given GR, but it is much higher than it is on GD. For according to GD,
we will be able to make such (apparent) instruments only if the deceiver has
fairly arbitrarily chosen to deceive us into believing that we are able to do so.
Another interesting aspect of the apparent detail in our sensory evidence
that tells in favor of GR over GD is the fact that we appear to be able to
investigate temporary lapses in the reliability of our senses and to assign
predictable causes for them—fatigue, drugs or alcohol, a malfunctioning
brain—using the resources of the physical world. Thus the GR hypothesis
hangs together even “at the edges,” where our sensory and memorial access
temporarily ceases to be reliable. These are, for the most part, not utterly
unexplained lapses into sensory disintegration.
GD, to be sure, might very well lead us to fear that the apparent external
world would sometimes “come apart at the seams,” like the town at Five
Dials, simply because the deceiver could not sustain it or decided not to
do so any longer. Occasional sensory unreliability may even be said to be
predicted by GD. The problem for GD lies in the apparent explanations within
what appears to be a coherent and plausible real world that encompasses
these lapses. For a deceiver sufficiently interested in his world and sufficiently
powerful and creative to tell us convincing and apparently well-supported
stories (medical and psychological) about occasional lapses in the fabric of
our sensory access would seem to be also sufficiently interested, powerful,
and creative not to allow the lapses to happen in the first place! Melichus
either voluntarily or involuntarily lets Five Dials melt. What he does not do is
produce a smooth-talking apparent-doctor (who doesn’t lapse into mumbling
and gibberish and does not seem to disappear) to explain to Prospero that the
entire Five Dials incident had nothing to do with Melichus but was the result
of a drug in his food at the previous town, an apparent set of detectives to
investigate the poisoning, and an apparent chemist to produce and explain
the drug to the victim.
The “overkill” theme as we compare GR and GD continues when we
consider the fact that we appear to have more than one type of sensory access
to the world and that our various types of apparent sensory access dovetail
together. Sight and touch produce correlative sensations of apparent three-
dimensionality that can be matched predictably after we have experienced
them. Hearing gives us the Doppler effect, which can be correlated with sight
to show objects apparently coming closer, bearing down upon us, and then
moving farther away again. We come to be able to predict certain tastes and
smells and to correlate them with what we see or touch (a particular smell
with what appears to be a skunk, for example, a particular taste with an object
that has both an appearance and a texture that we associate with sugar or a
lemon). All of this is unnecessary from the perspective of GD, and, as with
all these matters of detail, it is by no means built into GD ab initio that the
deceiver will have either the desire or the capability to give us the appear-
ance of multiple senses with these myriad interactions with one another. Sight
alone, for example, would be a good trick to pull off and could lead us to
believe what the deceiver wants us to believe.
On GR, we would not necessarily expect to have this many senses, but
we would certainly expect that whatever senses we do have (since they give
us relatively reliable access to a real and stable world) will work together
predictably as we use them to perceive real objects. Our experience of one
sense, together with GR, enables us to expect congruence with our experi-
ences with another sense, if we find ourselves having other types of sensory
lydia mcgrew 133
experiences (apparently of that same object) at all. The probabilistic impact

of this point will be discussed in a later section.
Similarly, there are far more apparent other minds (persons) all commu-
nicating coherently and in mutually reinforcing ways about what they say are
the same objects than any deceiver needs to conjure up, nor is there reason
to believe that any deceiver will be able to sustain the appearance of this
many other persons and their consistent and detailed interactions both with
me and with the other (apparent) objects in the world. If GR is true, of
course the objects in the physical world, being extramental, can be expected
to be intersubjectively perceptible, so it would be expected that multiple
subjects, if they do exist, will be able to interact both with them and with
each other.
5.2. Predictability and Perdurance
There is no question that our own sensory experiences appear to us to be

experiences of perduring and predictable objects. I have an experience at the
moment of a white shape in front of me, and I seem to remember not only
that this shape is called a “computer desk” but also that I have had many,
many other experiences of and types of interaction with this same object.
And the same is true for the whole rest of the world of apparent objects that
we seem to remember having encountered or presently seem to be encoun-
tering: we remember their being apparently solid and perduring, not sim-
ply fading away, and we seem to remember huge numbers of occasions on
which our expectation of their perdurance has allowed us to predict their
behavior and to use them in known or learned ways. We seem to ourselves
to have gradually built up more and more of such experiences over a period
of years from what seem to us to be our earliest memories of young child-
hood onward.
We have, moreover, experiences of what appear to be reports from
other people telling us of their interactions with the physical world, telling
us of their own experiences of the same things we now encounter at other
times, or telling us about when and how some new object or objects came
into existence—for example, the planting of a tree or the building of a new
neighborhood. The apparent reports to which we have access seem to indi-
cate that people have found an external world all around them for many
thousands of years. It is a world that changes, to be sure, but it changes in
ways that make sense and that appear to be explicable in terms of anteced-
ent causes—the actions of other agents, animals, or physical regularities.
Trees grow; houses are built. Trees are chopped down; houses fall; rocks are
gradually weathered away. Bodies decompose, and from the earth to which
they return new plants grow that, in turn, are consumed by creatures and
contribute to the growth of their bodies. Or so it seems in the world as I now
seem to be in contact with it and to remember it.
While there is no problem at all in accounting for these appearances
on GR (given that it includes the stability of the external world), GD yields
a much greater degree of causal independence among the various appar-
ent instances of our contact with the physical world. The deceiver must be
able and willing to sustain the apparent physical world through moment after
moment of apparent time; or, to put the matter the same way, he must be
able and willing to give me the mountainous quantity of present apparent
memories—memories that appear to have been built up during years of my
own life—of the ability to predict the continued existence of objects and
of encounters with the same objects. He must also have chosen to give me
the huge quantity of apparent reports from others of such encounters with
a relatively stable world spanning what are said to be thousands of years
of time, including descriptions of many scientific experiments that depend
crucially upon the expectation that physical objects will not suddenly fade
away and will continue to act in a predictable fashion. In inventing all of
this, the deceiver must not grow bored, must not change his mind, and must
desire to give me all of this, for no other reason than the incredible, super-
erogatory verisimilitude of his deception.
5.3.Verisimilitude
The concept of verisimilitude is relevant to the issue of likelihoods. As

Laurence BonJour points out, it seems clear that what he calls a “quasi-
commonsensical hypothesis” (which resembles GR) is a “relatively adequate
explanation of the details of our sensory experience.”10 What BonJour calls
a “digital” explanation of our experience (a deceiver hypothesis) explains
our experience by “some agent or mechanism that produces experiences in
perceivers like us in a way that mimics the experience that we would have
if the represented world were actual and we were located in it, even though
neither of these things is in fact the case.” In other words, the very notion
of a deceiver presupposes that the evidence we have is well-explained by some-
thing like GR. Otherwise, why does a deceiver produce these sensations rather
than others? This point is particularly clear when it comes to perdurance and
predictability.
Laurence BonJour, “Foundationalism and the External World,” Noûs 33

10
(1999): S13, 244.

lydia mcgrew 135
One could even go further: plausibly, if the skeptic were right in his
concept of what rationality requires (i.e., epistemic neutrality or something
extremely close to it concerning realism), it would not be meaningful to say
that the deceiver is producing experiences “as if ” of a mind-independent
external world or experiences that “look like” such a world. According to the
skeptic’s strictures, a truly rational subject would not believe that these appear-
ances look like an external world, for a truly rational subject would, after
receiving such experiences, be neutral on whether there is such a world. He
would take those experiences to look like (at most) realism-or-deceiverism.
The deceiver can deceive successfully only if we associate these experiences
with a stable, mind-independent world as opposed to associating them with a
deceiver, but given skepticism, this association is inexplicable in any objective
fashion. Should the skeptic say that we just happen to be irrational in such a
way that we find ourselves thinking that these experiences look like a real world?
If the deceiver has so much control over our thought processes, why could
he not manipulate us to think that vague experiences of amorphous sensa-
tions are like the existence of an external world, thus saving himself a lot of
trouble? It seems as though, in order to say that our experiences really are like
a mind-independent external world, the concept of verisimilitude must be
given some content such as that such experiences make it reasonable to think
that there is such a world. But that is precisely what the skeptic denies; deny-
ing it is of the essence of his position.
This point about the meaninglessness, given skepticism, of any norma-
tive idea that our experiences look like an external world (which is why the
deceiver produces them) brings us back to the piggybacking issue discussed
earlier. I have already argued that a deceiver hypothesis that is by definition
empirically equivalent to realism is indefinite in content and hence cannot be
a real hypothesis with a real probability. The same parasitism upon realism
means that the skeptic’s strictures against realism undermine the notion of
verisimilitude that is indispensable in describing the deceiver’s own activities
and motives.
In all the areas discussed so far—levels of detail, apparent ability to
discover laws, the use of apparent physical objects to investigate others,
appearance of multiple, mutually confirming senses and multiple, mutually
confirming reports from subjects, perdurance and predictability—it would
be difficult to overestimate the sheer quantity of our data in favor of the
external world. Indeed, despite the attempt here to itemize it, it would be
easy to underestimate it. It is only by even beginning to think about it, item
by item, that we begin to realize how much evidence there truly is and of
how great a variety. Every single story that every single person has ever told
you about any event whatsoever must be an invention of the deceiver if no
external world exists. Everything you have ever read in any history book,
as well as the book itself, must be his invention. All the apparent external-
world items encountered—in all their apparently multifaceted natures, in all
your memories of all your daily activities, and in all anecdotes and historical
incidents—were themselves the mere inventions of the deceiver. Day after
day, moment after apparent moment, you are interacting with what appear to
you to be extramental objects and are finding that they stand up to the test.
Apparent year after apparent year, your experience is going on without any
sudden appearance that the external world or any of the objects in it have
unaccountably wavered and melted away.
5.4. Independence
To understand what the evidence says and to see the previously stated argu-
ment most clearly, the issue of independence must be brought to the fore.
It may or may not be possible to atomize our bare sensory experiences as
they are experienced by (say) an infant or a drugged person—a patch of red
here, a meaningless sound there. Such atomic pure sensations may have little
to no evidential value taken individually in favor of GR as opposed to GD.
But once we gain even a little bit of inductive experience, and certainly by
the time we are (or seem to ourselves to be) children of several years old,
any such sensations are simultaneously accompanied by a wealth of seem-
ing memories. These include memorial data that associate sensations with
other types of sensation (sight with expected touch, for example), familiarity,
apparent perdurance, what we seem to remember having heard other people
say, and so forth. As most of us actually experience it, a single sensation is
part of a far more complex composite, synchronic experience, the parts of
which are related in various ways.
It is these relationships that give the greatest force to the evidence in
favor of GR. As I have shown elsewhere, the force of a set of evidence
{E1, . . . , En} vis-à-vis a hypothesis H can be expressed as the product of the
individual Bayes factors for each of the Ei and a correction factor showing
their relative probabilistic dependence given each of H and ~H, as follows:
P ( E1 | H ) P( En | H ) P( E1 &...& En | H )/P( E1 |H ) × ... × P( En |H )

× ... × ×
P( E1 |~H ) P( En |~H ) P( E1 &...& En |~H )/P( E1 |~H ) × ... × P( En |~H )
The terms on the left of this expression are the individual Bayes factors. These
express the force of the individual items of evidence concerning some
hypothesis H, treated as if they are independent. But what if they are not
independent? Then the term on the right comes into play. The term on
the right is a correction factor, which I have dubbed a measure of “relative
lydia mcgrew 137
consilience.”11 This is a “correction factor” in the sense that it shows how

much weaker or stronger the case really is than it would be if all of the items
of evidence were completely independent of each other given H and ~H,
respectively. For example, if we have several witnesses to some event and
also have serious reason to think that they may be colluding, and if H is the
proposition that what they all attest to is true, then the term on the right will
be bottom-heavy, partially sapping the force of the case as it would be if they
were independent. But it is also possible for dependence considerations to
favor some H over ~H.
The considerations discussed in earlier sections indicate that the relative
consilience factor for GR vis-à-vis GD is strongly top-heavy—that is, that
the evidence items are much more probabilistically dependent given GR than
given GD and that the case for realism is therefore stronger than it would be
if the items of evidence are merely considered individually.
As already noted, GR and GD do not form a partition, so GD is not
identical to ~GR. While the relative consilience ratio can be estimated for
two hypotheses H1 and H2 that do not form a partition, as can individual
Bayes factors, there is no guarantee in that case that a top-heavy Bayes factor
or RC factor indicates confirmation of the “winning” hypothesis of the two
in the Bayesian sense that it is confirmed by the evidence relative to its own
negation. However, I have addressed this by suggesting that the catchall,
which is the negation of both GR and GD, gives a probability of near zero to
even a portion of the evidence we have, much less all of it. If this is correct,
then, where GR is the salient hypothesis H,
P( E1 &...& En | H )/P( E1 |H ) × ... × P( En |H ) P( E1 &...& En | GR ) /P ( E1 |GR ) × ... × P( En |GR)
≈
P( E1 &...& En |~H )/P( E1 |~H ) × ... × P( En |~H ) P( E1 &...& En |GD)/P( E1 |GD) × ... × P ( En |GD)
To see in more detail how the RC factor produces confirmation of GR

vis-à-vis GD, consider some ways in which items of our evidence are more
causally independent given GD than given GR. Causal independence, of
course, is not the same as probabilistic independence, but when we realize
that two occurrences are causally independent, this consideration generally
(though not invariably) favors treating them as probabilistically independent.
If GR is true, different features of our experiential and memorial
evidence are not all independent either within each (loosely conceived) cate-
gory we have discussed or between categories. For example, various levels of
See Lydia McGrew, “Accounting for Dependence: Relative Consilience as a

11
Correction Factor in Cumulative Case Arguments,” Australasian Journal of Philosophy

95 (2017): 560–72. Lydia McGrew, “Evidential Diversity and the Negation of H:
A Probabilistic Account of the Value of Varied Evidence,” Ergo 3 (2016): 10.
detail in physical reality are not physically independent of each other; rather,
more readily observable levels have their features at least in part because of
the features of the less readily observable levels, which allows us to infer, in
relation to our experiences of regularity, the continuation of the less readily
observed physical traits that underwrite the observable traits. The various
senses fit together in a regular fashion because they are all in contact with
the same relatively stable external reality. The language, gestures, and actions
of other people interact smoothly both with physical objects and with each
other because the physical objects (including people’s own bodies) are real
and accessible to all. Our multiple senses appear to confirm the various levels
of detail in the objects about us because the objects are real and our senses
are really encountering them at various levels.
GD, on the other hand, introduces a much higher degree of indepen-
dence, even (I am strongly inclined to think) an irreducibly higher degree
of independence, among all of these features than does GR because of
the ongoing possibility that the deceiver’s desire or ability could stop at some
point. The deceiver must be able to create all of these impressions and must
desire to do so, and either his ability or his desire might (for all we can tell
from GD itself) fail at almost innumerable points along the way before we
reach the level of variety, detail, and interconnectedness that we find in the
evidence we actually possess.
One can attempt to reduce the apparent arbitrariness in GD by think-
ing of the deceiver as conceiving of (and presenting to us) imaginary wholes,
entire objects (tables, chairs, plants, insects) conceived in great detail and at
many levels of reality, whose parts are interrelated in the deceiver’s mind in
exactly the way we imagine the parts and levels of real objects to be, rather
than conceiving the objects merely in terms of their obvious surface prop-
erties. But, while this reduces the independence among the parts of such a
pseudo-object in any given case, it does so only for that particular object, and
it does so, moreover, at the cost of imagining a deceiver able and willing to do
so even for that particular object, when going that far is by no means neces-
sary for his goals qua deceiver. To imagine a deceiver even capable of inventing
pseudonature and other embodied agents as we seem to find them is to imag-
ine a deceiver at least very nearly omnipotent, omniscient, and omnicreative
(without being omnibenevolent). And once we have said that, we still have
left a significant amount of arbitrariness in the deceiver’s decision to use
those powers to produce so elaborate a hoax.
A couple of miniature examples give an idea of how the relative consil-
ience factor works to confirm GR over GD and also how memorial expe-
riences associating sensory experiences confirm GR over GD. Suppose that
I have a visual experience such as I usually associate with an apple. Simul-
taneously I have a memory experience that I generally associate this visual
lydia mcgrew 139
experience with a kind of fruit that can be touched, that produces a tactile
sensation like that, and that also produces an olfactory sensation like that. I
seem to myself to reach out and touch the apple and smell it, and indeed
I simultaneously experience the apparent memory of past experiences like
this, the expected tactile sensation (and the apparent memory of having just
predicted such a sensation), and the expected smell sensation. GR states
that a mind-independent external world exists and that I have fairly reliable
sensory and memorial access to this external world. Given my visual expe-
rience, my apparent memory of a coordinated tactile and smell experience
connected with this type of visual experience, and GR, I have some reason
to expect that tactile and smell sensation. Suppose too that I have a more
general apparent memory that visual sensations tend to be associated with
tactile sensations and a more general apparent memory that visible objects
sometimes have smells as well. (This is aside from my apparent memory of
the specific tactile and smell sensations that I associate with the apple-type
image.) These items of my experience (where apparent memory is included)
are fairly strongly probabilistically dependent given GR, which is to say that
the ratio
P( E1 & E 2 & E 3 & E 4 & E 5 & E 6 | GR )

P( E1 | GR) × P( E 2 | GR) × P( E 3 | GR) × P( E 4 | GR) × P ( E 5 | GR) × P( E 6 | GR)
is significantly top-heavy, where E1, E2, and E3 are the sensations, E4 is the
apparent memory of the connection among the three of them in the past,
E5 is the apparent memory of a general connection between sight and touch,
and E6 is the apparent memory of a general connection between sight and
smell. RC is a ratio of ratios, and this is its numerator.
What about the denominator of the RC ratio showing the unification of
these same items of evidence by GD? Suppose that one had the visual sensa-
tion (E1) and the apparent memories (E4, E5, and E6). Would this provide
as much reason for expecting the tactile and smell sensations (E2 and E3) as
does GR? Even if the deceiver desires to convince me that I have relatively
reliable access to a mind-independent world, and even if he has bothered to
give me these apparent memories correlating my senses (both generally and
specifically), his ability to deliver on that expectation might fail. Melichus
presumably knows that Five Dials, if it were fully real, would not dissolve,
but he lets it dissolve nonetheless. Or the deceiver’s attention might wander,
or he might change his mind and decide to let my experience be less unified
in this particular instance, to allow my experience to be less “like” that of a
real, mind-independent world than it would be if my expectations of touch
and smell were fulfilled. This is a far more live option given GD than a similar
failure is given GR.
Moreover, the argument made earlier suggests that E4, E5, and E6 each
have individual Bayes factors that are top-heavy in favor of GR, since a
deceiver might well not bother in the first place to give me all three apparent
senses of taste, touch, and smell and to give me apparent memories that
coordinate them. In contrast, if I have multiple senses at all given GR, they
would plausibly coordinate as part of my reliable sensory access to a real
external world. They would not generally be in conflict unless there were some
explanation such as specific hallucinations, drugs, and so on.
Another simplified case illustrate a similar point. Suppose that it seems
to me that multiple people who do not know one another and are not all
following the same source account tell me that they have seen Stonehenge.
Or suppose that I have one apparent in-person account of someone’s seeing
Stonehenge and one apparent report in a book of a different person’s seeing
Stonehenge. Obviously this case raises the point that GR allows me to have
reason to believe in the existence of other minds by inductively observing
their behavior, while GD does not. But to simplify further, consider for the
moment just the two apparent reports, R1 and R2, including the knowledge
that they attest to particular contents (such as the existence of Stonehenge),
and the RC ratio. R1 and R2 are not independent of each other modulo GR. If
GR is true, an apparent report from one person of the existence of Stone-
henge should lead me to expect more strongly than I did before that I will
receive another such apparent report. GR states that there are relatively stable,
mind-independent objects in the world and that I have relatively reliable
sensory and memorial access to them. Therefore, given GR, the first report
can give me some reason to believe that one of the real things in this relatively
stable world of intersubjectively accessible objects is, in fact, Stonehenge.12 If
GD is true, however, there is far less reason to expect more than one report
of Stonehenge than there is given GR, as the deceiver might not choose to
make it seem to me that more than one person has seen Stonehenge and
reported it. Since there is no real Stonehenge to be seen, it is entirely up
to the deceiver to decide how much apparent evidence to give me. And of
course he must be able to produce multiple apparent reports. Given GD,
some other person in the world cannot simply come upon Stonehenge and
choose to report it, regardless of the deceiver. The deceiver has to produce
everything. So R1 and R2 are more independent given GD than given GR, and
the RC ratio favors GR.
Depending on other information, R1 and R2 may be independent of each

