You are on page 1of 29

International Phenomenological Society

Modest A Priori Knowledge


Author(s): Donna M. Summerfield
Source: Philosophy and Phenomenological Research, Vol. 51, No. 1 (Mar., 1991), pp. 39-66
Published by: International Phenomenological Society
Stable URL: http://www.jstor.org/stable/2107819 .
Accessed: 22/03/2014 13:10
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

International Phenomenological Society is collaborating with JSTOR to digitize, preserve and extend access to
Philosophy and Phenomenological Research.

http://www.jstor.org

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

Philosophy and Phenomenological Research


Vol. LI,No.

1, March 1991

Modest A Priori Knowledge


DONNA

M. SUMMERFIELD

SouthernIllinois University at Carbondale

A prioriknowledgeis out of fashionthesedays,whatwithQuine'sclassic


attacks on the analytic/syntheticdistinction, with related attacks on
Wittgenstein'sdistinctionbetweencriteriaand symptomsandon the very
possibilityof knowledgethatis certain,incorrigible,and indefeasible,and
morerecently,with reliabilisttheoriesof knowledge.' Andyet, we really
coulduse somethinglike a prioriknowledge:Thereare some subjectmatters,for example,mathematicsandethics,thatexperiencearguablycould
not give us knowledgeof (even leavingasidecertainty,etc.), so thatif we
aregoing to haveknowledgeof them,we hadbetterhavesomeknowledge
not providedby experience.Further,there are principlesof inquiry2in
manyareasthatappearto be justified,but are not justifiedby empirical
evidence.Moregenerally,empiricalknowledgeitself maybe possibleonly
if there is something non-empirical,as Kant and Wittgensteinboth
insisted.We had betternot be too quick to throw the baby of a priori
knowledgeout with the bath water of extravagantclaims traditionally
madefor thatknowledge,or we may findourselveswith no knowledgeat
all. In this paper,I articulatea minimal,or modest, notion of a priori
warrantwhich shows that thereis a positionto be takenbetween those
See Quine, "Two Dogmas" [1961] and "Camapand Logical Truth"[1963] for his classic
attacks on analyticity. For attacks on the supposed special status of criteria, see Richard
Rorty [1973] and C. Chiharaand J. Fodor [1965]. For argumentsthat we do not or cannot
have knowledge that is certain, incorrigible,and indefeasible, see Wilfred Sellars [1963],
Rorty [1979]. Finally, Paul Benacerraf [1973] has argued that causal theories of
knowledge rule out knowledge of mathematics, the leading proponent of reliabilism,
Alvin Goldman, has denied that there is any a priori knowledge [1986], pp. 301-2, and
Philip Kitcher has rejected claims to a priori knowledge of mathematics because such
knowledge is not indefeasible, arguing that any warrant that deserves to be called
'independentof experience'must be indefeasible[1980, 1983].
For example, principles plausibly involved in learning language, such as the belief that
it won't look like a sheep is present unless one is present. This belief is prima facie
justified in many circumstances for persons who have learned the words in question in
the appropriateway, even though under other circumstances (e.g., if X knows that there
are accuratemodels of sheep in the vicinity) it won't be justified. See Aron Edidin [1984],
p. 384, who argues that there are many contingent beliefs that are justified, but not by
experience.

MODESTA PRIORIKNOWLEDGE

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

39

who makeextravagantclaimsfor our non-empirical


knowledgeand those
who deny that what is non-empiricalcan be knowledge at all. More
of a priorijustificationenablesus to see
specifically,my characterization
bothhow we mighthavea prioriknowledgethatis defeasibleandhow we
mighthavea prioriknowledgewithina reliabilistframework.
Some philosophersrequirethat no experiencesor empiricalbeliefs
could overrideone's a priorijustificationand that a priorijustification
mustprovidewhateversupportis neededto insurethatthe belief justified
has the statusof knowledgein all (relevant)counterfactual
circumstances
(PhilipKitcher[1983],Bob Hale[1987]).In sectionI, I distinguishpositive
fromnegativedependence;in sectionII, I distinguishthe statusof a belief
from the quantityof supportprovided for it by some source. These
distinctionsenableme to characterizea priorijustificationas justification
(of p for X at t) that dependspositively on no experienceor empirical
beliefs, and that, though it depends negatively on experience in that
experiencesor empiricalbeliefs may overridethejustificationit provides,
will neverthelessprovide the same degree of supportin all (relevant)
counterfactualcircumstances.3In section III, I respondto two alleged
to my characterization
of a priorijustification.
counterexamples
The characterization
offeredhereis neutralwith respectto a varietyof
epistemictheories,4but it is clearlycompatiblewithreliabilism,as can be
seen in two ways. First,as becomesevidentin sectionI. it can be used to
modifyKitcher'sanalysisof a prioriknowledgein a way thatpreservesits
"psychologistic"characterbut deflects Kitcher'sargumentsagainst the
possibilityof a prioriknowledgeof mathematics.Second,the assumption
that the degree of support provided by a process in counterfactual
conditions will remain constant, far from being a controversialnew
assumptionfor whichsubstantialevidencemustbe provided,is reallyjust
a condition of making distinctions that philosophers sympatheticto
reliabilisminsiston making.Thus,in sectionIV, I show thatphilosophers
sympatheticto reliabilismshouldwelcomemodestapriorism.
I. Positive and Negative Dependence
The centralconditionof Philip Kitcher'sinfluentialanalysis of a priori
knowledgerequiresthat,if a processis to be an a prioriwarrantfor X's
3

40

I make no assumptions here about the nature of the source or sources of a priori
justification -.g., I do not assume that all a priori knowledge is of propositions that are
analytic or necessary, or even that there are propositionsthat are analytic or necessary; I
do not assume that there is a special faculty of intuition, etc. In what follows, I will use,
merely as examples, processes that often are taken to be non-empiricalprocesses (e.g.,
calculation, reflection on mathematicalor conceptual relations, etc.).
Though the general characterizationof a priori justification does not presuppose reliabilism, I will often phrase my discussion in reliabilist terms.

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

belief that p, it must be indefeasible. In other words, no process will count


as an a priori warrantunless its ability to warrantthe belief that p for X is
not undermined' in any world in which X gains sufficient understandingto
believe that p. In Kitcher's terminology:
(3b):

If a is an a priori warrantfor X's belief that p then a is a process such that, given any life e, sufficient for X for p, if a process of the same type were to produce in X a belief that p,
then it would warrantX in believing thatp.6

But Kitcher thinks it is easy to describe counterfactual circumstances in


which mathematicalbeliefs (whether produced by the types of processes
favored by constructivists, platonists, or conventionalists) would fail to
be warranted,due to the existence of, e.g., social challenges,7 and therefore,
the processes that produce those beliefs fail to meet the central condition
on a priori warrants.
In response to Kitcher's charge that no warrantis independentof experience, and thus, properly a priori, since the warrantfor any belief p may be
undermined by alternative experiences, the modest apriorist will insist
that there is a significant difference between positive and negative
epistemic dependence. X's belief that p is positively dependent for its
epistemic status on the process that produces and sustains the belief, but it
is only negatively dependent on processes that would, in some
counterfactualcircumstances, undermine (or rather. override) its status as

The terms 'undermine' and 'override'-seem to be used roughly interchangeablyin the


literature, but I think that 'override' is preferable in that it does not suggest that the
supportprovided by the process that produces (or sustains) a belief is eroded. This will
become important later. Still, here I follow Kitcher's terminology and use
'undermined'.
By "any life e, sufficient for X for p," Kitcher means any life that is sufficient to enable
X to believe thatp.
Kitcher makes what he takes to be a rough distinction between direct, theoretical, and
social challenges. As applied to a Kantianaccount of geometry and, in particular,to the
proposition that the sum of the angles of any triangle is 180 degrees, the distinction is
this: a perceptual experience of a figure which apparentlycontradicts the proposition,
even judged by our best criteria, would be a direct challenge; a series of experiences
suggesting that a simpler total description of the phenomena would be possible in a
physics-cum-geometrythat lacked the proposition than in one that included it would be
a theoretical challenge; a series of experiences in which apparentlyreliable experts deny
the proposition and present hypotheses about errorsthat have led us to believe it would
be a social challenge. Most of his examples appeal to the possibility of social
challenges, though he thinks that it would often be possible, even if more difficult, to
describe possible theoretical challenges to our most cherished mathematical beliefs.
None of his examples depend upon the existence or possibility of direct challenges; in
fact, he concedes the necessity of the challenged truths, at least for the sake of the
argument.See [1983], p. 55.

MODESTA PRIORIKNOWLEDGE 41

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

knowledge.8RobertAudi([1983],p. 131),suggestsan analogyto illustrate


thedistinction.We canimaginesomeone'sinsistingthat
... when' one takes a walk in Washington Square, one's safety depends on one's not being
harmedby certain ruffianswho are several miles away stalking CentralPark, but could have
been stalking Washington Square. This is not obviously false, and if one has very high
standardsfor applying the term 'safety', one may accept it. But it is perfectly reasonableto say
that one's safety on a walk in Washington Square depends simply on what is happening
there-or relevantly near there-at the time.