12
other modulo the proposition “Stonehenge exists” and its negation. Multiple reports
may be independent given a specific proposition and its negation, while they are
dependent given a broader proposition such as GR.
lydia mcgrew 141
The power of GR to unify data as compared to GD is an important part

of the confirmation of GR.
6. Foundationalism and the Problem of the Priors
In the end, there is no putting off the problem of the priors indefinitely.
If one is not a Bayesian personalist, the question of where prior probabili-
ties come from has a fairly straightforward answer for the great majority of
inferences: prior probabilities for most inferences come from all of one’s
other concrete evidence—other, that is, than the evidence on which one is
conditionalizing at the moment or deeming oneself to be conditionalizing
for purposes of some given argument. From this perspective, the distinction
between the justification of prior probabilities and the conditionalization step
using likelihoods does not mark a gulf in the nature of justification between
subjective and objective probabilities. It is a desire for epistemic or analytic
clarity, or perhaps a matter of literally having newly received some evidence
at a given point in time, that leads one to group empirical evidence into that
which acts as background information and justifies the prior probability and
various likelihoods on the one hand and that on which one conditionalizes to
obtain a posterior on the other.
But no such analysis will do at what we might call the very beginnings
of inference, for those propositions (such as GR) so sweeping and so vital
that they serve as background to our other inferences about both everyday
matters and high-level scientific theories. But can we say anything definite
about objective prior probabilities of GR and GD, where “prior” means
“prior to all sensory and apparent memorial data”? How are such priors to
be understood in relation only to, say, conceptual truths and the direct Carte-
sian knowledge of the existence of one’s own mind—that is, what is available
other than the vast quantities of sensory and memorial data that we wish to use
to find their posterior probabilities?
A special problem arises here for the classical foundationalist, since
that position entails that no proposition has an intermediate probability—a
probability between zero and one—except in relation to some other evidence.
There are, on this view, no such things as propositions that are intrinsically
probable without being intrinsically certain. Nor are there intrinsically improb-
able propositions. All intermediate probabilities exist in virtue of the relation
of the proposition in question to some other known propositions.13 More-
over, the only propositions that can be regarded as foundational are such
in virtue of a form of special access the subject has to those propositions,
13
Timothy McGrew, Foundations of Knowledge, 60–65, 70–72.
which gives him certainty about them. In this way, the classical foundation-
alist is distinguished from the moderate foundationalist, the direct realist,
and the externalist. Obviously, neither GR nor GD satisfies any criterion for
being foundational in such a scheme; the entire discussion here would be
unnecessary if either of them did.
Rather surprisingly, though McGrew defends this type of foundation-
alism, his own argument concerning the differential prior probabilities of
realism and deceiverism, discussed at the outset, suggests the contrary. For
those prior probabilities must be construed in a probability distribution when
empirical data are not yet known. That is the point of their being prior proba-
bilities in the realism/deceiverism debate. The mere fact that we are consid-
ering the prior probabilities of those propositions because we are ultimately
interested in explaining the evidence we now have does not mean that the
value of the priors (nor even their comparative value) is set by the empirical
evidence. To the contrary, these must be “absolute” priors or “ur-priors,” yet
McGrew argues from ontological considerations that realism has a higher
prior probability than does deceiverism.
While it may be small comfort to the classical foundationalist, it is worth
stressing that, if we did regard the prior probabilities of GR and GD (and
of the catchall) as objectively real, these would not be (by my modeling here)
high probabilities. My model has treated GR and GD as if they have prior
probabilities that are at least equal to one another (though that of GR might
be higher than that of GD), not as if either of them is high in some stron-
ger sense. They are, in fact, both sufficiently complex propositions that it
seems quite legitimate to treat both of them, prior to all concrete evidence,
as having low probabilities. Therefore, allowing intermediate probabilities to
exist relative only to conceptual truths would not serve the usual purposes of
moderate foundationalism—that is, justifying our ordinary knowledge aside
from inference—since both GR and GD would presumably start out with
absolute prior probabilities less than .5. This would merely act as a basis for
beginning to conditionalize on all of the sensory and memory-type evidence
we actually have. A typical Moorean proposition like “I have a hand” would,
on such a view, not have some intermediate high probability without infer-
ence. In the absolute prior distribution, it would have a low intermediate
probability and would acquire a high probability only relative to foundational
evidence of the standard type that the classical foundationalist is interested
in—qualia, memory-type experiences, and the like, such as I have been
discussing throughout this essay.
The discussion so far suggests something like Richard Swinburne’s
notion of real, absolute prior probabilities for propositions, ranked accord-
ing to complexity, so that something like “That is a planet” has an abso-
lute prior probability higher than “That is 387 small planets with a common
lydia mcgrew 143
center of gravity.”14 Given the greater independence of the evidence given

GD than given GR, as was just argued, a highly elaborated subhypothesis
of GD that gave equal probability to our existing evidence to that of GR (if
such a thing could be found) would be strongly disfavored in a complexity-
sensitive ur-prior distribution because it would be significantly more complex,
requiring multiple assertions of the deceiver’s continued ability and desire to
produce the evidence.
It is understandable that philosophers might balk at accepting the exis-
tence of such a vast ur-prior distribution in which every possible empirical
proposition has a specific, objective probability relative only to conceptual
truths and to whatever a Cartesian subject could know without having any
concrete empirical experience. For one thing, it seems like the number of
propositions in such a distribution might be infinite, which is counterintu-
itive. The construct of such a prior distribution seems rather artificial. It
implies that there is a real answer to the question of what an ideal subject
would say about the probability of a proposition such as GR and millions
of other propositions—indeed, that he would be able to give them specific
probabilities—if he had no evidence on the topics in question at all other than
considerations of relative simplicity.
Another possibility is to regard the model here as involving a normali-
zation of the likelihood function over a threefold partition—GR, GD, and
(~GR & ~GD).15 If we think of it in this way, we can think of the likelihoods
as having maximal effect with no requirement to assume real prior proba-
bilities. In effect, this acts like a model in which GR, GD, and the catchall are
given equal prior probabilities and in which this prior distribution is maxi-
mally sensitive to subsequent data. But in fact one does not assume that there
are real probabilities in a prior distribution; rather, one turns the likelihood
function on the three possibilities into a posterior probability. My contention
has been that, if one does this, the catchall will (in essence) disappear and GR
will end up with a very high probability and GD with a very low probability.
This is somewhat different from the prior model in Swinburnian terms, both
because the Swinburnian model treats the ur-prior probabilities as real and
14
Richard Swinburne, Epistemic Justification (Oxford: Clarendon Press, 2001),
82–99.
15
The concept of normalizing the likelihood function for some set of evidence
and a set of hypotheses that form a partition has been suggested by the objec-
tive Bayesian Roger Rosenkrantz, Foundations and Applications of Inductive Probability
(Atascadero, CA: Ridgeview, 1981), 4.4–13. Rosencrantz implies that the concept
originated with Sir Harold Jeffreys, Theory of Probability (Oxford: Clarendon Press,
1948), 102, but Jeffreys speaks in terms of uninformative equal priors rather than in
terms of normalizing the likelihood function.
also because the Swinburnian prior probability of the catchall would presum-
ably be much higher than that of either GR or GD, given that it contains
metaphysically simple possibilities such as that no physical objects exist, no
deceiver exists, and nothing is making it look like any physical objects exist.
That, however, becomes a moot point when the priors are swamped.
A third possibility, rather similar to the second, would be to incorporate
simplicity and complexity considerations into a nonnormalizable prior distri-
bution that is not a probability distribution at all but that ranks a possibly
infinite number of propositions according to their relative complexity. This
idea, which remains relatively nonspecific, is that such a distribution (as in the
case of normalizing the likelihood function) would be maximally sensitive
to data and would yield a posterior probability when evidence comes in, in
accordance with the relative complexity of the possible explanations.16
The skeptic will be hard-pressed to argue that any of these ideas is unfair
to his position per se. Can he argue that GD has a real prior probability and
that it should be regarded as much higher than that of GR so that it can handle
all the disconfirmation I have argued for while still ending up at least equal
to GR? It is extremely difficult to see how one could go about making such
a case.
The skeptic’s better option would be to attempt to state that none of
these ideas is satisfactory, that we should throw up our hands in despair over
the problem of prior probabilities, and that his challenge is unanswerable
for that reason alone. If we cannot even begin to model a Bayesian or quasi-
Bayesian inference by making any sensible model of the situation prior to all
concrete evidence, then there seems to be no point in talking about GD as
“disconfirmed” or about GR as “highly confirmed” by the evidence.
I stated at the outset that my argument would show that realism is highly
confirmed if rational empirical inference is possible at all, and that point
comes back into play here. For if we despaired in that way, any inference lead-
ing to a high probability on the basis of empirical evidence would become
impossible. Nor would this be a result of the fact that we could not then
be justified in believing realism per se. Such an alleged problem would apply
even if realism were assumed to be true. There would be no way to say,
given realism, that an elaborate conspiracy theory was highly unlikely and a
rival, nonconspiracy theory highly probable, assuming that one could gerry-
mander the conspiracy theory to the point that it would “account for” the
evidence we currently have just as well as some rival, nonconspiracy theory,
pushing the issue back to prior probabilities. Or one could simply assert that
we have no coherent way to discuss the prior probabilities of the conspiracy
Something like this has been suggested by Timothy McGrew, personal

16
communication.
lydia mcgrew 145
and nonconspiracy theories. Then one could declare that, since the problem
of the (absolute) priors is insoluble, there is no way to tell which is correct
in the posterior distribution when evidence is taken into account. Such
despair, then, is really just another version of Trumping Empirical Equiva-
lence. If we do not allow rationality to be undermined by Trumping Empiri-
cal Equivalence, we can let the evidence speak.
Conclusion: A Note on Direct Realism
Despair of defending the propositions contained in GR on a more funda-

mental basis creates a significant motivation for the push in epistemology
to declare such propositions “basic” or to take them to be intrinsically
and directly justified in some way that does not involve even tacit rational
inference.
But once we see that the tacit principle on which the skeptic’s opposition
is based, Trumping Empirical Equivalence, would rule out the rationality of
inferences even after antirealism is dismissed, the Moorean or pragmatist has
just as much need as the foundationalist to point out that Trumping Empir-
ical Equivalence is irrational not just pragmatically but epistemically. Once
that is understood, the possibility opens up for justifying belief in the exter-
nal world “from scratch.”17
It is certainly true that we find ourselves believing in the external world
and trusting our own senses spontaneously and continuously, and if one
takes it that spontaneous belief and overwhelming trust are incompatible
with rational inference, one will naturally have a problem with the attempt to
justify such beliefs inferentially. But what if we reject the opposition between
spontaneity on the one hand and rational inference on the other? What if,
instead, we accept a robust notion of “inference” that allows rational infer-
ence to be incredibly rapid, spontaneous, and inexplicit—even to take place
nonlinguistically? In that case, we should be open to the following possi-
bility: these beliefs are so spontaneous, so pervasive, seem so obvious, and
act as a background to so much else we believe because they are so heavily
overjustified—hyperrational, one might say. It might seem surprising that over-
whelmingly justified propositions should be believed in such a way that they
appear (to some) to be held arationally, yet it is not surprising upon reflec-
tion. Where a tacit inference is justified by so great a wealth of data forming
so many interlocking and converging lines of evidence, and where a great
17
Contrast this argument with the move to pragmatism by Eric Olsson, Against
Coherence (Oxford: Clarendon Press, 2005), 173ff., after arguing that it is not possible
to justify realism “from scratch.”
quantity of that data first comes to the subject at a very early stage of his
existence, it is understandable that he should take the inference for granted,
find it nearly impossible to doubt, and find it difficult or impossible to artic-
ulate his reasons for it. This need not be a sign of the absence of good
reasons; it may be a sign of excessive amounts of evidence.
This view of the matter satisfactorily accounts for the fact that we would
consider a person insane if he seriously did doubt the existence of the exter-
nal world. It accounts for the fact that it seems that we cannot get along in
the world—as indeed we cannot—without assuming GR to be true and even
without assuming other things, such as the existence of other minds, not
contained in GR but readily justified on the basis of GR and of our own
specific evidence. To all of these points, the classical foundationalist can say
this: Yes, you are right. Yes, there is a sense in which we are compelled to believe
these things. Yes, it would be insane seriously to doubt them. Yes, we do
constantly believe them without stopping to worry, wonder, and tease out our
reasons for so doing, much less the metalevel explanation of the cogency of
those reasons. Let us go on doing so. And let us go on doing so because what
compels us in these beliefs is reason itself—reason that has been given so much
material to work with, material that is interwoven so beautifully and with
such complexity, that the reason suffers from an embarrassment of riches
and scarcely knows how to explain the matter to anyone who would doubt.
To us as philosophers is given the odd and off-putting task of creating
problems where the ordinary man, understandably enough, sees no problem.
But if we do so, let us also take on the task of trying to answer those ques-
tions thoroughly. Thus we can be both philosophers and real men living in
the real world, seeing no conflict between the two.
Fine-Tuning and the Search
for an Archimedean Point
Timothy McGrew
Western Michigan University
One of the benefits of working in the philosophy of religion today is the

chance to engage with genuinely new empirical arguments. The fine-tuning
argument, or FTA, ranks among the most intriguing of those arguments.
Although it is in a broad sense a version of the design argument, which is not
new, its structure and its empirical grounds are thoroughly modern and have
attracted the attention of philosophers, physicists, and mathematicians alike.
I am, at the moment, a reluctant and somewhat tentative skeptic regard-
ing the argument’s force. But I am not immune to its intuitive appeal. In this
paper I will sketch the argument as it is commonly discussed, explain where
in my view some key difficulties lie, and then engage with a recent thoughtful
attempt to circumvent those difficulties.
The Intuitive Picture and the Underlying Math
In your mind’s eye, imagine a single speck of dust floating in the air in a vast
cathedral. The mote is tiny, though not actually a mathematical point. Focus
on that speck, and then zoom out. For practical purposes, the speck vanishes;
it is lost in the immensity of the space around it. Once we lose sight of it, we
have little hope of finding it again. If we were asked to specify its location
with a random guess—say, while we were blindfolded—our task would be
hopeless.
This mental picture, or something like it, is a fundamental intuition pump
for the FTA; we need only change a few terms to exhibit the parallel. Instead
of the physical space of the cathedral, we are speaking of an abstract space
in which the values of certain physical constants can be specified. Instead
of a physical mote, we are looking at the region of that abstract space in
which the constants permit the existence of organic life in a recognizable
form. And it turns out that, on physical grounds, the life-permitting region
is small. The set of values picking out the constants in our own universe is
in it, obviously, for here we are. And within a very limited region around
© Timothy McGrew, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)

148 fine-tuning and the Search for an archimedean point
those constants, life could theoretically flourish—say, in a universe like ours,

but with a tiny increase in the gravitational constant. But the “ball” of such
values, like the dust mote in our analogy, is dwarfed by the space of life-
forbidding universes, as small variations in any of the constants (singly or
jointly) prove fatal to the possibility of life. Zooming out, therefore, we find
that the ball of life-friendly values is quickly lost to view in an immense space
of hostile, sterile universes that consist (for example) of nothing but black
holes or nothing but diffuse hydrogen. What are the odds, then, that the actual
universe would have just those values that permit life, merely on chance?
To make the intuition behind the FTA more precise, we have to connect
these intuitions to probabilities. The inside of the cathedral is the space of
logically possible universes, and the entire interior volume corresponds to
100 percent. The volume of the speck of dust corresponds to the ball of life-
permitting universes. The ratio of those two volumes—the speck divided by
the space—will stand for the probability that a randomly selected (or gener-
ated) universe will be hospitable to life. This ratio is, so to speak, the Archi-
medean point of the argument. If we can define it carefully and defend a
particular value (or even an approximate value) for it, then we have some
hope of moving forward; if we cannot, the argument fails.
To complete the argument, we need one more piece of information—or
two more pieces, depending on the conclusion we want to draw. First, we
need to know how that first small probability compares to the probability that
there would be a life-friendly universe, supposing that the physical constants
are not selected at random. It is not necessary, in order to make a case against
randomness, that we have an exact number for (say) the probability that a
deity would create a life-friendly universe; it is sufficient if we know that it
would be many times more likely if the universe is the product of divine
intention than if it were (in a probabilistic sense) a random accident.
We are still sliding over some contested points here. For one thing, it
requires an argument we have not provided to identify not-by-chance with
the theistic hypothesis or chance with atheism. Let us suppose, for the
moment, that this gap can be closed. Then, if all is well with the intuitive
version of the argument that we have sketched, we would be entitled to say
that the life-friendliness of the universe is evidence for theism. But to make
the stronger claim that theism is, all things considered, more probable than
not, we would also have to know the probability that there is a deity apart
from this specific kind of evidence. And the discussion of that wider issue
fills libraries.
timothy mcgrew 149
A Mathematical Difficulty
Even for the more modest version of the argument, there is a hidden prob-
lem. When we set up the intuitive case, we used the interior of a cathedral as
an image; and however vast the cathedral might be, its volume is unambigu-
ously finite. But in the case of the possible values of the physical constants,
this condition is not obviously met. How strong, for example, could the grav-
itational constant be—even assuming that it is positive? It would appear that
it could take any value greater than zero, without an upper bound. True, there
might come a point at which the possible universes generated by holding
other things equal and increasing the value of G would become, for practical
purposes, empirically indistinguishable. But the intuition behind the FTA has
nothing to do with how these universes look. If even one of the constants can
take on any positive value whatsoever, then even if others are neatly bounded
within finite ranges, the volume of the abstract space of possible universes
is infinite.
Infinity, unfortunately, does not play well with finite numbers. The
measure of a finite ball in an infinite space of the sort we are considering is
not merely small; within the ordinary real number system, it is strictly zero. If
we take that measure for a probability, we are starting with a probability of a
life-permitting universe of zero. Standard models of updating our probabili-
ties do not allow us to use propositions with probability zero for condition-
alization. Like division by zero, the operation is undefined.
A parallel problem arises when we ask what happens, in an infinite space,
if we expand the ball—if we suppose, for the sake of argument, that the set
of life-permitting universes were many times greater than in fact we believe
it to be. So long as the ball is finite, it still has measure zero on the infinite
background space. The awkward consequence is that, if the FTA worked,
so would a coarse-tuning analogue, which we can call the CTA, in which the
life-permitting values of the constants were billions of orders of magnitude
greater than, in fact, they are. The two arguments would be equally strong. But
the CTA plainly lacks the intuitive force of the FTA. If our analysis suggests
that they are equally good, something has gone wrong.
Finite and Countable Additivity
In a clear and thoughtful paper, Jeff Koperski raises the stakes in the argu-
ments over cosmological fine-tuning.1 His contention, put briefly, is that the
1
Jeff Koperski, “Should We Care about Fine-Tuning?,” British Journal for
the Philosophy of Science 56 (2005): 303–19.
pattern of inference at work in fine-tuning arguments is endemic to science

and that the rejection of the fine-tuning argument on formal grounds would
entail the defenestration of much work in statistical mechanics, cosmology,
and chaos theory.2
That is indeed a serious charge, too serious to be answered fully in a
brief essay. Of the many issues Koperski raises, I will touch on only two: the
transition from finite measure in a space with a nonfinite total measure to
zero probability and the use of nonnormalizable “priors” by objective Bayes-
ians in the name of maximizing entropy. In both cases I want to suggest
that there is more going on than meets the eye. The former issue points to
a deep and as yet unresolved issue in the epistemology of probability; the
latter raises considerations that tend to weaken the analogy Koperski draws
between the fine-tuning argument and other applications of indifference in
scientific reasoning.
As Koperski notes, there are many distribution functions that have noth-
ing to do with probability. We can speak, for example, of the measure of one
quadrant of an infinite plane (divided into quadrants by two perpendicular
lines) on the entire plane. Both sets have infinite area; there is no convenient
way to boil these areas down so that the whole plane is 100 percent of the
space and the quadrant in question is a quarter of that. If we could accom-
plish that reduction, we would be said to have “normalized” the distribution.
But in this case, despite the persistent illusion that 25 percent must be the
appropriate ratio, we cannot do it.
Two sorts of measures are of particular interest for our purposes: those
where a nonempty subset A of the set S has zero measure (as when S is a
finite interval of the real line, A is a collection of discrete points in S, and
μ is a flat measure across S) and those where A has finite measure while
the measure of S itself is not finite (as when S is the interval [0, ∞], A is an
interval [a, b] in S with finite a, b, and once again μ is a flat measure, this
time a nonnormalizable one). In either case, the accepted terminology is that
“almost all” of the space of possibilities in S under the measure μ will lie
outside of A.
In both cases the distributions that we must deal with here are flat: equal
areas of the space S have equal measures. And for the purposes of fine-
tuning arguments, this is important. Flat distributions are the natural tools for
representation of complete ignorance since they attempt to treat all possi-
bilities on a par. This epistemic desideratum is also what creates a good deal
of the trouble, for while it is easy enough to interpret a flat measure across
a finite outcome space as a probability, there is no good way to do so for the
As, for example, in T. McGrew, L. McGrew, and E. Vestrup, “Probabilities and

2
the Fine Tuning Argument: A Sceptical View,” Mind 110 (2001): 1,027–37.
timothy mcgrew 151
real line, much less a space generated by the consideration of k independent

nonnegative real variables in the abstract space Rk.
For the moment let us waive this point and ask what we get if we take
the proffered measures at face value as probabilities. In the fine-tuning case,
this amounts to accepting that the life-friendly ranges for the constants in our
own universe have measure (and hence probability) zero; and given Koper-
ski’s willingness to bite the bullet on coarse-tuning, this consequence extends
to all finite ranges in Rk.
Two difficulties follow immediately. First, no such distribution obeys a
principle of ordinary probability theory known as countable additivity.3 To
get a feeling for what is at stake, think of a large piece of toast and a modest
pat of butter. As long as the toast is finite and the butter is infinitely smooth,
we can in theory spread the butter evenly over the toast. The layer of butter
gets pretty thin if the slice of toast is large. But if the toast is infinite, the job
simply can’t be done. If there were a certain finite amount of butter on (say)
every square inch of the toast—the same amount on each, since we want it
to be spread smoothly—then there would be an infinite amount of butter all
told. And there is no way to divide infinity by something that will bring it in
at a tidy 100 percent.
The surface of the toast is, of course, a two-dimensional stand-in for the
space of possible universes; the pat of butter is the probability we are trying
to distribute over the space. The problem we run into is analogous to the
problem of nonnormalizability of a uniform density function over the space,
assuming countable additivity.
There is a long-standing controversy between the advocates of count-
able additivity and the advocates of merely finite additivity.4 One way of
focusing the controversy, due to de Finetti, is to ask whether it is possible to
have a fair lottery with a countably infinite number of tickets and to represent
that fairness in terms of the probability each ticket has of winning. If so,
then the probability for each ticket to win must be zero. But then the proba-
bility of the countable disjunction (T1 wins ∨ T2 wins ∨ . . .) cannot equal the
countable sum of the probabilities of the disjuncts, since each of these is
zero and yet some ticket or other must surely win. It was de Finetti’s position
that such a lottery makes sense, that the fairness of the lottery could indeed
be represented by an assignment of probability zero as each ticket’s chance
of winning, and that therefore countable additivity must be replaced by the
3
Here, in defiance of informal usage, “countable” indicates an infinite set with
the cardinality of the counting numbers.
4
Jon Williamson has proved a Dutch Book theorem against merely finitely addi-
tive measures in “Countable Additivity and Subjective Probability,” British Journal for
the Philosophy of Science 50 (1999): 401–16.
finite condition that finite disjunctions of these mutually exclusive events

have the probability equal to the sum of the probabilities of the disjuncts:
P(T1 wins ∨ T2 wins ∨ . . . ∨ Tn wins) = P(T1 wins) + P(T2 wins) + . . . + P(Tn wins)
Thus the probability that any finite collection of tickets will win is simply
zero, but there is no longer a contradiction involved in saying that the proba-
bility that some ticket or other will win is one.
Countable additivity has one highly desirable property: it coordinates the
probabilities of all subspaces with that of the whole of a space in a very
convincing way. If I am certain that a particular set of options exhausts the
set of possibilities, then I should, rationally, be certain that one or the other
of them will be instantiated; this follows conceptually from the fact that they
are exhaustive. My certainty of the disjunct is a credibility. But my rational
confidence levels regarding each option considered separately are also credi-
bilities, and if probability does anything for us it should relate these specific
credibilities to the certainty of the disjunct in a convincing way. Countably
additive probability does this. If the numbers assigned to the disjuncts sum to
less than one, then either I do not actually know that they are exhaustive or I
am not using these numbers to model my actual degrees of confidence. For
some of us, this line of reasoning is far more compelling than any positive
intuitions about de Finetti’s lottery; we are therefore inclined to say that there
is something wrong with the suggestion that we can model the fairness of
such a lottery with probabilities. The same reasoning applies, mutatis mutan-
dis, to the sorts of measures Koperski envisages. If we are putting equal
measures on equal finite areas across an infinite space, then whatever else
those measures may be, they are not probabilities. And if they are not proba-
bilities, then it is not clear how the argument is supposed to work.
Finite additivity aside, there is a second difficulty with the suggestion
that we should assign a probability of zero to each member of a countably
infinite set of events E1, E2, . . . even though we are certain that one or
another of them will occur. In the standard axiomatization of probability,
P(D|E) is equivalent to the fraction P(D & E)/P(E)—a fraction that is well
defined only when P(E) is not zero. This raises some interesting puzzles both
for the definition of conditional probabilities and for conditionalization upon
them. There are ways of constructing a probability theory in which such
expressions will be well-defined. We might, for example, attempt to appeal
to infinitesimals.5 We might take conditional probability to be the primitive
and define unconditional probability in terms of it. We might reformulate
I have explored this possibility in T. and L. McGrew, “A Response to Robin

5
Collins and Alexander R. Pruss,” Philosophia Christi 7 (2005): 425–43.

timothy mcgrew 153
our mathematics so that there are not, strictly speaking, any unconditional
probabilities. But so far no one interested in the fine-tuning argument has
suggested which of these alternative axiomatizations is compelling and why.
Nonnormalizable Distributions
Koperski delicately suggests that there may be some “mathematical hand

waving” going on when physicists move from the claim that μ is a positive
measure on the state space of FLRW universes that is independent of choice
of variables and captures certain symmetries to the claim that μ determines
a probability density. And indeed there is. The careful treatment of Ellis,
Kirchner, and Stoeger brings this point out explicitly: under certain con-
ditions “one has to accept that certain questions will have no well-defined
probabilities.”6 Had the chronology of the two papers been reversed, this
might have served as the epigraph for McGrew et al., 2001.
Can we use the method suggested in Ellis et al., 2003 to get around the
difficulty? Koperski’s point is that although ∫ f(k, Λ, χ)dμ is not a normalizable
density over all of the parameter space Γ, the ratio
∫ f(k, Λ, χ)dμ
___________________
A
∫ f(k, Λ, χ)dμ
may be counted as a probability when Α is a subset of Γ. If the numera-

tor diverges—as it would in our example of the quadrant of the infinite
plane—the ratio is not defined. But if not then, by definition, P(Α) = 0.
Of the many interesting aspects of this procedure, one is salient for our
purposes. A distribution that maximizes entropy, thereby minimizing infor-
mation, has a certain intuitive appeal as a means of letting the subsequent
data speak for themselves without prejudging what their character will be.
As Bernardo says in a passage quoted by Koperski, such distributions are not
intended to describe beliefs; they are only positive functions used formally in
Bayes’s theorem to obtain nonsubjective posteriors. The max-ent “prior” is
really a mathematical template rather than an epistemic probability. In finite
cases, where normalizability is not a problem, such priors tell us as little as
possible (consistent with their being probabilities at all) about a population in
question. If we were trying to determine the characteristics of a population
6
G. F. R. Ellis, U. Kirchner, and W. R. Stoeger, “Multiverses and Physical
Cosmology,” Monthly Notices of the Royal Astronomical Society 347 (2004): 921–36,
arXiv:astro-ph/0305292v3, 17.
of discrete individuals by conditionalizing on such a prior, the observation of

even one individual would not leave us shocked by the low probability that
an individual would have just those characteristics; rather, we would be (tenta-
tively) inclined to regard that individual as rather typical.
But here it seems to me that the defenders of the fine-tuning argument
want to do something rather different: they want to take specific likelihoods
from an indifference distribution very seriously before any data come in,
to make much of the fact that indifference despises alike all finite ranges
of possible values for the constants of our universe, and on that basis to
regard the one data point we have as, so to speak, an outlier. No one suggests
that the data might be interpreted as saying that our universe is rather typical
of universes; and that cannot be quite the right response in any event, since
no very clear sense attaches to the suggestion that our universe was fished
at hazard from a barrel of universes. But the radical difference between the
standard employment of max-ent priors in Objective Bayesianism and
the appeal to indifference in the fine-tuning argument suggests that the forms
of inference at work are actually rather different and that it is, at least, less
straightforward than Koperski supposes to argue from the legitimacy of the
former to the cogency of the latter.
Conclusion
One of the charming features of a really difficult philosophical problem is

that it brings strong intuitions into conflict. Koperski’s paper indicates that
in the case of the fine-tuning argument these conflicts of intuition threaten
to ramify into other areas, and in at least some ways this is likely right. The
upshot may not be the reinstatement of the idea that “the observations sup-
port fine-tuning if they are understood in measure theoretic terms,”7 though
the issues are certainly difficult enough that it would be rash to be quite sure.
But what seems clear is that the fine-tuning argument is now catalyzing seri-
ous work on fundamental problems in epistemology and probability theory
that have been lurking in the background for decades. How should we reason
about finite subsets of infinite sets? Under what circumstances is it legitimate
for us to employ the language of probability in such reasoning? When can
we employ something like a principle of indifference? What should we do
when such principles do not seem to apply? That is reason enough to take it
seriously and to continue to probe the complex network of issues that Kop-
erski’s excellent essay has begun to uncover.
7
Koperski, “Should We Care?,” 318.
Proper Functionalism and the Metalevel
A Friendly Reply to Timothy and Lydia McGrew
Tyler Dalton McNabb

Houston Baptist University
Over the years, Alvin Plantinga has developed an epistemological system that
allows beliefs to be warranted1 without requiring the subject to have inter-
nal access to those properties conferring warrant. Plantinga’s epistemology,
known as proper functionalism, allows a subject’s belief to be warranted,
insofar as the right conditions relating to cognitive proper function are in
place.2 Plantinga’s theory of warrant can be summarized as follows:
S’s belief that P is warranted iff,
1) S’s cognitive faculties are functioning properly,

2) S’s cognitive environment is sufficiently similar to the one for
which the cognitive faculties are designed for,
3) The design plan that governs the production of such belief is
aimed at producing true belief, and
4) The design plan is a good one in that there is a high statistical
(or objective) probability that a belief produced under these
conditions will be true.3
1
Plantinga takes warrant to be that epistemic property that separates mere true
belief from knowledge. See Alvin Plantinga, Warrant: The Contemporary Debate (Ox-
ford: Oxford University Press, 1993), 3.
2
For a defense of Plantinga’s epistemology, see Tyler Dalton McNabb, “War-
ranted Religion: Answering Objections to Alvin Plantinga’s Epistemology,” Religious
Studies 51 (2015): 477–95. Also see Kenny Boyce and Andrew Moon, “In Defense of
Proper Functionalism: Cognitive Science Takes on Swampman,” Synthese 193, no. 9
(2016): 2987–3001.
3
By design plan, Plantinga has in mind the way in which something ought to
work when it functions properly. By epistemic environment, Plantinga just means
the state of affairs in which our faculties are meant to operate. For the schematic
articulation of Plantinga’s theory used here, see Joseph Kim, Reformed Epistemology
and the Problem of Religious Diversity: Proper Function, Epistemic Disagreement, and Christian
© Tyler Dalton McNabb, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)
156 proper functionaliSm and the metalevel
In Warrant and Proper Function,4 Plantinga utilizes his theory of warrant

to show how belief in other minds, belief that the future will be like the
past, memorial beliefs, perceptual beliefs, and testimonial beliefs can all be
warranted, even apart from argumentation. In an even more controversial
move, Plantinga, in Warranted Christian Belief,5 applies his theory to religious
belief. Plantinga argues for what he calls “Reformed epistemology.” Roughly,
Reformed epistemology is the thesis that religious belief can be justified or
warranted apart from argumentation. Plantinga argues that it is epistemically
possible (that is to say, given what we know it could be the case) that we have
what John Calvin called the Sensus Divinitatis (SD)—a faculty aimed toward
producing belief about God and His activities. If the SD in a subject S meets
the aforementioned proper functionalist constraints when the SD produces
belief that God exists, then belief that God exists is warranted, even apart
from S possessing an argument for her belief.
Plantinga doesn’t actually claim that we possess such a faculty. He merely
argues that it could be the case that we have such a faculty, and, if we did
have such a faculty, religious belief would be rational apart from argument.
In a sense, Plantinga’s epistemology can be seen as a possible defense of the
rationality of religious belief. Plantinga, however, not only utilizes his theory
of warrant for the purposes of developing a defense; he also utilizes his
theory of warrant to discredit other worldviews. For example, Plantinga has
argued that naturalism can’t supply the preconditions needed to make proper
function intelligible.6 Moreover, Plantinga has also argued that on naturalism,
we have a defeater for thinking that our faculties are successfully aimed at
truth. He develops what he calls the evolutionary argument against natural-
ism (EAAN).7 The idea is that, if both naturalism and evolutionary theory
were true, then human cognitive faculties would not be directly aimed toward
producing true beliefs but rather would be aimed toward producing beliefs
that aid in the Darwinian requirement of survival and reproduction. Plant-
inga is skeptical that faculties that have come about through random genetic
Exclusivism (Eugene, OR: Pickwick Publications, 2011), 19. I have chosen Kim’s artic-
ulation of Plantinga’s theory, as it is in a helpful schematic glossing.
4
Alvin Plantinga, Warrant and Proper Function (Oxford: Oxford University
Press, 1993).
5
Alvin Plantinga, Warranted Christian Belief (Oxford: Oxford University Press,
2000).
6
Alvin Plantinga and Michael Tooley, Knowledge of God (Malden, MA: Blackwell,
2008), 20–30.
7
James Beilby, Naturalism Defeated? Essays on Plantinga’s Evolutionary Argument
against Naturalism (New York: Cornell University Press, 2002).
tyler dalton mcnaBB 157
mutation would be even indirectly aimed at producing true beliefs.8 Proper

functionalism, then, plays a fundamental role both in Plantinga’s assessment
that belief in God can be properly basic and in his development of defeaters
for belief in naturalism.
There have been, however, numerous objections to both Plantinga’s
proper functionalist theory of warrant9 and his applications of it.10 For
the purposes of this essay, I will focus on proper functionalism simpliciter
and not on how he applies his theory. Specifically, I will entertain Timo-
thy and Lydia McGrew’s metalevel objection to Plantinga’s proper function-
alism. The objection centers on the proper functionalist’s need to borrow
from internalist conceptions of rationality in order to consistently formulate
defeaters. After articulating their objection, I provide two reasons to think
that the objection fails. My ultimate aim is that the robust and plausible nature
of Plantinga’s system will be seen and that the larger project of Reformed
epistemology will be advanced.
Warrant-as-Proper Function and Its

Need to Presuppose Internalism
In Internalism and Epistemology, the McGrews argue that in order for Plantinga
to use counterevidence and formulate defeaters, he must utilize internalist
concepts of rationality and counterevidence to which he has no claim.11 The
McGrews advance their argument in two stages. Their first step raises the dis-
tinction between metalevel beliefs and object-level beliefs. For the McGrews,
an object-level belief is a belief that is not about the epistemic status of
one’s own belief or the epistemic status of another subject’s belief.12 Accord-
ing to the McGrews, on the object level, no terms of epistemic appraisal
appear. This is a consequence of the fact that an object-level belief p “is not a
belief that employs epistemic concepts and S’s reasons for believing that p do
not themselves employ epistemic concepts.”13
8
Thomas Crisp, “On Naturalistic Metaphysics,” in The Blackwell Companion to
Naturalism, ed. Kelly James Clark (Hoboken, NJ: John Wiley and Sons, 2016), 61–73.
9
Jonathan L. Kvanvig, Warrant in Contemporary Epistemology: Essays in Honor of
Plantinga’s Epistemology (New York: Rowman & Littlefield, 1996).
10
Dieter Schonecker, Plantinga’s Warranted Christian Belief: Critical Essays with a
Reply by Alvin Plantinga (Berlin: De Gruyter, 2015).
11
Timothy and Lydia McGrew, Internalism and Epistemology: The Architecture of
Reason (London: Routledge, 2007), 89.
12
Ibid., 57.
13
Ibid.
However, when one asks questions related to the epistemic status of an

individual such as “Is S warranted in believing p at t?” one passes beyond
the object level and enters the metalevel. The McGrews argue that Plant-
inga’s theory has a problem in that it cannot show how a belief is justified
at the metalevel. This is so, given that, Plantinga’s theory as an externalist
theory isn’t compatible with a robust metalevel criterion. And this is espe-
cially troubling for the McGrews, who at least partly define what it means
to be internally rational by way of the metalevel principles they advance.
Because it is unable to endorse any robust metalevel principles, the McGrews
make a case that Plantinga’s theory of warrant entails metalevel circularity.
In regard to this, the McGrews state, “Yet even there [metalevel circularity]
may arise. Within Plantinga’s own system, for example, the proposition ‘God
exists’ may be held as ‘properly basic’ without any premises. If one were to
defend the claim that one is justified (or, in Plantingian terms, ‘warranted’) in
holding it, using Plantinga’s own theory, one would state, inter alia, that God
has designed us to have non-inferred spontaneous beliefs in His existence.”14
Plantinga appears happy to admit that his theory would entail epistemic
circularity.15 For example, in one place Plantinga states, “Not even God
Himself, necessarily omniscient as he is, can give a noncircular argument
for the reliability of his ways of forming beliefs. God himself is trapped inside
the circle of his own ideas.”16 According to the McGrews, fellow Reformed
epistemologist, William Alston, likewise claims that metalevel circularity is
harmless. Alston states,
Surprisingly enough, [epistemic circularity] does not prevent

our using such arguments to show that sense perception is
reliable. . . . Nor, pari passu, does it prevent us from being
justified in believing sense perception to be reliable by virtue of
basing that belief on the premises of a simple track record argu-
ment. At least this will be the case if there are no “higher level”
requirements for being justified . . . such as being justified in
supposing the practice that yields the belief to be a reliable one,
or being justified in supposing the ground on which the belief is
based to be an adequate one.17
14
Ibid., 66.
15
Cited in ibid., 75. For direct reference, see Plantinga, Warranted Christian Belief,
125. Cf. Alvin Plantinga, “Respondeo: Ad BonJour,” in Warrant in Contemporary Epis-
temology, ed. J. Kvanvig (Lanham, MD: Rowman & Littlefield, 1996), 342.
16
Plantinga, Warranted Christian Belief, 125.
17
Timothy and Lydia McGrew, Internalism and Epistemology, 71. For original text,
see William Alston, Perceiving God: The Epistemology of Religious Experience (Ithaca, NY:
Thus for Plantinga and Alston, there are no good reasons to accept that one’s
belief must meet a robust and noncircular metalevel requirement. It seems
likely that the externalist will simply reject metalevel requirements, as these
requirements would appear to be strictly internalist requirements.18
While the McGrews give arguments for rejecting that one has to know
that they know (KK theory) or that one has to be justified in believing that
they are justified (JJ theory), they do advocate other conditions that need to
be met at the metalevel. In regard to justification, the McGrews advocate the
modal principle. The modal principle states the following:
MP: If it is in principle impossible to show decisively that S’s belief that

p is justified, then S is not justified in believing that p.19
The McGrews argue that if one wants to use the term warrant instead of
justification, then they would invoke the strong modal principle, which states
the following:
SMP: For any term E intended to indicate positive epistemic status, if it

can be the case for some belief p that Ep while it is not in principle
possible to show decisively that Ep, then E is not in fact a type of
positive epistemic status.20
Given that the McGrews give brief arguments for why one should reject the
JJ thesis and the KK thesis, they think that their modal principles have a firm
foundation. The alternative to their principle is to accept the permissibility of
metalevel circularity, which, as they go on to argue, has consequences. With
this stated, the McGrews believe that they have sufficiently established that
Plantinga’s epistemology must endorse metalevel circularity. The first step of
their objection completed, they proceed with the second step.
If one was to reject MP or SMP, then how could one formulate defeat-
ers? As discussed earlier, Plantinga wants to formulate defeaters for believing
in naturalism, but given his theory of warrant, how could he do this consis-
tently? For Plantinga, even if there is a defeater that shows that a belief
is internally irrational, the subject’s belief could still be warranted because
internal rationality isn’t a necessary condition for warrant (since the subject
doesn’t have to meet any metalevel requirements). If internal rationality is not
required for warrant, then how could one formulate defeaters that attempt
Cornell University Press), 16.