Analogously,accordingto the modestapriorist,it is perfectlyreasonable


to say thatX's warrantfor believingp is dependenton theprocessthatproducesandsustainsthe belief thatp in a way thatit is not dependenton the
circumstances.9
processesthatmayoverrideX's warrantin counterfactual
modestaprioristspointto an ambiguityin the notion
Correspondingly,
the apriorist'sclaimthatX has some beliefs
of experience-independence:
that are warrantedindependentlyof their relations to experience is
ambiguousas between (1) At t, X has some warrantedbeliefs whose
warrantdoes not, at t, dependon, in the sense that it derives from, the
warrantof any empiricalbelief(s) X has at t or the warrantprovidedby
beliefs
any experiencesX has at or beforet, and(2) X has somewarranted
whichwould (otherthingsequal)remainwarrantedno matterwhatother
empirical beliefs X forms and no matter what other experiences X
undergoes.To hold (2) is to hold that there are some beliefs that are
indefeasiblein the face of experience,but to hold (1) is to hold something
muchweaker.And yet, (1) articulatesa perfectlygood sense in which a
of experience.'0
independently
belief mightbe warranted
In responseto this suggestion,sone will arguethatif X can now (at t)
lose herjustificationfor believing thatp by coming to have subversive
experience,thenshecouldhavelost herjustificationin the sameway at any
earliertime duringwhich she was justifiedin believingthatp. Put more
8

10

42

In some circumstances,X may actually have beliefs the denial of which would override
her justification for believing p, but in other cases the person may not. For example,
suppose that p is a complex mathematicaltheorem; X may not have the belief that no
medical experts are trying to hide from her compelling evidence that she suffers from a
rare disease that destroys her ability to follow complex mathematical proofs, even
though, if she comes to believe the denial of that proposition, her justification for
believing p may be overridden.
Audi [1983] makes a parallel response to Hilary Komblith [1980]. However, Audi's
response is directedto the chargethat no belief is independentof other beliefs, and thus,
properly basic, since any belief may be underminedby alternative beliefs. Rather than
developing a notion of modest a priori knowledge, Audi is concerned in the paper
mentioned to develop a notion of modest foundationalism.
See Audi [1983], p. 129, for the parallel claim about foundationalbeliefs. The proposal
just articulatedis similar in certainrespects to a proposalmade by Edidin [1984].

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

generally,if a stateS can be affectedby a typeof conditionC at sometime


t, thenit is only a contingentfactthatS is not affectedby thattype of conditionat any othertime duringwhichS obtains.In short,so the argument
goes, thereis no significantdifferencebetweenpositiveandnegativedependence, since whateverempiricalbeliefs S might form in the futureand
whateverexperiencesX mightundergoin the futureonly contingentlyfail
to occur before t (for any t duringthe time that X's belief is justified).
Insofar as those empirical beliefs and/or experiences would, if they
occurred,overrideX's justification,they only contingentlyfail to override
herjustificationbeforet. Therefore,herjustificationat any time t reallyis
not independentof those empiricalbeliefs and experiencesthat could
overridethatjustification."
In responseto this objection,noticethatsomeonewho makesa distinction betweenpositiveandnegativedependencewill grantthatthereis one
sense of 'independence'in which X's justificationfor believingp is not
independent of the beliefs and experiences that may, in certain
counterfactualcircumstances,overrideit. However,the defenderof the
distinctionwill insistthatthereis anothersense in whichX's justification
for believing p is independentof what may potentiallyoverrideit. Of
course, it is true thatthe beliefs and experiencesthat may overrideX's
justification only contingently fail to occur before any particulart.
Nevertheless, intuitively, there is a difference between the factors
responsiblefor the initial productionand sustenanceof a belief and the
factorsthatwould,if theyoccurred,overrideX's justification.
'Producing'and 'sustaining'arecausalnotions,andthe kindof distinction I havein mindis familiarenoughfromordinarycausaltalk,even if it
is notoriouslydifficultto makeprecise.Ordinarily,we wouldsay thatthe
strikingof the matchcauses it to light, even though(a) thereare certain
backgroundconditionsthathave to be met (e.g., oxygenin the room)and
(b) thereis a ceterusparibusclause(e.g., not-x,y, etc.).Further,we maysay
that the match's lighting causally dependsboth on the strikingof the
matchandon the presenceof oxygenin the roomin a way thatit does not
dependon the absenceof conditionsthatwould,if they obtained,prevent
the matchfromlighting.Similarly,intuitively,we may say thata belief is
independent(in one sense) of certainempiricalbeliefs and experiencesif
theyplayno roleanalogouseitherto thatof causeor conditionsof kind(a),
even though they may play a role analogousto that of the conditions
excludedby theceterusparibusclause.
Thus,I believe, thereis a distinctionto be madebetweenpositiveand
negative dependence;this distinctionenables the modest aprioristto
This argumentderives from an argumentHale [1987] gives against Edidin's distinction
between forward-lookingindependenceand backward-lookingindependence.