18
Timothy and Lydia McGrew, Internalism and Epistemology, 77.
19
Ibid., 73.
20
Ibid., 74.
to demonstrate internal irrationality? Hence the McGrews conclude that

Plantinga’s system is circular at the metalevel and this renders the Plantingian
unable to present defeaters for other views. Since Plantinga’s theory of
warrant doesn’t give one any way to separate the epistemic sheep from the
goats, it appears that it is not a preferable theory of warrant.21
Clarifying Internal Irrationality and the Design Plan
If one’s epistemic theory of warrant entailed not being able to formulate

defeaters for another subject’s belief, one’s theory of warrant would not
seem plausible. However, I think the Plantingian can avoid the McGrews’
criticisms in at least two ways. Given that the McGrews are classical foun-
dationalists, I assume they think that for a subject to be internally rational
is for the subject to meet the traditional internalist requirements, including
internalist metalevel requirements. Moreover, I assume they think that inter-
nal rationality requires one to (ultimately) base one’s beliefs upon incorrigible
or self-evident beliefs. If this is what the McGrews mean by internally ratio-
nality, then I agree with Plantinga that internal rationality in this sense is not
necessary for warrant. However, if all it means to be internally rational is
that the subject has the correct doxastic response (which for Plantinga would
include reflecting for defeaters as prescribed by the design plan)22 to certain
phenomenological presentations, then it would appear that the proper
functionalist could endorse the necessity of some type of internal rational-
ity.23 Defeaters could be formulated to demonstrate that a subject has given
an improper doxastic response. Moreover, it is important to point out that
given Plantinga’s design plan requirement for having to reflect for defeaters,
one could not hold to a belief derived by proper function and either refuse
to reflect on possible defeaters or refuse to make an appropriate doxastic
response in light of a defeater.
21
Ibid., 88.
22
See James K. Beilby, Epistemology as Theology: An Evaluation of Alvin Plantinga’s
Religious Epistemology (Burlington, VT: Ashgate, 2006), 169.
23
Here I take phenomenological presentations to be synonymous or at least
closely related to what is more commonly referred to in epistemology as “seemings.”
By “correct doxastic response,” I just have in mind that a belief should be formed
in an appropriate way as a response to the specific phenomenology one has, and
forming the right sort of belief from the corresponding stimuli should be taken as
a necessary condition (and perhaps sufficient in some cases) for internal rationality.
If one has a particular experience that something is red, the right sort of internal
response would be to form the belief that something is red.
Furthermore, just because the design plan doesn’t require a subject to

always meet certain internalist requirements, such as having arguments for
their beliefs, it doesn’t follow that something like a propositional argument
or certain propositional evidence isn’t required for some beliefs.24 Perhaps the
design plan doesn’t require these sorts of internalist requirements for beliefs
like the belief in other minds or memorial beliefs, but it seems likely that it
would require certain internal conditions for things like the correct theory
of warrant, high-level scientific theories, or certain metaphysical beliefs such
as the belief in naturalism.25 If the design plan did require these sorts of
internalist requirements for beliefs like the belief in naturalism, then it would
appear that Plantinga isn’t being inconsistent when it comes to formulating
defeaters that demonstrate the subjective irrationality of naturalism.
Let’s return to the McGrews’ favored metalevel principle:26
MP: If it is in principle impossible to show decisively that S’s belief that

p is justified, then S is not justified in believing that p.27
The proper functionalist would resist this account. First, as an externalist, the
proper functionalist thinks that there are external factors that are responsible
for justifying or warranting a belief that she (or anyone else) lacks access to.
So it would be impossible for her (or anyone else) to show decisively that her
belief is justified or warranted.28 Second, the proper functionalist is likely to
think that there is a whole host of beliefs that don’t need to be based on an
argument in order for such beliefs to be justified or warranted (e.g., belief in
24
I take it that externalism just is the denial of internalism. That is to say, if
one denies that all beliefs need to meet some access requirement in order to be
warranted, then one is espousing a variation of externalism. If this is the case, then
Plantinga’s proper functionalism should still be considered as an externalist system.
25
One possible way to delineate between the beliefs that need to meet internal
conditions to be rational and those beliefs that don’t is to postulate that those beliefs
that fall under the category of animal knowledge do not need to meet such condi-
tions, while those beliefs that fall under the category of robust reflective knowledge
do need to meet such conditions. For more on the difference between animal knowl-
edge and robust reflective knowledge, see Ernest Sosa, A Virtue Epistemology: Apt
Belief and Reflective Knowledge (Oxford: Oxford University Press, 2007), 24.
26
I will deal with their favored principle, as I don’t see why their SMP principle
is necessary. As far as my view is concerned, it seems like their MP principle can do
the work it needs to for the concepts of both justification and warrant.
27
Timothy and Lydia McGrew, Internalism and Epistemology, 73.
28
On pages 75–76, the McGrews discuss why, on externalism, not even postulat-
ing God as an ideal observer could make SMP compatible with externalist accounts.
other minds). And if this account entails a rejection of this view, the proper
functionalist will likely reject the metalevel account instead.
However, given that she could think that there are some beliefs that need
to be based on an argument, she could think that in order for these types
of beliefs to be justified or warranted, it needs to be such that it is possible
to show that S’s belief that p is in part based on a successful argument. The
proper functionalist, then, could utilize something like the McGrews’ MP
principle and endorse the following:
Proper Functionalist MP: If it is in principle impossible to show S’s

belief that p is in part based on a successful argument, whenever it
is such that the design plan requires p to be based on a successful
argument, then S is not warranted in believing that p.
Objections and Replies
Perhaps the McGrews would argue in response that their modal principles are
analytic truths, and if one wanted to incorporate their principles (or some-
thing like their principles) in an ad hoc manner (i.e., in a way that was contin-
gent on the design plan of one’s cognitive faculties), it would be analogous
to an individual saying that 1 + 1 = 2 only when it is a sunny day in Hous-
ton. To treat an analytic truth as if it were contingent in this way would rob
it of its analytic status and render it absurd. Even if the McGrews pushed
the proper functionalist to see that this would be the case if their modal
principles were analytic truths, if the proper functionalist thinks about and
understands their modal principles and yet comes away unconvinced that
such principles are analytic truths, it would seem to me that the McGrews
would have to do more in order to motivate the proper functionalist to aban-
don her project. The McGrews could not simply declare that the principles
are analytic and yet give no positive reason for the proper functionalist to
affirm this. At best, such an approach relies on a seeming that a good number
of people do not share, and at worst, such an approach is question-begging.
If the proper functionalist, after considering such principles, is left uncon-
vinced of their analyticity, she is within her epistemic right in rejecting that
such principles are analytic truths. Moreover, if this is the case and such
principles aren’t analytic truths, the proper functionalist could incorporate
them or something like them into her own proper functionalist framework.
If the McGrews grant this, there seems to be at least one more thing they
could say in response to the proper functionalist who makes this move.
The McGrews could argue that I have not actually proven that their objec-
tion fails; I have only shown that it is epistemically possible that their objection
fails. This is so as I only argue that it is possible that the design plan of our
cognitive faculties requires that an internalist or internalist “lite” metalevel
requirement be met for beliefs like the belief in proper functionalism (i.e., a
theory of warrant). It is still possible, however, that our design plan doesn’t
require that such metalevel requirements be met for beliefs to be warranted.
And being that this is the case, I have failed to show that Plantinga doesn’t
need to borrow from the internalist’s conceptions of rationality or counter-
evidence, at least, borrow from those internalist conceptions of rationality
to which he has no claim. Rather I have only shown that it is epistemically
possible that he doesn’t need to borrow from internalist conceptions of
rationality and counterevidence to which he has no claim. And because
of this, even if Plantinga develops a defeater against naturalism, why can’t the
naturalist say that her belief in naturalism is still warranted as she isn’t moved
by Plantinga’s defeater (her belief in naturalism is so strong that it deflects
the attempted defeater) and she still thinks that her belief in naturalism is
produced from properly functioning faculties?
Though I think this last point is fair, I fail to see how this constitutes
an objection or at least a successful one. The objection is simply a version
of the conditionalization problem. The conditionalization problem can be
summarized as follows: which beliefs are relevant such that they could act as
defeater deflectors?29 Any good proper functionalist should respond to the
conditionalization problem by arguing that it is the design plan that deter-
mines when a belief should or shouldn’t be defeated given certain counter-
evidence. And this response shouldn’t be seen as wholly unsatisfying. The
whole externalist project is named in conditional terms, and those who are
sympathetic to it aren’t likely to feel the need to know that we know what
the design plan actually is in order to formulate epistemic defeaters. The
externalist could consistently formulate a defeater for the belief that p, if she
thinks belief that p isn’t part of the human design plan or if she thinks that
in order to hold to the belief that p, the belief that p needs to be based on a
successful argument. And the externalist could rest assured knowing that if
the design plan does require the belief that p be based on a successful argu-
ment that her defeater would strip away such warrant for any subject who
affirms that p and who reflects on the given defeater. Even if this type of
response doesn’t satisfy most, I’m dubious that many externalists, especially
of the proper functionalist stripe, will see the conditionalization problem as
something that proves too much for proper functionalism or any externalist
account, for that matter.
29
For a good discussion on the conditionalization problem, see Andrew Moon,
“Debunking Morality: Lessons from the EAAN Literature,” Pacific Philosophical Quar-
terly 98 (2017): 208–26.
Of course, the McGrews would not accept the response developed

here.30 Their internalist intuitions would leave them thinking that the condi-
tionalization problem, or something like it, should provide good reason for
one to reject proper functionalism altogether. But why should the proper
functionalist think that this is the case? Her intuitions are fine with the condi-
tional nature of how one knows that p, and because of this, I don’t see any
reason an already convinced proper functionalist should jump ship and aban-
don the project of proper functionalism.
Conclusion
In this essay, I have argued that Plantinga is not inconsistent in formulating

defeaters for two reasons. First, I have argued that under Plantinga’s epis-
temology, all beliefs must be formed along with some degree of internal
rationality. The design plan requires an appropriate doxastic response to cer-
tain phenomenological presentations; an appropriate doxastic response that
would even include one making the appropriate reflection for defeaters. If
this is so, then one could not hold to a belief derived by proper function
and either refuse to reflect on defeaters or refuse to make an appropriate
doxastic response in light of a defeater while remaining warranted in their
belief. Second, I then argued that one could incorporate certain metalevel
principles (even something closer to the McGrews’ own metalevel principles)
into Plantinga’s proper functionalism. This of course would not work for all
beliefs, but it seems epistemically possible that the design plan of our cogni-
tive faculties could require that something like their favored metalevel prin-
ciple be met for certain beliefs. And in doing so, Plantinga would be in his
right in directing defeaters toward those beliefs that appear to need to meet
the McGrews’ metalevel requirement. Given what I have established here,
it doesn’t appear to me that Plantinga must use internalist conceptions of
rationality and counterevidence to which he has no claim, but in fact, certain
internalist conceptions of rationality could rightly be seen within the proper
functionalist framework.
The McGrews seem to briefly allude to an externalist response that is similar to

30
the one developed and defended here. But they seem wholly unmoved. See ibid., 88.
Defeating Objections to Bayesianism by
Adopting a Proximal Facts Approach
Calum Miller
Uehiro Centre for Practical Ethics, University of Oxford
Introduction
One major line of attack against probabilistic approaches to the philosophy

of science has been to argue that certain results of theirs are in conflict with
intuitive notions of confirmation. Thus for example, some have suggested
not only that the Hempelian raven paradox1 counts against standard, prep-
robabilistic notions of scientific confirmation but also that it demonstrates
a problem with approaches based on confirmation theory: since P(nonblack
object being a nonraven|all ravens are black) is 1, it follows from Bayes’s
theorem that the observation of a nonblack nonraven constitutes evidence
that all ravens are black.2 Those who find the raven paradox persuasive, and
who retain their intuition that such an observation does not even slightly
confirm the black raven thesis, ought to find this a compelling argument
against Bayesianism, for the probabilistic account contradicts the ostensible
common-sense intuition.
Others see this as a strength of Bayesianism—that Bayesianism accepts
the otherwise plausible equivalence condition3 yet also accounts for the fact
that we do not hold such observations to significantly confirm the black
raven thesis. The reason for this is that the probability of a nonblack object
being a nonraven given that not all ravens are black is trivially close to 1,
even though it is not 1. This means that the observation—a nonblack
nonraven—is to be expected with a high degree of probability regardless of
whether all ravens are black. So the increase in the epistemic probability
of the black raven thesis is negligible.
1
Carl G. Hempel, “Studies in the Logic of Confirmation II,” Mind 54, no. 214
(1945): 97–121.
2
This assumes, of course, that P(nonblack object being a nonraven|~[all ravens
are black]) < 1, which is trivially true.
3
The equivalence condition says that if X is evidence for Y, then X is evidence
for anything logically equivalent to Y. Hence “all ravens are black” is logically equiv-
alent to “all nonblack things are nonravens.”
© Calum Miller, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)
166 defeating oBJectionS to BayeSianiSm
Inverse Gambler’s Fallacies and Bayesianism
Whatever one’s position on the raven paradox, it is clear that these kinds
of intuitive conflicts may in theory constitute a major group of objections
to Bayesianism. Let us turn to the difficulty I present in this essay. Take the
datum “tables exist”—a relatively noncontroversial proposition. And let our
hypothesis M be that there exists a multiverse—an enormous number of
disjoint space-times, or universes.4 According to the Bayesian account,
our observation of tables existing would thus seem to confirm the exis-
tence of a multiverse. This is because the probability of tables existing is
greater given the existence of a multiverse than the probability of tables
existing in the absence of a multiverse. Formally, where T = df tables exist,
P(T|M) > P(T|~M), and thus by Bayes’s theorem, P(M|T) > P(M). The
former inequality can be justified simply by appealing to the fact that the
more space-times there are, the higher the likelihood that tables would exist
in at least one of them and thus the higher the probability that the proposi-
tion “tables exist” is true. The latter inequality just says that the probability
of the proposition “a multiverse exists” is raised by the truth of the propo-
sition “tables exist.” So on the standard Bayesian account of evidence, T is
evidence for M.
This can, it seems, be applied to any existential proposition. The afore-
mentioned reasoning would suggest that any existential proposition, if true,
would provide evidence for the multiverse hypothesis. Thus “plants exist”
and “sandwiches exist” would also constitute evidence for M. The existence
of absolutely anything would seem to confirm M, and thus we would seem
to have overwhelming, perhaps infinite, evidence for M. Moreover, this is not
trivially confirmatory evidence, as in the raven paradox. Rather, the existence
of even a pen would seem to strongly confirm M, since we might suppose
P(a pen exists|~M) to be reasonably low, while P(a pen exists|M) would
be moderately high.
Worse still, objects’ existence would confirm M even more insofar as the
object is specified in a more detailed way. For example, let Rx = df x is a red
pen in a pot, along with several other pens, on a kitchen counter, in a terraced
house, in a city whose name begins with “L,” in a continent whose name
begins with “E,” on a planet whose name begins with “E,” in the Milky Way.5
Then P((∃x)Rx|~M) would seem to be extraordinarily low, whereas P((∃x)
Rx|M) would be moderately high, provided M was a sufficiently extravagant
4
The precise nature of the multiverse hypothesis under consideration is
irrelevant.
5
Evidently I wrote this sitting at my desk in London; more imaginative philos-
ophers will be able to cite more interesting examples.
calum miller 167
multiverse hypothesis. This would imply that (∃x)Rx acts not only as one
out of many pieces of evidence for M but counts as overwhelmingly strong
evidence for M. It seems that any mundane object we come across could be
described in such a detailed fashion, and so could each constitute extraordi-
nary evidence for M.
Finally, it is not clear that the evidential force of each object’s existence
would be greatly limited by the evidential force of other objects’ exis-
tence. The Bayesian approach typically shows that the evidential force of
some observations is limited by the evidential force of previous similar
observations. For example, if one already knows the results of a particular
experiment after a thousand repetitions, the results of the 1,001st repetition
will not dramatically alter the relevant hypotheses’ epistemic probability. In
Bayesian terms, this is because the results of the first thousand experiments
ought to lead us to expect the same result no matter whether the hypothesis
we are testing is true or not—and so there will be very little confirmation
by repeating the same experiment. Let us call this phenomenon evidential
limitation.6
But there does not seem to be a huge amount of evidential limitation here.7
For even though (∃x)Rx entails the existence of some other things—other
pens in the pot, for example, one can easily find the existence of an enor-
mous number of objects, largely independent, each of which confirm M
immeasurably. In addition to (∃x)Rx, then, one might also offer D in support
of M, where D = df there exists a precisely shaped dent (specified in detail)
in the biggest planet in a universe. P(D) does not seem to depend much on
P((∃x)Rx), and so would seem to give independent support to M—and again,
this would seem to be overwhelmingly strong support.
Some might perceive the primary difficulty here to be the seem-
ing commitment, from such trivial data, to an unfavorable metaphysical
position—namely, the multiverse. Even if the prior probability of a multi-
verse is minute, surely the evidence previously described is sufficient to
6
This phenomenon arises because the evidence is not independent—that is, the
probability of the latter data is affected by the knowledge of the first piece of data.
7
There is some evidential limitation. For physicists have shown—by apparent
consensus—that the laws of physics had to be very precise in order for any complex
matter to exist at all. Thus the laws of physics are said to be “fine-tuned.” Insofar as
the existence of material objects requires such laws, the existence of some material
object will significantly raise the probability of other physical objects existing. Nev-
ertheless, since the latter physical objects can be specified in arbitrarily more detailed
ways, the probability of their existence is still low even given the existence of other
material objects.
overcome it. And this seems troubling for those who are averse to a multi-
verse, for whatever reason.
But one solution might be to consider this a reductio ad absurdum against
the multiverse: if sound probabilistic reasoning leads us to consider these
phenomena as extremely strong evidence for the multiverse, perhaps the fact
that we do not really consider a multiverse to probably exist only serves to
demonstrate that the prior probability of a multiverse must be infinitesimally
low. This proposed solution would account for our intuitive conviction that
we are not really compelled to accept the multiverse on such trivial grounds.
I have some sympathies with this view, and it is plausible that quantita-
tive parsimony is indeed a theoretical virtue.8 Thus a multiverse has lower
prior probability insofar as it becomes more extravagant.9 But this kind of
Moorean shift, in this case, is entirely inadequate. For one thing, many will
consider such a move to be extremely ad hoc and will feel that staunch Bayes-
ians who do not have independent reason to appraise P(M) as insuperably
low must accept M.
But a more worrying problem arises for Bayesians: even those who have
no problem with accepting M are faced with the difficulty that the confirma-
tion of M provided by these existential claims is at odds with intuitive notions
of confirmation. Indeed, such confirmation seems to be a clear instance of a
well-known probabilistic fallacy: the inverse gambler’s fallacy.
The paradigmatic case of this fallacy involves a gambler who walks into a
casino and immediately sees that a 6 has been rolled on a die. This, he thinks,
suggests that the die has been rolled a large number of times. After all, P(a
6 is rolled on the die|the die has been rolled many times) is far greater than
P(a 6 is rolled on the die|the die has been rolled only once). Indeed, the
critique of Bayesianism currently under consideration may adduce this as a
further example: according to Bayesianism, this datum would significantly
confirm the many-rolls hypothesis over the single-roll hypothesis (one can
amplify the strength of such ostensible evidence by using an even more
improbable example: a result of 35 on a roulette wheel, for example). But
clearly such an inference is absurd: we realize that the gambler was always
bound to see something rolled on the die, and it just happened to be a 6 in this
case. It is hardly clear that his seeing a 6 on the first roll he sees confirms the
many-rolls hypothesis more than any other number would—but then we are
8
See Daniel Nolan, “Quantitative Parsimony,” British Journal for the Philosophy of
Science 4, no. 3 (1997): 329–43, and Michael Huemer, “When Is Parsimony a Virtue?,”
Philosophical Quarterly 59, no. 235 (2009): 216–36.
9
At least insofar as it is posited as a brute fact: if it is a consequence of a simpler
or more well-evidenced mechanism, then a multiverse may not become monotoni-
cally more implausible as its constituents multiply.
calum miller 169
left with the possibility that any result would confirm many rolls or that none
of the results confirms it. The first of these options seems clearly wrong,
since he would see some result whether there had been many rolls or just the
one. But then the only option is that no particular result confirms the many-
rolls hypothesis, and this is what we generally take to be true.
The problem, then, is that the Bayesian approach seems to contradict
this account and that the multiverse example is another prime instance
of this fallacy. For it seems that we were bound to see that some things exist so
long as only one universe exists (and has observers) and, intuitively, we should
be just as likely to observe tables and pens if only our universe existed, as if
many universes existed. If the other universes are not observable to us, then
they should have no impact on how likely it is that we see particular objects
existing. But if we use these existential claims as data in our Bayesian schema,
then we seem compelled to accept that they confirm the multiverse hypothe-
sis. The central problem can be summarized thus: the Bayesian approach
seems to advocate confirmation in these inverse gambler’s fallacies, when
really there is none.
A Possible Solution
All Bayesians are therefore liable to this objection and must come up with
some solution if they are to maintain that Bayesianism is consistent with
standard epistemological praxis. One more promising solution might be sug-
gested by reflecting on the standard inverse gambler’s fallacy: we realize, in
the case of the gambler, that the fact can be construed so as to avoid any
confirmation. We noted that we would expect the gambler to see something as
he walks into the casino and sees his first die roll, and in this case it happened
to be a 6. So although it is the case that P(a 6 is rolled on a die|the die has
been rolled many times) > P(a 6 is rolled on a die|the die has been rolled
only once), it is not the case that P(the gambler observes a 6|the die has been
rolled many times) > P(the gambler observes a 6|the die has been rolled only
once), provided the relevant background information is also included in the
conditional (i.e., that he has entered a casino and that this is the first roll of
the die he has witnessed). This is because what the gambler observes is not likely
to depend at all on how many times the die has been rolled previously, even
if the existence of some roll of a 6 does depend on it.
So it can be seen that how data are construed can make a huge difference
to the degree to which they confirm a certain hypothesis. Perhaps, then, the
Bayesian will want to restrict Bayesian propositions in some way. One way
this could be done is to restrict propositions to observational propositions.
Thus “I observe a pen” might be an appropriate proposition to appraise,
whereas “a pen exists” is not. This could potentially go a long way toward
alleviating the ostensible commitment to inverse gambler’s fallacies, but it
faces some forceful questions. First, what type of observational statements
are most appropriate? Second, what rationale can one give for using these
observational data and construing them this way rather than some other
data or a different construal? Third, are we to limit all Bayesian propositions
(including, for example, the hypotheses that we are supposed to be apprais-
ing) to these kinds of observational propositions? And finally, if not, by what
criteria can we judge when nonobservational propositions are appropriate,
and is it possible to find non–ad hoc criteria that do justice to the scientific
method? If this kind of solution is utilized, it ought to be able to answer
these questions persuasively.
I submit that the optimal solution lies in restricting the data used in
confirmation to what I will call “proximal facts.” This, rather than simply
providing some ad hoc solution to the aforementioned paradoxes, is the most
natural and appropriate way of utilizing Bayesian confirmation and can be
demonstrated to be so by the classic problem of uncertain learning.
Uncertain Learning and Proximal Facts
Suppose there is a murder case, and you, the detective, have been called
to the crime scene to view any evidence that might help in your investiga-
tion. On reaching the crime scene, you come across what you immediately
recognize as one of Mr. Wood’s kitchen knives—and indeed, after a brief
assessment, you are confident (i.e., think it probable) that it is his knife (call
this fact—that Mr. Wood’s kitchen knife is found at the scene—K). Since
Mr. Wood is already one of the primary suspects, you are inclined to think
that this provides significant confirmation of his being the culprit (W). Being
a good Bayesian, you form a probabilistic assessment and conclude that
P(K|W) >> P(K|~W) and that P(W|K) is now very high. You leave the
knife at the crime scene and return to the station to contact Mr. Wood.
As it happens, Mr. Wood is currently in some distant country and so you
are limited to a video interrogation with police at the other end. His quick
escape puzzles you, but even more so when he shows you his kitchen knife.
When the police with him verify that it is, indeed, his (P), you are astounded,
for you saw the same knife a very short time ago, far too short for it to have
travelled abroad since. You rush to find a way of reconciling these, coming
up with one implausible hypothesis after another—that he quickly boarded
an airplane faster than ever before, that he somehow teleported abroad, that
some deity intervened to bring this about, and so on. After all, you know
that his knife was with you at the crime scene just recently, and so you must
calum miller 171
come up with some hypothesis that takes account of this and includes it
among its explananda. All these hypotheses seem a priori unconvincing to
you, but you must search for the most probable among them in light of the
evidence.
Of course, in such a situation, where all the remaining explanations that
account for the data are so staggeringly improbable, the prudent detective
will recheck his facts. Despite his initial (and reasonable) assessment that
the knife belonged to Mr. Wood, most would agree that the more probable
situation, in light of these other facts, is that the knife did not really belong to
Mr. Wood at all. Most of us recognize that, rather than adopt a phenomenally
improbable hypothesis just to account for what we consider to be probably
true facts, we should more readily give up one or more of those facts, even
if they seemed to us more probable than not before considering their implica-
tions (and even if we thought we knew them). And ceteris paribus, we should be
more inclined to give up those facts that seemed only slightly more probable
than not.
In probabilistic terms, the detective has found that, despite all remaining
hypotheses (call these hypotheses A1 . . . An, whose disjunction might be
called A) having inconceivably low prior probability, nevertheless P(A|K & P)
is moderate. But rather than think that the most probable of these ad hoc
hypotheses are true, or even that a disjunction of some of them is true, we
seem more compelled to give up K and to remove it from our conditional.
This is an instance of the “problem of uncertain learning,”10 and a good
solution ought to make sense of this situation as well as other instances.
But oughtn’t K be replaced by something? After all, we cannot simply
ignore the knowledge we have, so it seems reasonable that we include it some-
how. But in light of P, K seems to become much more improbable. Since we
were originally less certain of K than we are of the falsehood of A1 . . . An,
we have given it up—and with full awareness that our primary circumstan-
tial data might not be as it seems, we now look for more certain and well-
founded data with which we can confirm or disconfirm our hypotheses. Thus
we might instead use the datum “a kitchen knife resembling Mr. Wood’s
was found at the crime scene” (K′). This is surely much more well-founded
than K and does not entail (when conjoined with P) that a disjunction
of absurd, ad hoc hypotheses is true. Indeed, one might much more easily
find some priorly plausible hypothesis H, accounting for K′ & P, such that
10
For a helpful discussion of uncertain learning in the context of uncertain aux-
iliary hypotheses, see Michael Strevens, “Notes on Bayesian Confirmation Theory”
(June 2017), http://www.nyu.edu/classes/strevens/BCT/BCT.pdf.
~[P(H) << 1]11 and P(H|K′ & P) > 0.5.12 One such hypothesis is that the
other primary suspect, Mrs. Walker, who is widely known to resent Mr. Wood,
bought a kitchen knife resembling Mr. Wood’s and used it to commit the
crime, thus framing Mr. Wood at the same time. Such a hypothesis would be
rendered extremely unlikely (or impossible) conditioned on K, but on K′ it
seems most reasonable.
K can still be included as a possibility in this circumstance and can
be worked into the probability assessments. In this case, K would work
as an uncertain auxiliary hypothesis in the same way that W is an uncer-
tain hypothesis, rather than as a datum taken as certainly true. Thus in our
example, P(K|K′) would be moderately high—accounting for our initial
judgment—but P(K|K′ & P) would be very low indeed.
To show how uncertain data such as K can still be included, and how
using a much more certain datum instead (K′) can give us a more realistic
probability assessment, let us consider how such a probability assessment
might work. We have said that P(A) is negligible, but P(A|K & P) = 1. If K
and P are to be taken as certainly true, then we have no choice but to accept
that A is true—that is, that some absurd, inconceivably implausible hypothe-
sis (though we know not which) is, given the data, correct. But if K′ is used
instead, we can calculate P(A|K′ & P), accounting for the uncertain auxiliary
hypothesis K thus:
P(A|K′ & P) = P(A|K & K′ & P)·P(K|K′ & P) +