MODESTA PRIORIKNOWLEDGE 43

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

require for a priori justification only that a belief be positively dependent


upon no experience or empiricalbeliefs. Since X's belief that p will be only
negatively dependent upon the experiences and empirical beliefs that may
override her justification in counterfactual circumstances, the possibility
of such overridingwill not affect p's status as a prior.
Some may insist that, if the apriorist can answer Kitcher without
forcing a distinction between positive and negative independence, she
should; perhaps there is a lingering suspicion that Kitcher may be right to
require that, if one's knowledge is really independent of experience, no
experiences or empirical beliefs could override one's a priori justification.
Consider, then, the following attempt to answer Kitcher while accepting
Kitcher's central condition on a priori warrants:12If X has knowledge of p
(where p is any statement, mathematicalor otherwise), then p is true. But
if p is true (where p is a mathematical statement), then (since it is a
mathematical truth) it is necessarily true, so there will be no
counterfactualcircumstancesin which p is false. Kitcher has grantedthat X
has knowledge of p, so p is true. Thus, there will be no counterfactual
circumstancesin which p is false. Therefore,no evidence X could have will
succeed in showing that p is false (or that it needs to be revised); whatever
such evidence X appears to have will be misleading. But misleading evidence does not overrideX's original (and, by hypothesis, adequate)grounds
for believing p.
There are several serious objections to this argument: First, Kitcher
could make his argumentagainst the possibility of a priori justification of
mathematics equally well without assuming that the belief in question is
true, but only rationally believed.13
12
13

This argumentseems to express one of the lines of thought underlyingHale's discussion


of Kitcher[1987].
In fact, a passage from Kitcherquoted by Hale suggests this possibility. Kitcher objects
to Saul Kripke's claim that a priori knowledge or belief need not involve certainty.
Kripkedescribesa case in which you have reada proof in a mathematicsbook and think it
is correct, but you are not able to rule out the possibility of error. Against Kripke,
Kitcher tries to show that whereas uncertainty is compatible with knowledge and
rationalbelief, it is not compatible with a priori knowledge and rationalbelief:
In this case, a nonempiricalprocess engendersbelief. However, the statement
known (believed) is not known (rationally believed) a priori. The process
producing it does not meet condition (3b) on a priori warrants.Experiences
which cast doubton the accuracyof the book (by appearingto expose errorsin
many "theorems,"let us say), and in which eminent mathematiciansdenied
the conclusion, would interfere with the ability of the process to warrantthe
belief.
...Reasonable uncertaintyis typically compatible with knowledge because of
the kindly nature of background experience. Transform the quality of our
lives, and that knowledge could no longer coexist with the uncertainty.
(Kitcher,[1983], p. 43)

44

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

Second, this argument falls prey to the same confusion commentators


have accused Kitcherof making in a slightly different context:14
On the face of it, there is simply a confusion at this point. Obviously, for X to know a priori
that p, it must be true thatp (just as it must be true that p, if X is to know a posteriorithat .p).
It clearly does not follow that the satisfaction of this condition, in the case of knowledge a
priori, must be ensuredby the fact that X has .a priori groundsto believe thatp. It seems that,
by conflating the truth-entailingcharacterof knowledge (of any sort) with the conditions on
the way in which a belief is held requiredfor it to qualify as knowledge a priori, Kitcher
gratuitously inflates the concept of knowledge that is independentof experience into that of
infallible knowledge.15

Finally, in general, beliefs may be warrantedthough false, so we need


some reason to think that mathematicalbeliefs are different in this respect.
The argument given above does distinguish mathematical beliefs from
many others, in that the formerare taken to be necessary, but that does not
by itself show that they may not be warrantedeven when they are false.
What the argument shows is at best only that if a mathematical belief is
warranted,it is either necessarily true or necessarily false.
However, there is anotherpossible argumentwhich provides a response
to the objection just made:16 If X has knowledge of p (where p is any
statement, mathematical or not), then her warrantfor believing p will be
pretty good, in fact, sufficient for knowledge of p. If X has knowledge of p
(where p is a mathematical statement), and if her warrant for believing
that p is (plausibly) a priori, then it will be, for example, the process of
following a proof or the process of reflecting upon concepts. But if these
types of processes are to provide warrant at all, then the proof must be
sound and the conceptual reflection must be accurate. The process of
following a flawed proof and the process of confused reflection do not
really provide any warrantfor believing that p, although they may appear
to do so. Kitcher has granted that X sometimes knows mathematical
statements on the basis of following a proof or engaging in conceptual
reflection. When this is so, X's grounds reveal the (necessary) truthof that
for which they are the grounds, so nothing could override them. (When X
fails to know a mathematicalstatementbecause what appearedto be her a
priori warrantis faulty, then the statement must be rejected or revised, but
14

15
16

Given one of Kitcher's conditions on a prioriwarrants,(3c), no process will count as an a


priori warrantunless it is what Kitcher calls 'ultra-reliable';processes that are a priori
warrantsmust guaranteetrue belief; they must never lead us astray. The passage quoted
is Hale's objection to Kitcher's condition (3c), yet it works equally well against Hale's
own argument.
Hale, [1987], p. 129. See also AlbertCasullo [1988].
This may be the argument Hale [1987] had in mind, rather than the one previously
discussed.

MODEn A PRIORIKNOWLEDGE 45

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

this is for a priorireasons and it does not overrideher warrant,since she had
none to begin with.) Thus, no empiricalevidence could show that the belief
that p is false or that it needs to be revised (though it may appearto do so).
Unlike the first argument, this argument at least gives us a reason for
thinking that mathematical beliefs are different from other beliefs in a
relevant respect: the types of processes that provide a priori warrant for
mathematical beliefs (if any do) are processes that confer warrant only
when they guarantee truth. According to this argument, it is not possible
for mathematicalstatements to be warranteda priori and yet false (though
it is possible for them to appearto be warrantedand yet be false).
However, I do not find this argument compelling. First, its success
depends upon the plausibility of its central premise, that a priori grounds
must guarantee truth if they are to provide any warrant at all, and that
premise is not very plausible. It seems reasonable to say that a capable and
careful mathematicianmay be warrantedon the basis of following a proof
in believing a theorem, even though the proof may contain an extremely
subtle flaw that she has not (or not yet) detected. It seems gratuitous to
insist that she merely appears to have warrant,or to insist that any warrantshe has must be empiricalratherthan mathematical.17Second, thereare
probabilistic proofs in mathematics; they do not guarantee truth, but
presumably the beliefs that result from following them have as great a
claim to a priorijustification as other mathematicalbeliefs.18 Third, even if
we grant that a priori warrants confer justification only when they
guarantee truth, it won't follow that the knowledge that results is
indefeasible by experience." In general, even if a priori warrantsconfer
justification on p only when they guarantee p's truth, we still may have
misleading evidence that casts doubt on the reliability of the non-empirical
process that produces the belief p. When we do, the warrantfor p may be
overridden.

17

18

19

46

This does not imply that we always know when we are warrantedin believing something. For example, the reliabilist grants justification only when the process that
produces a belief is sufficiently reliable, and it may be unreliable without our knowing
that it is unreliable. In the case described in the text, we may suppose that our mathematician is extremely reliable when she follows proofs, but, so long as we do not
requirethat she be 100%6reliable, she may be warrantedin believing a theorem that is
false. Hale's argument, on the other hand, assumes that the non-empirical process that
produces a mathematicalbelief must be 100%6reliable in orderto provide any warrantat
all for the belief. For more on the issue of whether flawed proofs may provide
justification, see the section "PurportedCounterexamples"below.
I owe this insight to an anonymous referree for this journal. See Michael Detlefsen
[1980].
Casullo, who emphasized this point in correspondence,gives an example in [1988]. The
Claudia and Elizabeth cases I discuss in the section "PurportedCounterexamples"also
show this.

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

Thus, I conclude that we should reject Kitcher's strong central requirement on a priori warrants,in favor of the weaker requirementthat beliefs
justified a priori must not positively depend for their warrant on
experience or empirical beliefs, though they may negatively depend on
experience or empiricalbeliefs.
II. Status of Belief vs. Quantity of Support
Thus far, we have characterizeda priori justification as justification of a
belief p (for X at t) that depends positively on no experience or empirical
beliefs, though it may depend negatively on experience or empiricalbeliefs,
in the sense that it may be overriddenby, e.g., experiences that cast doubt on
the reliability of the non-empirical process that produces the belief p. But
now we may wonder what will happen to X's a priori justification for
believing p in those counterfactualsituations in which it is overridden. If
X's a priori justification in those situations is eroded or eliminated, how
could it really have been independentof experience after all?
In this connection, consider anotherdistinction, between the quantity of
support provided for a belief by a process in various counterfactualcircumstances and the status that belief has in those various counterfactual
circumstances. A modest apriorist may insist that if a is an a priori
warrant, then, if a process of the same type as a were to occur in a
counterfactual situation in which there are defeating experiences or
defeating empirical beliefs, it would provide the same degree (quantity)of
support for the belief even though the status of the belief would be
different (i.e., it would fail to be knowledge or to be fully justified).20 The
modest apriorist will admit that overriding conditions may affect the
standards for knowledge or full justification-e.g., they may raise the
standardsso that the degree of supportprovided by a process is no longer
sufficient to produce knowledge. But the modest aprioristneed not say that
overriding conditions strip the process producing the belief of its ability to
provide the same degree of supportfor the belief as would be provided in
the absence of those conditions.2'
This proposal does not require that X's belief that p meet the standards
required of knowledge in every relevant counterfactual situation in order
to be justified a priori (as does Kitcher's central condition (3b)). Instead, it
requires only that, if the process that produces or sustains p is to be an a
priori warrant for X, the empirical conditions that may override X's
20
21

I owe the 'status'/'quantity' terminology to Audi.