P(A|~K & K′ & P)·P(~K|K & P)
Here, even though P(K|K′) is high, we are asked to calculate the probability
of K not just given K′ but also given P. And when P is considered, we have
much less confidence in K. Thus P(K|K′ & P) is inconceivably low (and
P(~K|K′ & P) is nearly 1). This will effectively make the whole first conjunct
negligible, leaving a close approximation:
P(A|K′ & P) ≈ P(A|~K & K′ & P)
And it is far from clear that P(A|~K & K′ & P) is anywhere near 1. Indeed,
if P(A) is negligible and we have at least one other plausible hypothesis under
~A that is compatible with ~K & K′ & P (a much easier condition to fulfill
than compatibility with K & P, since it allows for a similar-looking knife to
have been found at the crime scene—a relatively plausible proposition), then
it seems that, on this account, P(A|K′ & P) and P(A|~K & K′ & P) remain
11
That is, such that P(H) is not too much less than 1 (i.e., is somewhat plausible).
12
For example, that the probability of H given the data is greater than 0.5.
calum miller 173
incredibly low. So we arrive at the intuitive result: P(A|K′ & P) is low. This,
unlike our initial account, fits far better with our intuitive understanding of
theory appraisal and shows how K can nevertheless be fitted into the schema
as one possible (and priorly probable) hypothesis but also how the prior plau-
sibility of K can be overcome such that we can reject it if necessary. Thus
we can avoid concluding that absurd, counterintuitive hypotheses are correct.
So we have an account here that accords with our intuitions and that
can be generalized to all other cases of uncertain learning. And it suggests
that we should conditionalize on the most certain facts possible. This means
that facts like K should be disregarded as data in favor of facts like K′. This
approach of looking for the most certain facts is what I call the proximal
facts approach to Bayesianism.
What will such an approach involve? It seems that even highly probable
data such as K′ will prove inadequate for a comprehensive Bayesian analysis,
though they may be useful in practice. This can be illustrated by modifying
our original example. Suppose that, after modifying your datum to K′ and
having become satisfied with this more proximal fact, you then come across
some more information—when discussing your case with your team, they
have no recollection of any knife being found at the crime scene. Moreover,
when you next return to the scene, you find no trace of any knife, despite
high, trustworthy security. On returning home, you find what appear to be
hallucinogens sitting on your kitchen counter beside your early morning
coffee cup. Call these data H.
Again, we have a similar situation to before. In light of H, K′ itself seems
very implausible, and intuitively we should no longer accept K′. Rather than
trying to come up with a new set of implausible hypotheses that account for
K′ & H, then, we recognize that it may seem better to abandon K′ in spite
of the fact that P(K′|your initial observations) is high. K′, though a far more
certain fact than K, is now improbable itself in light of H. So to provide
even more certainty, you appeal to K″, a fact that is surely as certain as any
other: “I have a percept (whether veridical or not) of a knife that seems to
me to resemble Mr. Wood’s.” Rather than having to face another disjunc-
tion of wildly improbable explanatory hypotheses (call these A2) so as to
allow K′ & H to feature in our probabilistic antecedent (which would give
P(A2|K′ & H) = 1), we are faced with a new analysis:
P(A2|K″ & H) = P(A2|K′ & K″ & H)·P(K′|K″ & H) +

P(A2|~K′ & K″ & H)·P(~K′|K″ & H)
As with the previous case, P(K′|K″ & H) will now be seen to be negligible
despite P(K′|K″) being relatively high. Since this will make P(~K′|K″ & H)
roughly equal to 1, we are again left with:
P(A2|K″ & H) ≈ P(A2|~K′ & K″ & H)
And once again, this is nowhere near 1 because we no longer have to account
for K′ & H. Again, this delivers the right result. Having only to account for
K″ & H, we have a much better hypothesis not included in A2—that you
hallucinated the knife at the crime scene. And so we can see, once again,
that including only the most proximal facts—that is, those most certain to
us—provides the most complete and compelling analysis. This latter example
demonstrates that even data such as “I observe a knife” are inadequate (inso-
far as “observing” is understood as entailing veridicality). In order to include
the possibility that even our basic observations are mistaken interpretations,
we must appeal to the most immediate facts—that is, our ostensible percepts
of certain phenomena. Thus the proximal facts approach to Bayesianism
should be using this basic perceptual information as its data, with nonprox-
imal facts being used as data only for practical and expediential purposes.
Finalizing a Defense of Bayesianism
One can now see how this natural, powerful, and comprehensive approach
to Bayesianism may provide a compelling solution to the original objection.
In particular, our first two questions seem now to have decisive answers. In
response to the first, the type of observational statements that are most
appropriate are those that constitute the most fundamental, certain aspects
of our experience—namely, the data describing our percepts, with no implied
appraisal of the veracity of these percepts. And in response to the second, the
rationale is that using these data allows us to account for cases of uncertain
learning. Doing so also allows us to take all the explanatory possibilities into
account when assessing the probability of different hypotheses rather than
limiting our pool of possibilities by assuming the certainty of given proposi-
tions that are really less than certain and that are potentially defeasible.
Moreover, we can now begin to formulate an answer to the final two
questions. When reflecting on the inverse gambler’s fallacies, we noted that
we might wish to limit our propositions to those pertaining to observation.
But if this is the case, should we make all the propositions included in our
Bayesian framework these kinds of proximal facts? Including those hypoth-
eses that we are supposed to be using observational evidence to confirm or
disconfirm? If so, what is the use of Bayesianism? All our hypotheses would
be claims about what we observe—but we already know what we observe,
and the whole point of scientific investigation (at least, for the realist) is to
use observed data to help appraise propositions regarding what we do not
calum miller 175
observe! But if we shouldn’t restrict all Bayesian propositions in such a way,

then how do we rationalize our decision to allow nonproximal propositions
and nonobservational data anywhere in our schema?
The answers to these questions come from a consideration of the
purpose of Bayesianism and its intended use in science. A brief overview
of how it is alleged to work is therefore necessary. According to Bayesians,
science is characterized by a sharp distinction between two kinds of prop-
ositions to be appraised: those recording observed phenomena and those
not recording observed phenomena. The idea, for scientific realists at least,
is that observed data somehow determine, or contribute to, our appraisal
of the unobserved. In some cases, the unobserved will be propositions
regarding what will happen in the future; in other cases, the propositions
will pertain to the behavior of entities not directly observed. The classic
problem of induction (as well as a host of other riddles) relates to these and
concerns how one justifies the generalization from observed data (say, the
color of emeralds or the behavior of atoms) to unobserved situations (such
as the color of emeralds in the future or the behavior of atoms elsewhere in
the universe). In addition to this, there are a wide range of theories purport-
ing to explain these data, which are generally not held to be directly observed,
or entailed by observed data, but justified by some other method. Thus the
truth of Einstein’s general theory of relativity is not directly observed but
is held to give a good explanation of observed data and is also considered a
relatively parsimonious explanation. The problem of how to decide between
these “unobservable” theories is the subject of theory appraisal and is typi-
fied by the problem of underdetermination.
How observations are supposed to support a theory or not has been
the subject of enormous controversy for centuries, with all the proposed
solutions being attempts to find a rational justification for holding theories
to be true, despite arguments from the data to the theories generally being
deductively invalid. Bayesianism attempts to demonstrate the rationality of
such moves by appealing to the probability calculus. Here, so long as certain
axioms are granted (particularly the definition of conditional probability), it
can be demonstrated that for any h and e, P(h|e) = P(h) x P(e|h) / P(e), where
h represents the hypothesis and e represents the observational evidence. This
equation, Bayes’s theorem, demonstrates that the epistemic probability of a
hypothesis given the observational evidence is higher than the probability of
the hypothesis before13 considering that evidence, if the evidence is more to
13
“Before” is here used in a logical, rather than temporal, sense and is not in-
tended to connote any allegiance to predictivism—the thesis that novel predictions
confirm a theory more than previously known observations that are similarly accom-
modated or “predicted” in retrospect by the theory. It is a matter of some debate
be expected given the hypothesis than otherwise. This makes sense of the
notion that observations “confirm” a hypothesis if they are predicted by
that hypothesis and more so insofar as they are unexpected otherwise. The
hope, then, is that one will eventually come up with an epistemic probability
P(h|k), where k includes all observations so far made, in order to judge how
probably true the hypothesis is.
It seems clear, then, that there is no comparable reason to limit these
propositions to observational propositions. For the whole point is that we
are considering the epistemic probability of the hypothesis, given the data.
The reason we require proximal facts at all is that we need certainty (or as
near as possible) for the data that form the probabilistic antecedent of our
conditional probability in order to attain the most accurate appraisal of a
certain hypothesis. But since we are not including these uncertain hypotheses
in our antecedent knowledge, we have no need of their certainty—the whole
point is that we are trying to assess their probability, without assuming that
they are known for certain. So it is the need for certainty (or as close as we
can get) that is the criterion by which we require some propositions in the
Bayesian analysis to be of a particular kind, and this provides the answer to
our final question.14
Implications
We are now in a position to demonstrate why this most plausible reading of

Bayesianism is not committed to endorsing the fallacies described earlier. Let
us take the example of the gambler first. Instead of the gambler walking into
the casino with his background knowledge and using the datum “a 6 has been
rolled on a die,” which would give confirmation of the many-rolls hypothesis,
the gambler is rationally compelled to use the more proximal datum “it seems
that I observe that a die has been rolled to give a 6.” This, unlike the previous
whether predictivism is true, though I am heavily inclined to think not. See Richard
Swinburne, Epistemic Justification (Oxford: Oxford University Press, 2001), 221ff.
14
A clarification is required here. I am not saying that only the most proxi-
mal facts can feature in any probabilistic antecedent (i.e., in the conditional). I am
only saying that when we are trying to determine on what we should conditionalize
when trying to determine our credence in some proposition, we should use such
facts—that is, if we want to know what our credence in F should be and E represents
all our evidence as proximal facts, we should use the probability P(F|E). That is,
of course, compatible with conditionalizing on other propositions as part of the
calculation. Indeed, this is necessary for Bayes’s theorem in the first place, since part
of the calculation includes P(E|H), where H—the hypothesis—is far from certain!
calum miller 177
datum, does not seem to give any confirmation to the many-rolls hypothesis
over the single-roll hypothesis, and no fallacy is committed.
In the case of the multiverse, we are in a similar position. Instead of
using the more distal, uncertain facts, like “tables exist” and so on, the rational
observer ought to instead use, “it seems that there is something resembling
a table in my visual field” or something similar. Again, unlike the previously
construed datum, this does not seem to give any confirmation to the multi-
verse hypothesis, since the multiverse hypothesis does not seem to affect
what we observe at all.
We have, therefore, a resolution of the problems set out at the beginning
of the essay, and we have a well-motivated framework for what we should
include in our conditional probabilities. But there is one final implication I
would like to draw out related to our thinking about cosmic fine-tuning.
Although, given my embodied existence,15 it is just as likely that I would
observe the existence of tables (or any other given material object or detail)
given a multiverse than given only one universe, we might reasonably ask
whether my embodied existence itself is evidence for the multiverse. And
it appears that it is. For P(I observe that I exist|multiverse) appears to be
significantly larger than (I observe that I exist|universe). We cannot use the
same reasoning as in the other cases, for it is not inevitable that I observed
something—I might not have existed at all! And if the multiverse makes
it more likely that I observe my existence by making it more likely that I
exist in the first place, then it seems as though my existence is evidence for
the multiverse.
This is not as counterintuitive as my initial examples. Indeed, since the
discovery of cosmic fine-tuning, many physicists and philosophers have been
keen to posit a multiverse as an explanation of why the laws of the universe
are thus. So this need not be a problem for Bayesians. Indeed, it appears as
though whichever precise construal of the datum is used in the conditional,
the multiverse makes embodied life more likely than otherwise. Our intuitions
and the Bayesian result coincide: the existence of embodied life, given our
knowledge of fine-tuning, is evidence for a multiverse. But given this fine-
tuning and the existence of embodied life, observations of other material
objects or details is not so surprising: it seems as though we can observe just
as much whether there is one universe or many, since we can only observe
the universe we are part of, even if there are many. This is analogous to the
15
The assumption of embodied existence is a necessary simplification here. If
it were possible for me to exist unembodied (and so not requiring fine-tuned laws of
physics), then a multiverse would perhaps make it much more likely that I observe
material objects than if there were only one universe. But here I will assume that I
am essentially embodied.
gambler: he will witness the same thing whether or not there have been many
rolls of the die before the first roll he witnesses. But imagine we change the
gambler case to make it relevantly analogous. Given the nature of the case,
this will be somewhat contrived. But suppose that the gambler witnesses
not a 6-sided die roll but a lottery with ten million possible outcomes. And
suppose his entry into the casino was dependent on one particular outcome
of this lottery. To be sure, if he does enter the casino, he will not be surprised
to witness this outcome. But he should nevertheless be surprised that he
entered the casino in the first place if only one lottery took place. But if many
lotteries took place, it is much more likely that he would enter the casino at
some point. So in this case, it does not seem as though he has committed the
inverse gambler’s fallacy—his inference that there were probably many lotter-
ies is a reasonable one. Likewise, our existence is evidence for a multiverse.
How strong the evidence is depends on other considerations. In
general, the strength of a piece of evidence E for a hypothesis H is limited
by P(E)—the probability of the evidence in general. This, in turn, is a func-
tion of the probability of the evidence given each member of a partition (i.e.,
a set of mutually exhaustive and exclusive propositions) of possible expla-
nations for that evidence. These probabilities are to be weighted depending
on the prior probability of those alternative explanations. Thus:
P(E) = ΣP(E|Hi) x P(Hi).
The lower P(E) is, the stronger evidence E is for H. So if there is another
possible explanation Hi of E that has a moderate prior probability and that
leads us to expect E—that is, P(E|Hi) and P(Hi) are both moderate—this
will limit the strength of the evidence. In the case of fine-tuning and our
embodied existence, some philosophers and scientists suppose that they have
alternative explanations, whether theism, some deeper scientific explanation,
or something else. And in the case of fine-tuning arguments for theism, the
fact that our existence is evidence for the multiverse does not undermine
such arguments. For just as our existence is evidence for the multiverse, in
the same way it is evidence for theism (given certain assumptions that I do
not have space to defend here). And again, the strength of this evidence for
theism will be limited by the prior plausibility of alternative hypotheses—
like the multiverse itself. But note that the evidential force is limited only by
the prior plausibility of such alternative hypotheses—that is by their intrinsic
plausibility or by independent evidence for them. Since our existence is not a
piece of independent evidence but the piece of evidence to be explained, the
fact that it also supports a multiverse theory will not limit the strength of
the fine-tuning argument for theism. And as I have argued, since the exis-
tence of other material objects does not support the multiverse theory once
calum miller 179
the initial datum of fine-tuning has been taken into account, there is not
further independent evidence from this source. If the multiverse is to be
independently credible, it will need considerable independent theoretical or
empirical substantiation.
Conclusions
In this essay, I have defended Bayesianism from a potential paradox of con-