Here is an analogy. My five-year-old,Anthony,dependsupon me for his financialsupport
in a way that he does not depend upon most other academics. Still, the practices of other
academics might shift the standards for acceptable standards of living in a way that
interfered with my ability to provide a decent standard of living for Anthony. But I
might well still be providing the same quantity of financial support.

47
A PRIORI
MODEST
KNOWLEDGE

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

claims to knowledge of p may not also underminethe quantity of support


for p provided by the process. Whereas Kitcher and Hale both assume that
an a priori warrantwill provide whatever support is needed to insure that
the belief that p has the status of knowledge in all (relevant)
counterfactual circumstances, I replace that assumption with the weaker
assumption that an a priori warrant will provide the same degree of
support for p in all (relevant) counterfactualcircumstances.
In contrast with Albert Casullo [1988], I focus on the ability of a type
of process, if it occurs, to provide a particulardegree of epistemic support.
Casullo recognizes that the epistemic support provided by a process may
not be sufficient to sustain full justification of p in the face of recalcitrant
experience, so he distinguishes between the existence of an a priori warrant
and the strength of an a priori warrant:the existence of a process which is
in fact a warrant for a belief p may be independent of experience even
though the strength of that process may not be independentof experience.
He then recommends that we requireonly that the existence of the process
be independentof experience.
However, this proposal will not do: Intuitively, it seems that
experiences other than those requiredto gain the relevant concepts may be
causally necessary for the production of X's belief that p without playing
any role in the justification of X's belief that p; if X could acquire the
belief that p without undergoing those experiences, then, all else being
equal, the belief that p would still be justified. Given Casullo's proposal,
in such cases we must deny that X's belief that p is justified a priori, but
that seems implausible, if the "extra" experiences really play no part in
justifying her belief. For example, Jaegwon Kim [1981] claims that the
empirical process of counting something (X's fingers, etc.) may play a
causally necessary role in producing X's knowledge that 2 + 3 = 5, even
though it is not essential to X's being justified in believing that 2 + 3 = 5.
In short, Kim's
... guiding idea is that the role of perception is that of a causal cue of a certain kind, not as
justificatory evidence, for the a priori truths apprehendedthrough its aid. It is what triggers
the human cognitive mechanism into appropriate action, and given the particular sort of
cognitive apparatusthat humans are genetically endowed with, certain types of perceptual
stimuli may in fact be causally necessary to generatea prioriknowledge ([1981], pp. 351-52).

Of course, the reliabilist may stipulate that the relevant process is to be


type individuatedin a way that excludes the finger counting, but that won't
help with the present problem: The empirical process of counting on one's
fingers may still be causally necessary for the occurrenceof the non-empirical process that does the justifying.

48 DONNAM. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

The moral to be drawn from these difficulties is that the mere existence
of a process, no matterof what type, is not likely to be what is relevant to
an adequate notion of a priorijustification. Instead of distinguishing the
existence of a process from the strength of that process, the modest
apriorist distinguishes the question of what quantity or degree of support
is provided for a belief (by some process) from the question of whether
that degree of support is sufficient to enable the true belief to have the
status of knowledge. If someone insists that the strength of a process will
vary in counterfactual circumstances, defenders of modest a priori
knowledge will point out that, given the distinction between quantity of
supportand status of belief supported,there is a correspondingdistinction
between two senses of 'strength': strength, = degree of support provided;
strength2= ability to close the gap between true belief and knowledge.
Defenders of modest a priori knowledge will then insist that the strength,
of a process that provides a priori justification will not vary in
counterfactualcircumstances,and that it doesn't matterwhether or not the
strength2varies.
Thus, the modest apriorist may characterize a priori justification as
justification (of p for X at t) that depends positively on no experience or
empirical beliefs, and that, though it depends negatively on experience in
that experiences or empirical beliefs may override the justification it provides, will nevertheless provide the same degree of support(have the same
strength,) in all (relevant) counterfactualcircumstances.
In The Nature of Mathematical Knowledge, Kitcher anticipates the
charge that his argumentagainst apriorismpresupposes too strong a notion
of apriority,but he insists that such a charge
... is relatively easy to rebut. Previous chaptershave shown, systematically, that the processes
which aprioriststake to generate our mathematicalbeliefs would be unable to warrantthose
beliefs against the backgroundof a suitably recalcitrantexperience. If aprioristsare to escape
this criticism on the grounds that the analysis of apriorityis too strong, then they must allow
that it is not necessary for an a prioriwarrantto belong to a type of process members of which
could warrantthe belief in question given any sufficientexperience.To make this concession is
to abandonthe fundamentalidea that a prioriknowledge is knowledge which is independentof
experience. The aprioristwould be saying that one can know a priorithatp in a particularway,
even though, given appropriateexperiences, one would not be able to know thatp in the same
way. But if alternativeexperiences could undermineone's knowledge then there are features
of one's currentexperience which are relevantto the knowledge, namely those features whose
absence would change the currentexperience into the subversive experience. The idea of the
support lent by kindly experience is the obverse of the idea of the defeat brought by
uncooperativeexperience. To reject condition (3b), the condition of my analysis on which the
central arguments above have turned, would be to strip apriorism of its distinctive claim
[1983], pp. 88-89.