firmation: the idea that trivial existential claims appear to confirm a multi-
verse theory according to Bayes’s theorem. I noted how this can be resolved
by appealing only to the most certain data—the most proximal facts—when
appraising probabilities. I noted parallels between this case and the more well-
known inverse gambler’s fallacy and explained, likewise, how our intuitions
surrounding the inverse gambler’s fallacy can be accommodated by Bayesian-
ism. All this motivates further the proximal facts approach to Bayesianism.
I ended with one exception: our own existence as embodied observers
does appear to confirm a multiverse hypothesis. I explained how this is, in
fact, consonant with our intuitions and does indeed provide the multiverse
with a significant measure of support. However, this support is limited by
the credibility of alternative explanations for cosmic fine-tuning and does
not limit the strength of fine-tuning arguments for theism. In order to limit
the strength of such arguments, substantial independent support for a multi-
verse will be required. And that is not possible simply by appealing to the
diverse range of material objects and details in our universe.
The Myth of a Pure Virtue Epistemology
Joshue Orozco
Whitworth University
1. Introduction
G. E. M. Anscombe’s trenchant critique of consequentialist and deontolog-

ical moral theories helped bring virtues back into moral philosophy.1 Ethi-
cists committed to consequentialist or deontological frameworks gave virtues
renewed attention by developing theories of moral virtue that assimilated
virtue into their prior and more fundamental moral commitments.2 Others,
rather than assimilating moral virtue, developed a pure virtue ethic that gives
virtue and related aretaic notions of excellence and admirability a fundamen-
tal role in one’s moral framework. Some pure virtue ethicists address the tra-
ditional problems and questions (e.g., giving an account of right action) asked
by consequentialists and deontologists, some argue that there is something
flawed or importantly deficient with these traditional projects and questions
addressed by the other moral frameworks.
Virtue epistemology has experienced similar developments since
Ernest Sosa’s “The Raft and the Pyramid.”3 Some virtue epistemologists
offer theories of intellectual virtue that assimilate virtue into some more
fundamental epistemic framework (e.g., reliabilism or evidentialism). Some,
however, argue for a pure virtue epistemology that takes intellectual virtues as
personally excellent or admirable intellectual character traits analogous to
Aristotelian moral virtues, which purportedly play a fundamental role in one’s
epistemic framework. As in the moral realm, some pure virtue-epistemic
1
G. E. M. Anscombe “Modern Moral Philosophy,” Philosophy 33, no. 124: 1–16
(1958).
2
For consequentialist theories, see Julia Driver, Uneasy Virtue, Cambridge, UK:
Cambridge University Press (2001); and Thomas Hurka, Virtue, Vice and Value, Ox-
ford: Oxford University Press (2001). For a great resource on deontological (specifi-
cally Kantian) theories, see Lawrence Jost and Julian Wuerth (eds.), Perfecting Virtue:
New Essays on Katian Ethics and Virtue Ethics, Cambridge, UK: Cambridge University
Press (2011).
3
Ernest Sosa, “The Raft and the Pyramid: Coherence versus Foundations in the
Theory of Knowledge,” Midwest Studies in Philosophy 5, no. 1: 3–26 (1980).
© Joshue Orozco, Quaestiones Disputatae, Vol. 8, No. 2 (Spring 2018)
JoShue orozco 181
theories address traditional epistemic projects (e.g., understanding the nature

of knowledge or justification), others attempt to replace traditional projects
and questions, and still others attempt to simply redirect our attention to
neglected areas of evaluation that the other frameworks seem incapable of
adequately handling.
In this essay, I argue that “pure virtue epistemologies” are untenable and
suffer from crippling circularity. Specifically, I argue that the aretaic concep-
tions of intellectual virtues employed by pure virtue epistemologies are
unable to play the fundamental role they are assigned. Recent works by Julia
Annas and Jason Baehr perceptively point out that all personally excellent
or admirable character traits (whether moral or intellectual) involve behav-
ioral dispositions properly grounded in beliefs of good epistemic standing.
The problem for any pure virtue epistemology is that it is unable to give a
noncircular account of “good epistemic standing.” In light of this problem,
I conclude that any viable account of intellectual virtue, and of intellectually
virtuous performance, will have to be assimilated into some broader frame-
work that reduces the epistemic status of intellectual virtues and intellectually
virtuous belief to other more fundamental epistemic factors (e.g., truth or
evidential norms).4
I begin by developing the aforementioned distinction between impure
and pure virtue theories—both moral and epistemic—and the different
strategies that pure virtue theories can employ. I then unpack the conception
of virtue—as personally excellent character traits—commonly endorsed by
pure virtue theories. Following the recent work of Annas and Baehr, I discuss
why this conception of intellectual virtue involves a behavioral, motivational,
4
Some have argued that all virtue epistemic theories offer either “virtue reliabi-
list” and “virtue responsibilist” accounts of intellectual virtue. According to virtue
reliabilists, intellectual virtues are cognitive faculties or capacities (like our perceptual
and memory faculties) that are stable within an individual and reliable in getting
at the truth under appropriate conditions. According to the virtue responsibilists,
intellectual virtues are character traits—like intellectual courage, carefulness, and
open-mindedness—for which we have some responsibility in developing or main-
taining them. Although this classification has a lot to recommend it, I believe it has
obscured a fundamental problem with many character-based (i.e., “responsibilist”)
views. Such aretaic conceptions of intellectual virtue have generally been employed
by pure virtue epistemologies that give intellectual virtues a fundamental role in one’s
epistemic theory in the way that some virtue ethical theories give moral virtues a
fundamental role. See Guy Axtell, Knowledge, Belief, and Character, New York: Row-
man & Littlefield (2000); John Greco, “Virtue Epistemology,” Stanford Encyclopedia
of Philosophy, http://plato.stanford.edu/entries/epistemology-virtue/#Aca (2011);
Jason Baehr, The Inquiring Mind: On Intellectual Virtues and Virtue Epistemology, Oxford:
Oxford University Press (2011).
182 the myth of a pure virtue epiStemology
and cognitive structure. I then show how virtue’s cognitive structure poses
insuperable challenges for pure virtue epistemologies, regardless of the
strategy employed.
2. A Pure Virtue Ethic
Perhaps the central distinction between consequentialist, deontological, and

virtue ethics is the moral factor(s) that each theory takes to be fundamen-
tal in moral evaluations. Consequentialist theories take intrinsically good
and bad states of affairs as the fundamental moral factor. Deontological
moral theories take norms or principles of action (i.e., duties)—sometimes
in addition to good/bad states of affairs—as fundamental. Virtue ethics, in
contrast, take virtues, characterized as personally excellent and admirable
traits, as fundamental. With their preferred fundamental moral factor(s),
the various theories go on to evaluate acts, motives, characters, institutions,
and so on. For example, consequentialists might claim that an act is right
if and only if it maximizes good states of affairs among available courses of
action. Deontologists might claim that a right action follows the rule to never
treat people as a mere means and always as an end to themselves. Virtue eth-
icists might claim that an act is right if and only if it is what a virtuous agent
would characteristically do in similar circumstances.
In order for each of these theories to provide a fundamentally different
approach to moral evaluations than the others, the normative status of the
preferred moral factor(s) of each theory will not be reducible to, or wholly
explained by some natural nonnormative relation with, those of the other
two. For example, on virtue ethics, the good of moral virtues and virtu-
ous performances is not reducible to (i.e., cannot be fully explained by)
its tendency to promote intrinsically good/bad states of affairs or to rule
following. Similarly, on a deontological framework, the moral status of duties,
or norms of action, is not reducible to their tendency to promote intrinsically
good/bad states of affairs or virtuous—that is, personally excellent—actions
or character traits.
Of course, consequentialists and deontologists can incorporate virtue
into their broader framework by offering theories of moral virtues. However,
what distinguishes these impure virtue theories from a pure virtue ethic
is that they subordinate virtues by deriving virtue’s moral status from some
other, and more fundamental, moral factor(s).5 For example, consequentialist
The subordinationist strategy by consequentialists and deontologists and how

5
it contrasts with a pure virtue ethic is nicely discussed in David Solomon’s “Keep-
ing Virtue in Its Place: A Critique of Subordinating Strategies,” in Recovering Nature:
JoShue orozco 183
theories might have a conception of moral virtues in terms of some causal

relation to intrinsically nonvirtuous good states of affairs—perhaps as dispo-
sitions to maximize, or to bring about enough of, the good. Consequen-
tialists, à la Hurka (2001), could also offer a nondispositional consequentialist
account of moral virtue in which virtues are intrinsically valuable positive
orientations—for example, desiring, actively pursuing, or taking joy or plea-
sure in—toward nonvirtuous good states of affairs. Hurka’s account is
consistent with consequentialist commitments because virtue’s moral status
is wholly derived from some natural relation to (i.e., being positively orien-
tated toward) more fundamental moral factors.6 Consequentialists could even
combine the “dispositional” and “proper orientation” views so that moral
virtues are traits or dispositions reliable in bringing about nonvirtuous good
states of affairs and that involve a proper orientation toward the good.
Deontological theories can also offer an impure virtue ethic by under-
standing moral virtues in terms of some nonnormative relationship to fulfill-
ing one’s duties; for example, as dispositions of a person to fulfill one’s duty.
On this approach conscientiousness—that is, the disposition to act for the
sake of duty even in the face of contrary desires and inclinations—might be
recognized as the chief deontological virtue, while other virtues—like cour-
age, honesty, and benevolence—may involve more specific dispositions to
act in accordance with more specific principles or duties. Deontologists, as
Hurka (2001) recognizes, can also give a proper orientation account of moral
virtue in terms of some positive orientation toward fulfilling one’s duty.7 And
deontologists could advance an account of virtue that combines both of
these views.
Those wishing to provide a genuine alternative to consequentialist
and deontological theories of moral virtue have offered a pure virtue ethic
Essays in Natural Philosophy, Ethics, and Metaphysics in Honor of Ralph McInerny, Notre
Dame: University of Notre Dame Press (1999). Also see David Solomon, “Virtue
Ethics: Radical or Routine?,” in Intellectual Virtue: Perspectives from Ethics and Episte-
mology, Oxford: Oxford University Press (2003), for a related discussion.
6
Hurka (2001) states, “In my view, virtue cannot plausibly be treated as morally
primitive: it consists in an appropriate response to other moral considerations and
must be analyzed in terms of those considerations. At the same time, virtue has its
own intrinsic significance. Its presence contributes to a life’s goodness not just as a
means, but also in itself ” (42).
7
He states, “Non-consequentialists should not reject the recursive account; in-
stead, they should think about extending it. . . . If there are situations where telling
the truth is right even though it does not result in the most good, then desiring, pur-
suing, and taking pleasure in telling the truth in these situations should be virtuous
and good. And there should be other non-consequentialist virtues corresponding to
other non-consequentialist grounds of rightness” (52).
in which aretaic properties of “excellence” and “admirability” are taken as

fundamental and moral virtues are understood as personally excellent or
admirable character traits.8
Pure virtue ethical views can be differentiated by their relationship to the
traditional projects of moral philosophers—for example, giving an account
of right action.9 I will highlight three general strategies that a pure virtue ethi-
cal framework may take toward this project. Identifying them will be impor-
tant since we see parallel strategies adopted in epistemology, and it will allow
us to see the scope of the problem for pure virtue epistemic theories.
Continuity and Competition Strategy: Pure virtue ethicists may attempt

to engage the central project of consequentialist and deontological
theories of giving an account of right action and argue that having
virtues as the fundamental moral factor provides a better analysis than
the other frameworks (e.g., Hursthouse [1999], Slote [1992]).10
8
Some might worry that Aristotle’s virtue ethic might not properly fall under
what I have called “a pure virtue ethic,” since on his view, virtues are understood
in terms of eudaimonia and eudaimonia depends on nonvirtuous goods like friends,
beauty, and so on. However, for Aristotle, virtues are not good because they causally
contribute to some other nonvirtuous good states of affairs. Rather, virtues are good
because they are constitutive of the good life for human persons. Aristotle is clear
that happiness just is virtuous activity. Given that the moral status of eudemonia is
explained in terms of virtues, Aristotle’s view still falls under a pure virtue ethic as
I’m understanding it. Additionally, even though one cannot be fully happy without
other nonvirtuous goods, this is not because these other goods are constitutive of
happiness but rather because they provide the resources for a fully virtuous life.
Importantly, on Aristotle’s view, virtuous activity cannot be good because of its con-
nection to other goods, otherwise they could not be constitutive of the summum
bonum. The summum bonum, by definition, is that good that is only sought after its
own end and not for some other end. If virtues were considered good because of
their connection to other goods (like friendship, beauty, etc.) then virtuous activity
could not be constitutive of the summum bonum. Indeed, Kraut (2014) argues that on
Aristotle’s view some of the nonvirtuous goods that happiness requires are properly
understood only in relationship to virtuous activity (e.g., pleasure and friendship).
See Richard Kraut, “Aristotle’s Ethics,” Stanford Encyclopedia of Philosophy, http://plato
.stanford.edu/entries/Aristotle-ethics (2014). Thanks to Jason Baehr for raising this
worry to me.
9
I am assuming that this has been a predominant project of moral philoso-
phers. I recognize that there are others. Not much hangs on the specific examples,
however, since the three strategies listed could apply regardless of the specific proj-
ects that have traditionally garnered most attention.
10
Rosalind Hursthouse, On Virtue Ethics, Oxford: Oxford University Press
(1999). Michael Slote, From Morality to Virtue, Oxford: Oxford University Press (1992).
JoShue orozco 185
Paradigm Shift Strategy: Pure virtue ethicists may advocate abandon-

ing the traditional project altogether. For example, Anscombe (1958)
calls for jettisoning “the concepts of obligation and duty, . . . and of
what is morally right and wrong, and of the moral sense of ‘ought.’”
Although she is still concerned in evaluating actions, she advocates
using “thick” evaluative concepts like “courageous,” “honest,” and
“just,” rather than the “thin” evaluative concepts of “right” and “wrong.”
Continuity and Supplementation Strategy: Pure virtue ethicists may
advance a theory that is intended to supplement rather than compete
with the other moral frameworks and the questions they attempt to
answer. For example, one may endorse a consequentialist or deonto-
logical account for right action while maintaining that those frame-
works are incapable of adequately capturing or explaining other
kinds of moral evaluations. Adams (2006) holds that “the ethics
of virtue and the ethics of action are about different questions.”11
Whereas the notion of obligation governs our evaluations of action,
the notion of virtues governs our evaluations of character. Adams
denies (1) that there is some equivalence between the class of vir-
tuous acts and the class of morally right acts; (2) that the concept
of virtues explains what makes an act morally right or wrong; and
(3) “that virtue can be defined as a tendency to act rightly, nor even
as a tendency to try to act rightly.” Provided that one’s notion of
virtue is not reducible to other more basic moral factors, this sort
of compartmentalization or supplementation strategy is consistent
with keeping virtue a fundamental notion and with offering, what
I’ve called a “virtue ethic” (6–11).12
Some might be tempted to think that the key difference between the
normative frameworks is that consequentialism and deontology take evalua-
tions of actions as primary and evaluations of people as secondary, whereas pure
virtue ethics reverses the order. However, this temptation should be resisted.
It is consistent with consequentialism that the right act is understood as that
which arises from a virtuous character, provided that a virtuous character is
11
Robert Merrihew Adams, A Theory of Virtue: Excellence in Being for the Good,
Oxford: Oxford University Press (2006).
12
On this way of carving the theoretical landscape, a pure virtue ethic need not
commit one to “the view that a theory of virtue provides the right foundation for
all of ethics, and that the ethics of duty should be reduced to, or replaced by, the eth-
ics of virtue” (Adams [2006], 6). Although this more “imperialistic” view certainly
qualifies as a form of virtue ethics, it does not exhaust all the ways in which virtue
can play a fundamental role in one’s ethical theory.
reduced in some way to intrinsically good states of affairs (e.g., as those traits
that tend to maximize good consequences). Deontological theories can have
a similar ordering of evaluations. Furthermore, a pure virtue ethic can take
evaluations of actions as primary and evaluations of traits and persons as
secondary provided that the actions are evaluated in terms of fundamental
aretaic properties. As Adams (2006) points out, in Anscombe’s seminal paper
“[virtue provides] the terminology of moral assessment, but it is actions that
she seems absorbingly interested in identifying as ‘untruthful,’ ‘unchaste,’ or
‘unjust’” (5).13
3. Virtue Epistemology and Theories of Intellectual Virtue
Virtue epistemic theories, like virtue ethical theories, can be distinguished by

whether they offer an impure or pure theory. With impure virtue epistemic
theories, intellectual virtues do not play a fundamental role in the epistemic
framework. Instead they are assimilated and subordinated into a broader
framework in which their epistemic status is derived from more fundamental
epistemic factors. “Virtue reliabilist” views—with their conception of intel-
lectual virtues as stable cognitive faculties, powers, or capacities in an indi-
vidual that, when in appropriate conditions, reliably yield true beliefs—are
paradigm examples of an impure virtue epistemology. Such conceptions of
intellectual virtue arise from externalist epistemic frameworks in which true
belief is the fundamental epistemic good and intellectual virtues are under-
stood in terms of some causal or dispositional relation to truth.14 Unsurpris-
ingly, some have likened externalist frameworks as the epistemic analogue
13
Similarly, Adams argues that MacIntyre (2007) is also primarily concerned
with evaluations of individual and political actions. See Alasdair MacIntyre, After
Virtue: A Study in Moral Theory (3rd ed.), Notre Dame: Notre Dame University Press
(2007).
14
Even if truth is the fundamental good, it may be that not all truths are objec-
tively valuable. For example, some reliabilists claim that the truth’s value depends, in
part, on our interests, needs, and on the questions we want answered. Others, how-
ever, have objected to relativizing truth’s value. Cf. Ernest Sosa, “The Place of Truth
in Epistemology,” in Intellectual Virtue: Perspectives from Ethics and Epistemology (eds.
M. DePaul and L. Zagzebski), Oxford: Oxford University Press (2003); Alvin Gold-
man and Eric Olsson, “Reliabilism and the Value of Knowledge,” in Epistemic Value
(eds. A. Haddock, A. Millar, and D. Pritchard), Oxford University Press (2009); Ste-
phen R. Grimm, “Epistemic Goals and Epistemic Values,” Philosophy and Phenomeno-
logical Research 77, no. 3: 725–44 (2008).
JoShue orozco 187
to consequentialism and virtue reliabilism as an extension of this broader

framework.15
Virtue reliabilist views generally take the faculties of persons as the
primary point of evaluation and give beliefs a derived epistemic evaluation
based on whether they arise from virtue. Sosa initiates this move in his semi-
nal paper: “Here primary justification would apply to intellectual virtues, to
stable dispositions for belief acquisition, through their greater contribution
toward getting us to the truth. Secondary justification would then attach to
particular beliefs in virtue of their source in intellectual virtues or other such
justified dispositions” (1980, 23). However, as we saw, this does not make
such views pure virtue epistemologies.16 Virtue reliabilist views are impure
because the epistemic status of intellectual virtue is nonfundamental and
wholly derived—in this case, instrumentally—from the more fundamental
epistemic factor of truth or true belief.17
15
See Linda Zagzebski, “Virtue Epistemology,” Routledge Encyclopedia of Philoso-
phy, https://www.rep.routledge.com/articles/virtue-epistemology/v-1/history-of
-virtue-epistemology-and-its-varieties (2005).
16
Another, albeit hypothetical, version of virtue reliabilism could take beliefs as
the primary point of evaluation in terms of knowledge or justification and go on to
evaluate virtues in terms of traits that tend to yield knowledge or justification.
17
Some have argued that intellectual virtue’s derived status prevents virtue reli-
abilism from giving an adequate account of knowledge, since it fails to explain how
knowledge is intrinsically more valuable than mere true belief. See Linda Zagzebski,
“From Reliabilism to Virtue Epistemology,” Proceedings of the Twentieth World Congress
of Philosophy 5: 173–79 (2000). In response, Sosa (2003) attempts to explain how in-
tellectual performances can have extrinsic and intrinsic value while maintaining that
truth is the fundamental epistemic value. He argues that an intellectual performance
can have extrinsic (i.e., instrumental) “praxical” value in virtue of reaching the truth.
Intellectual performances can also have extrinsic “performance” value in virtue of
being of sufficiently high quality, “measured by how well such performance would
provide the expected goods [i.e., truth], if the system were properly installed, in
a suitable environment.” In addition, Sosa argues that an intellectual performance
can have intrinsic “praxical” value when its reaching the truth is attributable to the
intellectual agent’s own doing. This intrinsic “praxical” value arises only when one’s
believing has sufficiently high performance value and when one’s believing truthfully
is attributable to such a performance. Whether Sosa provides an adequate response
to the value problem of knowledge is beyond the scope of the paper (for a different
reliabilist approach to the value problem, see Goldman and Olsson [2009]). How-
ever, it seems clear to me that regardless of whether or not a virtuous performance
can sometimes get added intrinsic value, the epistemic status of an intellectual virtue
and of a virtuous performance on Sosa’s account is still wholly derived from its
dispositional and causal relationship to the more fundamental good of truth. For
on Sosa’s view, an intellectually virtuous performance (what he, in later work, calls
Nondispositional accounts of intellectual virtue could also be assimi-