MODESTA PRIORIKNOWLEDGE 49

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

I believe that I have articulateda sense in which a belief that p might be


both warranted independently of experience and defeasible: The belief
depends positively on no experience or empirical beliefs, though it may
depend negatively on experience or empirical beliefs; alternative
experiences could affect the status of the belief that p as knowledge, but
they could not affect the quantity of support it provides. Hence, in one
perfectly good sense, the belief produced is warranted independently of
experience.
III. Purported Counterexamples
Before we can rest content with the proposed characterizationof a priori
justificationas capturingthe central core of the notion of justification independent of experience, we must consider a couple of purportedcounterexamples to it.22 First, suppose that an excellent mathematician, call her
'Claudia', undergoes a psychological process of producing a proof for a
theorem, and thereby comes to believe the theorem. In the actual situation,
we suppose, Claudia is warranted in believing the theorem and she has
knowledge of it. In a bizarrecounterfactualcircumstancein which leading
members of the mathematicalcommunity perversely persuade Claudia that
her proving abilities are substandard,she is not warrantedin believing the
theorem,and she fails to have knowledge.
It is importantto notice that a modest apriorist will admit that in the
actual world Claudia is warrantedin believing the proposition in question
and has knowledge of it, while in the counterfactual situation she is not
warrantedin believing the proposition and does not have knowledge of it.
So the case described does indicate that the beliefs in question are
defeasible, certainly. But the modest apriorist insists that Claudia may
nevertheless have a priori knowledge in this world. I shall argue that the
case described above does nothing to show that the aprioristcannot have it
both ways, by granting that the relevant beliefs are defeasible and by
holding that they may be justified independentlyof experience.
Someone who presents the case described as a counterexample to the
modest apriorist's proposal wants us to characterize the case as one in
which the same type of process varies in the strength with which it
supports a belief p in various counterfactualcircumstances. However, this
characterizationseems plausible only so long as we conflate the two senses
of 'strength' outlined above. If we describe the case with the notion of
strength, clearly in mind, then, whether we describe it as a case of a priori
knowledge or not, it will turn out to be a case in which the same type of
22

50

The first was presentedby Derk Pereboomin his response to my "Against Kitcher on the
A Priori" at the Pacific Division Meetings of the APA, Spring 1988; the second was
presentedby CharlesChiharain correspondence.

DONNA M. SUMMERmELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

process provides the same degree (quantity) of support in counterfactual


circumstances.
On one hand, if we suppose that Claudia's belief in the theorem is warrantedindependentlyof experience in the sense that its justification for her
at t derives from the psychological process of proving and not from
empirical beliefs she has about the reliability of her proving abilities, then
it's hard to see how the absence of those beliefs in counterfactual
circumstances could affect the quantity of support provided by the nonempirical process that does the warrantingin the actual situation.
On the other hand, if we suppose that the belief in question is not warranted independently of empirical beliefs Claudia has about her proving
abilities, i.e., if its justification for her at t is taken to depend positively
upon empirical beliefs she has at t, then it's no wonder that a change in
those beliefs would affect the quantity of support provided by the nonempirical process plus empirical beliefs that do the warranting in the
actual situation.But in that case, the situation has been described in such a
way that the belief is not a plausible candidate for an a priori belief in the
first place. No defender of a priori beliefs will be disturbed by a case in
which beliefs whose justification is taken to depend (positively) upon
empirical beliefs are shown to be epistemically dependent on other
empirical beliefs.
Described either way, then, the case turns out to be a case in which the
same type of process provides the same degree of supportin counterfactual
circumstances.Thus, I conclude that the case described above does nothing
to count against the weaker condition on apriorityproposed by the modest
apriorist.
The second proposed counterexampleis, at least prima facie, a bit more
complicated: Suppose that anothermathematician,Elizabeth, proves, using
modus ponens somewhere in the proof, that p, where p is the proposition
"Every bounded set of real numbershas a least upperbound."But in some
possible world that is sufficient to enable Elizabeth to believe p, modus
ponens is regardedas a questionablerule of inference. Furthermore,in that
possible world, people learn very early in their education a number of
examples in which proofs involving modus ponens have yielded some
grotesque results. Suppose that, in that possible world, merely discovering
that some purportedproof uses modus ponens is sufficient to give rise to
great skepticism about the proof. Suppose that in the possible world in
question, even Elizabeth is skeptical about modus ponens. It just happens
that in this case, Elizabeth has not noticed that some step in the proof was a
modus ponens inference. (She simply overlooked that fact.) In the actual
world, Elizabeth is warrantedin believing p (that every bounded set of real
numbers has a least upper bound), and she has knowledge of it. In the