lated into an externalist epistemic framework in which truth is the funda-
mental value. In analogous fashion to Hurka’s consequentialist virtue theory,
in which the moral status of a moral virtue is recursively related to more
fundamental good states of affairs, one may conceive of intellectual virtues
as involving positive orientations toward—that is, desiring, actively pursu-
ing, being motivated by, or taking joy or pleasure in—truth or other epis-
temic goods, like knowledge or understanding. Such accounts, presumably,
would give intellectual virtue intrinsic epistemic value that is wholly derived
from—because it is recursively connected to—the more fundamental intrin-
sic good of truth.18
Moreover, it is consistent with a commitment to truth as the funda-
mental epistemic good to combine the reliabilist and motivational accounts
of intellectual virtue. For example, intellectual virtues could be thought of
as intellectual traits or dispositions that involve a proper orientation toward
truth and that are reliable in yielding true belief when situated in a suitable
environment.
Given the close structural analogy with consequentialist theories of
moral virtue, I’ll call any virtue epistemology that attempts to derive the epis-
temic status of intellectual virtue from truth in the aforementioned ways
“Consequentialist Virtue Epistemologies,” with the following corresponding
conception of intellectual virtue:
Consequentialist Conception (CC): The epistemic status of intellec-

tual virtues—or of intellectually virtuous performances—is wholly
derived from the more fundamental epistemic factor of true belief
by being either dispositionally or motivationally connected to true
belief or a combination of the two.
If reliabilism is the epistemic analogue to consequentialism, eviden-

tialism is the epistemic analogue to deontology. On evidentialism, evidence
adroit performance) just is a believing with high performance value while in a suit-
able environment.
18
As we will see, pure virtue epistemic theories also incorporate a positive ori-
entation toward truth in their conception of intellectual virtues. See Baehr (2011)
and Linda Zagzebski, Virtues of the Mind: An Inquiry into the Nature of Virtue and the
Ethical Foundations of Knowledge, Cambridge, UK: Cambridge University Press (1996).
However, what makes such conceptions instances of pure virtue epistemic theories
is that the epistemic status of intellectual virtues is not reduced to the value of such
motivation. Rather, on such conceptions, the properties of admirability or personal
excellence are given a fundamental and irreducible place in the theory.
JoShue orozco 189
and evidential norms are fundamental epistemic factors by which beliefs and
intellectual character are evaluated. And like reliabilism, evidentialism can
assimilate intellectual virtue in its broader framework. One might adopt a
dispositional account that takes intellectual virtues as dispositions of an
agent to fulfill one’s epistemic duties (i.e., believe in accord with the proper
evidential principles). A nondispositional account may conceive of intel-
lectual virtues as some positive orientation toward fulfilling one’s epistemic
duties. And evidentialists could even adopt an account of intellectual virtue
that combines the dispositional and proper motivational accounts.19
Given that what is fundamental here is fulfilling one’s epistemic duties,
we’ll call such virtue theories “Deontological Virtue Epistemologies,” with
the following conception of intellectual virtue:
Deontological Conception (DC): The epistemic status of intellectual

virtues—or of intellectually virtuous performances—is derived
from the more fundamental epistemic factor of evidential norms
or duties by being either dispositionally or motivationally connected
to the epistemic good of adhering to proper epistemic norms or a
combination of the two.
In contrast to CC and DC, some have attempted to develop a pure virtue

epistemology in which intellectual virtues, as personally excellent or admira-
ble character traits, play a fundamental role in one’s epistemic theory. On this
approach, the epistemic status of intellectual virtues is not wholly derived
from more fundamental epistemic factors (e.g., truth or evidential norms).
Although intellectual virtues are importantly related to various epistemic
goods like truth, knowledge, or understanding, their status as virtues depends
on exemplifying irreducibly normative properties of personal admirability and
excellence in relation to such goods. Specifically, such intellectual virtues are
generally modeled after an Aristotelian conception of moral virtues and
are understood as acquired traits—like open-mindedness, intellectual cour-
age, carefulness, humility—for whose cultivation and sustainment one is
responsible and which are individuated by a psychological, behavioral, and
cognitive profile, consisting of specific attitudes, and actions toward various
19
Evidentialists have not followed reliabilists in attempting to evaluate beliefs
(e.g., as justified or unjustified) in terms of intellectual virtues, and for good reason.
Evidentialism’s central intuition seems to be that whether a belief is justified for
S at some time T depends on S’s mental states (i.e., S’s beliefs and experiences) at
T. Whether a belief arises from a disposition to follow one’s evidence or whether
one is positively oriented toward one’s epistemic duties, however, seems irrelevant
for determining whether that belief is justified.
epistemic goods (e.g., truth, knowledge, understanding). Accordingly, a pure

virtue epistemology seems to endorse the following conception of intellec-
tual virtues:
Personal Excellence Conception (PEC): The epistemic status of intel-

lectual virtues—or of intellectually virtuous performances—is
fundamental (i.e., not wholly derived from other more fundamental
epistemic factors) since intellectual virtues are irreducibly personally
excellent/admirable character traits involving an underlying psycho-
logical (i.e., motivational and cognitive) and behavioral structure that
is properly related to epistemic goods.
We can see this emphasis on personal excellence and admirability by many

advocates of a pure virtue epistemology.
A virtue, then, can be defined as a deep and enduring acquired

excellence of a person, involving a characteristic motivation to pro-
duce a certain desired end and reliable success in bringing about
that end. (Zagzebski [1996, 137]; emphasis mine)20
We think of human beings as persons and of the virtues as excel-

lences of persons, traits that make one excellent as a person. (Roberts
and Wood [2003, 65]; emphasis mine)21
[Let us] assume that there is a reasonably familiar and intuitive

way of being good or excellent qua person that is distinctively
intellectual—or a way of being intellectually good or excellent
that is also personal. Let us also assume an initial correspon-
dence between this dimension of value and the traits we intui-
tively regard as intellectual virtues. (Baehr [2011, 94])
We can also see different strategies among pure virtue epistemologists that
parallel the three different strategies previously mentioned for a pure virtue
ethics. The differences among strategies involve how virtue epistemologists
20
Although Zagzebski endorses a more specific, eudaimonistic conception of
intellectual virtues, this will not be relevant for my purposes. Any problems for the
more general conception (i.e., PEC) will also be problems for this more specific
conception.
21
Robert C. Roberts and Jay Wood, Intellectual Virtues: An Essay in Regulative Epis-
temology, Oxford: Oxford University Press (2011).
JoShue orozco 191
approach the traditional epistemological projects of giving an account of

knowledge and justification and answering the problem of skepticism.
Continuity and Competition Strategy: Zagzebski (1996) develops

“a pure virtue theory” in which “intellectual virtue is the primary
normative component of both justified belief and knowledge.” She
goes on to make the similarities to virtue ethics explicit by stating
that “[the] justifiedness of beliefs is related to intellectual virtue as
the rightness of acts is related to moral virtue in a pure virtue ethics”
(xv).
Paradigm Shift Strategy: Hookway (2003) argues that developing a
pure virtue epistemology allows for the possibility of doing away
with evaluations of knowledge and justification altogether without
thereby threatening our ability to adequately “describe our practice
of epistemic evaluation” (202).22 Kvanvig (1992) also argues that
we abandon the Cartesian project of evaluating singular beliefs for
knowledge and justification so that the true importance of intellec-
tual virtues in our intellectual lives can be adequately captured.23
Continuity and Supplementation Strategy: Roberts and Wood (2007)
attempt to give a conceptual roadmap of intellectual virtues for the
purposes of offering guidance for our epistemic lives “without any
pretension that [virtues are] somehow the foundation or source of
derivation of everything else.” Although they are pessimistic about
the project of providing necessary and sufficient conditions for
knowledge and justification, they are less concerned with abandon-
ing the project altogether than with developing a separate regulative
theory in which intellectual virtues “[like the] love of knowledge
and . . . epistemic humility, caution, courage, tenacity, openness,
charity, and generosity are the chief regulators” (27–28).
Baehr (2011) also endorses a version of the supplemen-
tation strategy—which he calls “Weak Autonomous Virtue
Epistemology”—and maintains that “there are theoretical—and
indeed broadly epistemological—issues and questions that virtue-
minded epistemologists might pursue but that are largely indepen-
dent of traditional epistemological issues and questions” (199).
22
Christopher Hookway, “Cognitive Virtues and Epistemic Evaluations,” Inter-
national Journal of Philosophical Studies 2, no. 2: 211–27 (1994).
23
Jonathan Kvanvig, The Intellectual Virtues and the Life of the Mind, Savage, MD:
Rowman & Littlefield (1992).
Given the structural analogy between a pure virtue ethics and a pure
virtue epistemology, it appears that PEC provides the resources to make a
pure virtue epistemology a viable third alternative to reliabilist and eviden-
tialist frameworks—and their corresponding virtue theories. I will now
argue that such appearances are illusory. The problem can be clearly seen
once we unpack what it is for an intellectual character trait to qualify as a
virtue on PEC. In what follows, I will briefly develop the structure of intel-
lectual virtues on PEC and argue that, in light of the cognitive structure of
personally excellent traits, the hopes for a pure virtue epistemology vanish.
4. The Structure of Intellectual Virtues as Personal Excellences
In claiming that virtues are personal excellences, virtue ethicists and episte-
mologists emphasize the way virtues—both moral and intellectual—go to
make one a better person qua person. This distinguishes virtues from other
excellences persons may possess—like excellent vision, speed, strength,
and beauty—which we admire persons for possessing but not qua persons.
What makes virtues personal excellences is that they necessarily speak to those
features that are distinctive and unique to persons—that is, a person’s will,
broadly speaking—in a way that other traits do not. In other words, virtues
reveal a person’s overall orientation to the world by revealing how a person
is disposed to act, feel, and reason.
First, virtues are dispositions of an individual to reliably behave in various
ways. Generally, such dispositions are acquired and sustained by repeated
practice throughout one’s life. With intellectual virtues, the relevant behav-
ioral dispositions are intellectual in nature; characterizing how an individual
is disposed to conduct one’s investigations, deliberations, and inquiries.
Someone who is intellectually courageous, say, will be disposed to continue
in their intellectual pursuits despite threats—whether actual or perceived—to
their well-being.
Second, virtues are emotional and motivational dispositions that reveal
what the person loves, desires, hopes for, is motivated by, and finds attractive
and valuable. The virtuous individual is disposed to act in various ways and is
also disposed to have her attitudes and emotions endorse those actions. For
example, an intellectually courageous individual’s willingness to risk her well-
being when engaging in intellectual activities will be grounded in a love and
concern for various epistemic (rather than moral) goods.24
Baehr (2011) suggests that the intellectually virtuous motivation involves a

24
love for epistemic goods for their own sake.

JoShue orozco 193
Third, and for my purposes most importantly, virtues involve cognitive

dispositions. Intellectual virtues reveal (1) what an individual is likely to believe
in light of her investigations, deliberations, and inquiries and (2) what she
believes about her investigations, deliberations, and inquiries. Because I will
argue that this feature proves problematic for pure virtue epistemic theories,
I will develop the cognitive structure of virtues on PEC. Annas (2011) does
just this with moral virtues, while Baehr (2013) deals primarily with intellec-
tual virtues. Once we recognize the cognitive structure PEC requires, we can
then see the problems that face a pure virtue epistemology.25
4.1 Annas and Intelligent MoralVirtue
In her book Intelligent Virtue (2011), Julia Annas defends the cognitive struc-
ture of moral virtues by arguing that they involve “the kind of reasoning that
is found in the development and exercise of a practical expertise” (169).26
This similarity between virtues and skills is perhaps best seen in the way that
both are learned dispositions. In learning any skill, there is a certain amount
of imitation involved. The novice pianist, for example, will typically learn
where and how to place her fingers on the keys, her feet on the pedals, and
so on by simply observing and imitating a role model. However, if the novice
imitates the role model’s actions without also understanding the reasons for
the particular actions, then she will fall short of becoming a master pianist.
According to Annas: “The learner needs to understand what in the role model
to follow, what the point is of doing something this way rather than that,
what is crucial to the teacher’s way of doing things a particular way and what
25
Julia Annas, Intelligent Virtue, Oxford: Oxford University Press (2011); Jason
Baehr, “The Cognitive Demands of Intellectual Virtue,” Knowledge, Virtue, and Action
(eds. David Schweikard and Tim Henning), New York: Routledge (2013).
26
That she is endorsing a PEC view of virtue is clear throughout the book but
perhaps most clear in the following passage:
Can we say anything more specific about the distinctive kind of ad-
miration called forth by virtue? The first answer is that in the case of
the virtues our admiration is for the person’s character: possession of
virtues indicates something about what the person is like, whereas pos-
session of traits such as tidiness or wittiness indicates only traits that
the person has. . . . In our society appealing traits such as wittiness and
even physical skills may attract more attention and focus than virtues,
and so they may be more salient, and encouraged, in the progress of
our everyday lives; our approval of and admiration for them may be
more vivid in our lives. But from the perspective of admiration for the
person, virtues indicate character in a way that these traits do not. (124)
is not. A learner who fails to do this will simply copy the teacher’s manner-
isms and style along with the teacher’s exact way of doing things. But this is
clearly a failure to learn the skill, not a success” (17).
Being an expert pianist requires more than memorizing how to play
Mozart, Bach, and so on. It involves learning the principles for skillful play-
ing. These principles are grasped mainly through practicing Mozart, Bach,
and so on, but practice alone is often not sufficient. As Annas puts it, “[the]
way she plays exhibits not only increased technical mastery but increased
intelligence—better ways of dealing with transitions between loud and soft,
more subtle interpretations of the music, and so on” (14). Similarly, the coura-
geous individual, say, is not someone who repeatedly but mindlessly risks her
life. She is someone who through experience has gained understanding of
when to rush to action and when to first carefully assess a situation.
By involving an understanding of the relevant ends and why specific
behaviors are needed in order to best achieve those ends, virtues, like prac-
tical skills, “enable us to respond in creative and imaginative ways to new
challenges” (15). Moreover, the best role models will give their pupils expla-
nations and reasons for why they act as they do. As Annas states, “[the] expla-
nation enables the learner to go ahead in different situations and contexts,
rather than simply repeat the exact same thing that was done” (19). Once
the pupil grasps the relevant principles and reasons, she will then be able to
act on her own in new situations and, if given enough practice and varying
experiences, perhaps even surpass the virtue possessed by the role model.
4.2 Baehr and Intelligent IntellectualVirtue
In “The Cognitive Demands on Intellectual Virtue” (2013), Baehr similarly

argues for the cognitive structure of intellectual virtues and that such a struc-
ture flows out of a PEC conception of intellectual virtues. His general point
is that in order for traits to be personally excellent, they must be properly
integrated in one’s character and that such integration requires the holding
of various beliefs.
Baehr argues that intellectual virtues possess a “two-tier psychological”
structure in which intellectual virtues are unified by a general motivation for
truth or other epistemic goods while distinguished from each other by a more
specific behavioral/psychological profile. For example, intellectual courage is
distinguished from other intellectual virtues by involving a tendency and will-
ingness to overcome fears or threats for the sake of various epistemic goods.
However, simply having these two psychological profiles is not sufficient for
personal excellence unless they are properly related to each other. Specifically,
the behavioral/psychological profile that characterizes the individual virtues
JoShue orozco 195
must be properly grounded in the “more basic concern with truth, knowledge,
or the like” (100). An intellectually virtuous individual’s care and concern for
truth is not disconnected from, but rather informs, the more specific moti-
vations and behaviors. Baehr argues that the most plausible explanation for
what constitutes “proper grounding” is that the agent believes that “engaging
in the relevant cognitive activity is an effective or reliable means to reaching
the truth” (100).27 For example, a disposition to engage in characteristically
open-minded behavior is intellectually virtuous only if one engages in such
behavior out of an overall concern for truth and because one believes that such
behavior is a reliable means to reaching the truth. Without such a belief, even
if one’s behavior is in fact reliable in reaching the truth, the individual’s
general love for truth fails to properly inform, explain, or “ground” the rele-
vant behavior. In fact, without such grounding it seems that the behavior
would more properly be described as rash than courageous.
In addition, Baehr discusses how cognitive states play an essential role
in what many regard as the most important of all virtues: wisdom. Just as
the morally wise individual is able to perceive and identify the relevant moral
values in a given situation and determine their relative weight, the intellec-
tually wise individual is able to perceive the relevant epistemic values in a
given situation and determine their relative weight. Moreover, the wise indi-
vidual uses her assessment to guide her actions. Aristotle thought wisdom
was necessary for the possession of any virtue, for as he saw things acting
virtuously entails acting appropriately, “at the right times, about the right
things, towards the right people, for the right end, and in the right way” (II. 6,
1106b21–24). Accordingly, Aristotle regarded wisdom as the virtue bridging
the intellectual and moral realms of life. Those in agreement will extend the
cognitive structure of wisdom to all virtues.
Baehr also echoes Annas’s aforementioned points and discusses how
intellectual virtue’s cognitive structure helps explain the difference between
what Aristotle calls “natural virtue” and full virtue. When we look around
us and reflect on the individual psychological and behavioral profiles of
individual virtues, we quickly realize that some people are naturally more
disposed to act in a way characteristic of virtue than others. For example, some
individuals are less averse to risking their well-being, some are more inclined
to share their resources, and for some telling the truth comes easy. However,
if we think of virtues as personal excellences, then these natural abilities or
27
Baehr seem open to the idea that the relevant belief is not about the reliability
of the particular activity but rather about the particular activity constituting a fulfill-
ment of one’s duty. At this point, however, one might begin to see how this cognitive
requirement will lead to problems for a pure virtue epistemology. More on this in
section 5.
innate traits do not rise to the level of full virtue because, as Baehr puts it,
these natural traits have not “become integrated into [one’s] psychology or
character in a reasonably deep and personal way” (103). Such integration of
one’s natural virtue requires that one recognize and approve the trait’s behav-
ior’s or activities’ overall value. For example, the honest person—if fully
virtuous—will not just be disposed to tell the truth but will also recognize
the value of truth-telling and in light of such recognition be motivated to
tell the truth. Additionally, integrating one’s natural traits requires that the
individual herself, rather than simply the vicissitudes of nature, plays a signifi-
cant explanatory role in the sustaining and strengthening of the relevant trait.
This again requires that she recognize the value of the relevant activity and
that she go on to use such recognition as a basis or motivation for engaging
in future behavior characteristic of the trait. Only when such traits are inte-
grated in this way is she deserving of the sort of admiration characteristic of
virtue in the full sense.
Finally, having established the cognitive structure of intellectual virtues,
Baehr goes on to argue that in order for the relevant beliefs to support a
personally excellent trait in the aforementioned ways they must be in good
epistemic standing. He focuses primarily on one’s belief about the goods one
aims at—whether epistemic or moral—and argues that it is not sufficient
for virtue that one “attempt to achieve what [one] regards as (epistemically
or morally) good.” Working tirelessly and with great risk to one’s life in
committing genocide, no matter how sincere, is not acting out of virtue. Nor
is it enough for virtue that one “attempts to achieve what is in fact (epistemi-
cally or morally) good.” Although this would avoid the conclusion that some-
thing like genocide could be virtuously pursued, it fails to admit that someone
could through no fault of their own virtuously attempt to achieve ends that
are not in fact good. Consequently, Baehr claims the following: “A person
S is good or admirable qua person only if and to the extent that S attempts
to achieve what S has good reason to believe is (epistemically or morally) good”
(112–13). He rightly goes on to extend the normative requirement for the
other cognitive states involved in intellectual virtues. Therefore, one must not
only reasonably believe that one’s aims are good but one must also reasonably
believe that one’s behavioral dispositions are “suitably related to S’s more
general epistemic goals.” This extension comports nicely with Annas’s discus-
sion on how being virtuous involves not mindless habituation but having an
understanding—gained through practice, teaching, and modeling—of the prin-
ciples and reasons for acting that enable one to act appropriately in a variety
of, and perhaps completely novel and unforeseen, situations. Someone who
believes that one’s behavior is reliable at securing some goods could hardly
be said to have understanding without good reasons supporting those beliefs,
even if by good luck he happens to be right.
JoShue orozco 197
5. The Myth of a Pure Virtue Epistemology
Let’s take stock. I have argued that a pure virtue epistemology is distinguished
from consequentialist and deontological virtue epistemologies by giving
intellectual virtues a fundamental role in one’s epistemic theory. In a pure
virtue epistemology, the epistemic status of intellectual virtues (or virtuous
performances) is not reducible to, or wholly explained by, some natural and
nonnormative relation to other, more fundamental, epistemic factors (e.g.,
truth, or evidential norms). A pure virtue epistemology has been coupled
with PEC, thereby making a character trait’s aretaic properties of excellence
and admirability an epistemically fundamental good that doesn’t reduce to,
for example, either a dispositional or motivational relation to the epistemic
good of truth or to the epistemic good of fulfilling one’s evidential norms
(nor does it reduce to a combination of the two relations).
I have also identified three different strategies that have been pursued
by pure virtue epistemologists: the continuity and competition strategy, the
paradigm shift strategy, and the continuity and supplementation strategy.
Finally, I argued that if intellectual virtues are going to be personally excel-
lent character traits in the way advocates of PEC suggest, then they will
necessarily possess a cognitive structure involving beliefs in “good epistemic
standing” that connect the agent’s motivation and behavior.
In light of all this, we can now see that pure virtue epistemologists face
intractable problems. For what still needs to be explained is what “good epis-
temic standing” amounts to. I argue that answering this question puts pure
virtue epistemologists in a dilemma: they must either give a viciously circular
account of good epistemic standing or abandon the idea that intellectual
virtues play a fundamental role in one’s epistemic theory. We can see this
most clearly by looking at each of the different strategies employed by virtue
epistemologists.
5.1 The Continuity and Competition Strategy
Those endorsing the continuity and competition strategy argue that having
intellectual virtues plays a fundamental role in one’s epistemic theory best
equips one in the traditional projects of giving an account of the nature
of knowledge and/or justification. For example, Zagzebski (1996) endorses
this strategy and defines knowledge as a state of true belief arising out
of acts of intellectual virtue.28 She goes on to describe acts of intellectual
28
She also defines a justified belief as a belief that a virtuous person would be-
lieve in the same circumstances and be motivated by virtuous motives.
virtue as follows: “An act of intellectual virtue A is an act that arises from the
motivational component of A, is something a person with virtue A would
(probably) do in the circumstances, is successful in achieving the end of the
A motivation, and is such that the agent acquires a true belief (cognitive con-
tact with reality) through these features of the act” (270). In order to stave
off skeptical worries, Zagzebski is clear that performing an act of intellectual
virtue does not require possessing the virtue in question. She is less clear,
however, about whether acting as a person with intellectual virtue amounts
to a personally excellent or admirable performance. Zagzebski claims that an
act of virtue generally “has all of the morally desirable features of an act, of
the agent performing the act, of the relation between agent and act, and
of the relation between act and end” (253). However, as we saw in the pre-
vious section, it is not enough for a personally excellent performance that one
simply performs the acts that a virtuous person would do; one must also
perform them in the way that a virtuous person would do them. Specifically, in
order for an act of intellectual virtue to display personal excellence, the agent
would need to have a variety of beliefs—specifically, beliefs connecting one’s
goal for various epistemic goods and one’s intellectual practices—that are in
good epistemic standing.
Such an account of knowledge would be incredibly rigorous. But
Zagzebski, to her credit, is sensitive to this and wants “an account of knowl-
edge according to which true beliefs in normal circumstances based on unre-
flective perception, memory, or introspection qualify as knowledge” (279).
Accordingly, she claims that imitating the behavior of an intellectually virtu-
ous person simply requires behaving reliably, since an intellectually virtuous
individually is likely to engage in reliable intellectual behavior. Understanding
an act of intellectual virtue in this way allows her to accommodate the intu-
ition that many of our unreflective beliefs (e.g., many of our perceptual and
memorial beliefs) qualify as knowledge. However, such an understanding
seems to abandon the project of offering a pure virtue epistemology and
amounts to nothing more than a version of CC. Recall the following:
Consequentialist Conception (CC): The epistemic status of intellec-

tual virtues—or of intellectually virtuous performances—is wholly
derived from the more fundamental epistemic factor of true belief
by being either dispositionally or motivationally connected to true
belief or a combination of the two.
Zagzebski’s understanding of an act of intellectual virtue combines being

positively oriented toward the truth and acting reliably. The aretaic properties
of personal excellence and admirability that are supposed to be central to a
pure virtue epistemology play no role in her account of knowledge.
JoShue orozco 199
If Zagzebski were to have conceived acts of intellectual virtue in a way

consistent with a pure virtue epistemology, and in keeping with PEC, then
her account would face insuperable problems in explaining what is required
for the beliefs that ground personal excellence—that is, beliefs connecting
one’s motivation for truth to one’s intellectual behavior—to be in “good
epistemic standing.” These problems, I argue, beset any pure virtue episte-
mology, no matter the strategy employed.
One option is to suggest that the relevant beliefs are in good epistemic
standing iff known (i.e., arising out of an act of intellectual virtue). However,
it should be obvious that this move would make her account of knowledge
hopelessly circular. Knowledge would be defined in terms of acts of intellec-
tual virtue, which, in turn, requires having knowledge.
Another option is to understand “good epistemic standing” in terms
of some other good epistemic state; for example, justification. But now one
needs to explain what it is for a belief to be justified, with seemingly only two
options available. Zagzebski can either appeal to intellectual virtues to help
understand justification or appeal to yet some other epistemically good state.
If she appeals to intellectual virtues, then circularity returns. Acts of intellec-
tual virtue will require justified beliefs, which in turn require an intellectually
virtuous performance.29
The only way out of circularity is to not conceive the good epistemic
standing of the beliefs constitutive of acts of intellectual virtue in terms of
intellectual virtue. One might opt for a reliabilist or evidentialist conception
of justification. But once that move is made, then it seems that one has
abandoned having intellectual virtue play a fundamental role in one’s epistemic
theory. Although such a move provides a noncircular account of knowledge
in terms of acts of intellectual virtue, it does so by reducing intellectual
virtues to either a consequentialist or a deontological conception. In other
words, one will have abandoned the project of offering a pure virtue epis-
temology since the theory of intellectual virtue will now be subsumed and
assimilated into a broader, and more fundamental, epistemic framework (e.g.,
reliabilism or evidentialism).
To see this more clearly, let’s take a look at a case that Zagzebski describes
of a belief that amounts to an act of intellectual virtue: “Suppose that a
medical researcher [forms a true belief and] does what intellectually virtuous
persons do in the conduct of her research. She is careful, attentive, thor-
ough, evaluates her methods and results critically, is open to the criticisms
of others, is flexible, and acts as if she is unprejudiced, open-minded, and
tolerant of conflicting views. Also, we can assume that she routinely reaches
29
Given that she also gives an account justification in terms of intellectual vir-
tue, then even this option forces her toward the circularity horn of the dilemma.
the truth through these procedures. . . . Her procedures are reliable, and she
is even aware that they are reliable, so she is aware that her procedures are
likely to lead to the truth” (313). We may even add that the researcher has
an intrinsic love for truth so that her intellectual behavior and her awareness
of its reliability are unified in an overall intellectually excellent character. The
question we have been asking is, What does such awareness amount to? We
have seen that it requires “good epistemic standing” and that such stand-
ing can’t be understood in terms of intellectual virtue without circularity.
If the relevant epistemic standing is understood along reliabilist lines, then
an act of intellectual virtue amounts to an intellectual performance that is
reliable in securing the truth, that is motivated by one’s love for truth, and
is grounded on other reliably produced beliefs regarding the reliability of
one’s performance. Clearly, this account of an act of intellectual virtue falls
under a reliabilist framework—one that is subsumed under truth as the
fundamental epistemic value.
If one opts for an evidentialist understanding of “good epistemic stand-
ing,” then one will require that one’s intellectual performance be properly
grounded by a well-supported belief that such a performance will likely lead
to truth. Additionally, one may also adjust the motivational component so
that the performance arises not just from a love of truth but also from a
love to pursue the truth according to proper epistemic norms. Regardless of
whether this adjustment is made, the resulting account of an intellectually
virtuous performance again becomes subsumed under an evidentialist (or
deontological) framework in which the fundamental epistemic value is fulfill-
ing one’s epistemic duties.
To be clear, my objection to the continuity and competition strategy is not
that the resulting accounts of knowledge are implausible once we recognize
the cognitive structure of intellectual virtues and of intellectually virtuous
performances. My more basic worry is that the theories of intellectual virtue
or of intellectually virtuous performances, and the corresponding accounts
of knowledge/justification, are either circular or fail to provide an alterna-
tive framework to the reliabilist (CC) and evidentialist frameworks (DC) in
the way pure virtue epistemologists advertise. And this is a problem for pure
virtue epistemology regardless of the strategy being pursued.
5.2 The Paradigm Shift Strategy
According to the paradigm shift strategy, making intellectual virtues funda-

mental in one’s epistemic theory allows one to do away with the problematic
accounts of knowledge and justification without compromising “our prac-
tice of epistemic evaluation.” A pure virtue epistemology, rather than helping
JoShue orozco 201
with the traditional projects and questions, provides the resources for an
alternative epistemic paradigm in which evaluations of knowledge and justi-
fication plays no role.
Hookway (2003) shows some sympathy with this strategy when stat-
ing the following: “Elizabeth Anscombe argued that moral ‘ought’ and the
related idea of moral obligation should be expunged from moral thought. . . .
I have tried to make attractive a similar possibility in epistemology. . . . A
virtue epistemologist can describe our practice of epistemic evaluation and
elucidate the fundamental vocabulary of epistemic evaluation without giving
a central role to knowledge and justification” (201–2). Kvanvig (1992) also flirts
with this strategy by arguing that the “Cartesian perspective [i.e., an approach
to epistemology that focuses on individual and abstract analyses of knowl-
edge and justification] must be abandoned in order to appreciate what is right
about a virtue approach to epistemology” (187).
Still, this strategy faces the same problem facing the continuity and
competition strategy. If intellectual virtues and intellectually virtuous perfor-
mances require holding various beliefs in good epistemic standing, and if one
has abandoned appealing to traditional epistemic evaluations (e.g., knowl-
edge or justification), then there will no noncircular way to understand what
an intellectually virtuous performance amounts to while still offering a pure
virtue epistemic theory. For when trying to explain what it is for the various
beliefs, constitutive of such a performance, to be in good epistemic standing
one has to either appeal again to intellectual virtues—enter circularity—or
introduce some other epistemically normative state that is neither knowledge
nor justification and that is not itself understood in terms of intellectual
virtues. But this new normative state—whatever it is—will have to be assim-
ilated into some broader and more fundamental epistemic framework (e.g.,
reliabilism or evidentialism) and along with it intellectual virtues and intellec-
tually virtuous performances that it helps explain. Once again, this abandons
the project of offering a pure virtue epistemology.
5.3 The Continuity and Supplementation Strategy
Those endorsing the supplementation strategy do not attempt to give

accounts of traditional epistemic evaluations in terms of intellectual virtue,
nor do they try to replace the traditional epistemic evaluations with evalua-
tions of intellectual virtue and intellectually virtuous performance. Rather,
they suggest that evaluations of intellectual virtue and intellectually virtuous
performance can run alongside and supplement the more traditional evalua-
tions of knowledge and justification. Moreover, an account of intellectual
virtue or of virtuous performance may not be relevant for an account of

knowledge or justification.
Since there is no attempt to provide accounts of the traditional epis-
temic evaluations in terms of intellectual virtue, and since there is no attempt
to abandon those evaluations either, this strategy allows virtue epistemol-
ogists to appeal to knowledge and justification in explaining the epistemic
standing of beliefs constitutive of an intellectually virtuous performance or
of an intellectually virtuous character without circularity. However, as we
have seen, avoiding circularity comes at a cost. First, it undermines any hope
of not having evaluations of intellectual virtue depend on more traditional
epistemic evaluations. Although Baehr (2011) is right that there are many
questions concerning intellectual virtue that can be pursued without address-
ing the traditional epistemic questions concerning knowledge and justifica-
tion,30 what the foregoing shows is that a complete understanding of intellectual
virtue and intellectually virtuous performance will not be able to escape these
questions. And once this is acknowledged, it forces one to abandon the proj-
ect of offering a pure virtue epistemology and having intellectual virtues or
intellectually virtuous performances play a fundamental role in one’s epis-
temic theory. For one will have to adopt some alternative framework through
which to understand knowledge or justification. And whatever framework
one adopts will assume a fundamental status that thereby assimilates an
understanding of intellectual virtue and intellectually virtuous performance.
5.4 One Last Attempt to Save a PureVirtue Epistemology
The only remaining option that I can see for pure virtue epistemologists is to
claim that although the cognitive structure of intellectual virtues, or of intel-
lectually virtuous performances under the PEC, shows that their epistemic
status depends on some alternative framework (e.g., reliabilism or eviden-
tialism), its epistemic status is not wholly explained by these frameworks. In
other words, one may argue that there is something beyond being positively
oriented toward truth, performing reliable intellectual behavior, and having
justified beliefs connecting one’s motivation and intellectual behavior that
explains what makes an intellectual trait or intellectual performance person-
ally excellent or admirable.
Such an argument would preserve the possibility of articulating a pure
virtue epistemology. However, this is an argument that pure virtue epistemol-
ogists have yet to make, and it is not clear how such an argument would go
See Baehr (2011: 199–202) for a detailed and helpful list of the various ques-
30
tions that might be pursued.

JoShue orozco 203
and what one might appeal to in order to maintain the fundamental status of
intellectual virtues. In the case of moral virtues, the central trait or property
that explains personal excellence is phronesis, or wisdom. Recall Aristotle’s
thought that wisdom bridges one’s moral and intellectual life by involving
“knowledge of which ends are most valuable or worth pursuing and how
best to balance and achieve these ends.”31 Baehr (2013) follows this explana-
tion regarding intellectual virtues and appeals to wisdom to explain how one’s
intellectual traits or performances qualify as personally excellent or admirable
qua persons. However, it is precisely the cognitive nature of wisdom that
makes problems for pure virtue epistemologies. If there is some other prop-
erty beyond one’s positive orientation, reliable behavior, and the wisdom that
allows one to bridge one’s orientation and behavior that explains how one’s
intellectual traits or performances are personally excellent, then it remains a
mystery what that might be.
6. Conclusion
In this essay, I have argued that, contrary to recent work in virtue episte-
mology, intellectual virtue, understood in terms of personally excellence or
admirability, cannot play a fundamental role in our epistemic theory. In other
words, I believe that a pure virtue epistemology is a myth. Nothing I have
said, however, brings into question what I have called impure virtue epis-
temologies and the various projects that they might help advance. In fact,
the only viable accounts of intellectual virtue are those that subsume virtue
into some other epistemic framework. For example, one may opt for reliabi-
list conceptions of intellectual virtues, like those developed in the work of
Ernest Sosa and John Greco. The prospects of such conceptions in help-
ing give an accurate account of knowledge remains untouched. Moreover,
it remains open for impure virtue epistemologists (e.g., virtue reliabilists) to
incorporate “character-based” conceptions of intellectual virtue into their
broader theory. For example, virtue reliabilists can offer a conception of
intellectual virtues that goes beyond reliable cognitive faculties, abilities, or
powers and that are more akin to intellectual traits—like open-mindedness,
intellectual carefulness, intellectual humility—involving complex behav-
ioral and motivational dispositions that are acquired and sustained through
repeated agency. Provided that these traits are subsumed under the broader
reliabilist framework, such conceptions can still be developed. In fact,
Baehr (2011) makes a good case for why reliabilists might find it desirable
to develop and employ character-based conceptions of intellectual virtue
31
Baehr (2013: 102).
into the broader framework, since they can serve to complement traditional
epistemic theories and projects.32 He argues that there are many scenarios in
life in which “reaching the truth is not simply or even primarily a matter of
having good eyesight, a good memory, or making valid logical inferences.
Rather, the individuals in question reach the truth because they exhibit cer-
tain attitudes or character traits. These traits seem to account most saliently
for or to best explain why the individuals form true beliefs” (54). To the
extent that reliabilists want to account for the full range of human knowledge
and understanding, they should “give a significant epistemological role to
the intellectual character virtues” (56). I have much sympathy for this sug-
gestion provided that there is no illusion that the virtues being considered
are to be countenanced as personally excellent traits that purportedly play a
fundamental and irreducible role in the broader epistemic theory.33
32
See chapter 4.
33
I do not intend to disparage impure virtue theories by suggesting that they
are not genuine virtue theories or that virtues don’t play a central role in the epis-
temic theory. Sosa’s view that reliabilist intellectual virtues are primary in epistemic
evaluation, while beliefs are secondary places intellectual virtues front and center in
an account of knowledge or justification. Similarly, it remains open to evidentialists
to develop accounts of intellectual virtues subsumed under their broader framework
and give virtues either a primary or secondary point of evaluation. As I’ve said, how-
ever, it seems to me unlikely that they would follow reliabilists in making intellectual
virtues evaluatively, normatively primary, and beliefs secondary.

Gage Epistemology QD Fus-qd0802-All

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gage Epistemology QD Fus-qd0802-All

Uploaded by

Copyright:

Available Formats

QUAESTIONES DISPUTATAE

FRANCISCAN UNIVERSITY OF STEUBENVILLE

SPECIAL ISSUE ON EPISTEMOLOGY

ISSN: 2150-5756 VOLUME 8, NO. 2 SPRING 2018

Copyright © 2018 Quaestiones Disputatae

Vol. 8, No. 2 Spring 2018

Special Issue on Epistemology

Newman on the Grounds of Faith

Complexly Based Beliefs and the Generality Problem for Reliabilism

Preemptionism and Epistemic Authority

The Unassertability of Contextualism

Can Experience Fulfill the Many Roles of Evidence?

The World, the Deceiver, and The Face in the Frost

Fine-Tuning and the Search for an Archimedean Point

Proper Functionalism and the Metalevel:

Defeating Objections to Bayesianism by Adopting a

The Myth of a Pure Virtue Epistemology

Logan Paul Gage

© Logan Paul Gage, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)

as evidence, testimony, and important theories about knowledge and reason-

Logan Paul Gage

An epistemological issue that preoccupied John Henry Newman was the

© Frederick D. Aquino, Quaestiones Disputatae, Vol. 8, No.2 (Spring 2018)

is best construed as a soft rationalist insofar as he thinks that (1) faith is

In the University Sermons, Newman offers some rich epistemological insights

be articulated (or demonstrated) formally. Thus, his dissatisfaction with exist-

Some of the University Sermons, as we have seen, can be misleading. However,

What, then, is Newman’s position on the grounds of faith? William

unconditional assents to propositions that fall short of demonstrative proof and

Second, Newman broadens the scope of what counts as rational by

Newman as a Soft Rationalist

In an essay on the importance and relevance of Newman as a philosopher,

judgment plays in evaluating evidence and forming arguments. With this in

inaccuracies and determines whether a proposition yields true rather than

irrefragable.” People who refuse to depend on the durability of a cable and

the scales of rational evaluation” but not constrained by or reducible to

The Generality Problem for Process

In broad terms, the generality problem is the problem of finding a principle

(MR) S’s belief p is warranted only if S wouldn’t have falsely believed p in

(MR*) S’s belief p is warranted only if S wouldn’t have falsely believed p

causal belief-forming process. At the broad end of the spectrum, we could

Proposals for Solving the Generality Problem

the objection is that these various proposals fail to do enough to narrow

Complexly Based Beliefs and Modal Reliabilism

By speaking of “complexly based belief,” what I mean to denote is a case

E1: A careful Theusian chronicler writing approximately a century after

E5: An inscription on the tomb of Alberich, bearing the words

On the basis of surveying this evidence, Richard concludes that Alberich

Mythrian slaves. The historian Richard’s actual-world belief in p is counted

characterizing Richard’s method, w is not counted as a world in which Richard

assassination attempt succeeded. And so Richard’s actual-world belief satis-

Complexly Based Beliefs and Process Reliabilism

it is typically quite easy in a given case of simply based belief to identify a

by Richard’s actual-world belief, based as it is on excellent evidence, and the

to speak of the type’s being either “reliable” or “unreliable” across such a

reliabilism. What characterizes these cases is that they involve complex

© Donald J. Bungum, Quaestiones Disputatae, Vol. 8, No. 2 (Spring 2018)

“replaces” a subject’s own consideration of the evidence.1 Arnon Keren, for

Is preemptionism plausible? That is, when you receive authoritative

In section 3, I describe a general strategy that the preemptionist can use

2. A Difficulty for Preemptionism: The Unhinging Objection

It will be useful to begin by setting out Jäger’s unhinging objection in a little

cardinal: Francis is not very good at identifying birds, but

3. Defending Preemptionism: A General Strategy