MODESTA PRIORIKNOWLEDGE 51

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

possible world described, however, Elizabeth is not warrantedin believing


p and she does not have knowledge of it.
In the possible world described, so the objection goes, not only is it not
the case that the process of going through the "proof' warrantsElizabeth
in believing p, this process does not provide the same degree of supportfor
believing p as is provided by the correspondingprocess in the actual world.
In fact, one might maintain that this "proof"does not provide any support
for p. To put the worry generated by this purported counterexample
another way: What grounds are there for claiming that alternative
experiences could affect the status of the belief that p as knowledge,
though they could not affect the quantity of support provided by the
process that warrantsit?
In the first purported counterexample, doubts were raised about the
reliability of Claudia's proof-following ability. In the purported
counterexamplejust described, there are no doubts about the reliability of
Elizabeth's proof-following ability in particular; instead, doubts are
raised about the reliability of a particular pattern of inference.
Nevertheless, though the natureof the doubts is different in the two cases,
the upshot is the same: in both cases, there are counterfactualcircumstances
in which the reliability of the process that produces the belief is brought
into question.
It thus seems to me that one way of responding to the second purported
counterexamplewould be to respondjust as we did to the first. What does
the warranting in the actual world? Is it the non-empirical process of
following a proof, or is it the non-empirical process of following a proof
together with empirical beliefs about the reliability of the process? If the
former, then, if the process that produces the belief that p in a
counterfactualsituation is really of the same type, it's difficult to see how
the quantity of support it provides could change due to the presence of
additional empirical beliefs. If the latter, then the warrantElizabeth has
for the belief that p is not a good candidate for a priori warrant.In either
case, the defender of modest a priori knowledge is not forced to admit that
the same (type of) process could provide different degrees of support in
counterfactualcircumstances.
I think this response to the proposed counter-examples is, in fact, adequate. However, I recognize that there are strong intuitions on both sides.
Confronted with the Claudia and Elizabeth cases, some people feel
strongly that doubts about the reliability of the process that produces
their beliefs lessens or eliminates altogether the support provided by the
process, whereas others feel strongly that such doubts leave the degree of
supportunchanged. Thus, I think, we need to uncover the sources of these
differing intuitions. When we do so, in section IV, we shall see that, given

52

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

distinctions to which reliabilists cling, reliabilists ought to have the latter


intuition. Insofar as they have the former intuition, they are in danger of
abandoning reliabilism. Before turning to that task, however, we shall
consider two responses to the alleged counter-examples different than the
one alreadyproposed.I shall arguethat, in each case, the response gets us no
further than the response already given. However, discussing the two
responses will enable us to uncover importantreliabilist insights that can
be used to strengthenthe initial response.
There is one difference between the two counter-examplesthat suggests
a possible reply to the second that is ruled out for the first: In the counterfactual situations described, Claudia actually believes that her proving
abilities are unreliable, whereas Elizabeth does not actually believe that
modus ponens occurs in the proof in question, since she has overlooked that
step. Though Elizabeth may discover that modus ponens occurs in the
proof, if she examines the evidence again, she is not aware of that fact at the
time in question. Thus, Claudia's justification is overridden by factors
internal to her psychological state, whereas Elizabeth's justification is
overridden by factors external to her psychological state. The question of
whether justification can be overridden by external factors is
controversial,23 so some may respond to the second counter-example by
insisting that, in the counter-factual situation, Elizabeth is justified in
believing the theorem after all. The reasoning would be something like
this: If the process of following a proof is a sufficiently reliable process,
and if Elizabeth's belief in the theorempositively depends for its epistemic
status on that process, then Elizabeth is in a strong position epistemically,
whetheror not she justifiably believes that the process is reliable.24
This response expresses an important insight: Objectively speaking,
Elizabeth is in a strong position epistemically. Though in section IV something like this kind of reasoning will turn out to be important in
responding to the purported counterexamples, it will not do as things
stand, since Elizabeth's supposed situationmay simply be modified to more
closely resemble Claudia's situation. We may simply suppose that, in the
counterfactual situation, Elizabeth does notice that modus ponens constitutes a step of the proof and yet continues to believe the theorem on the
basis of the proof. In this kind of case, reliabilists undoubtedly will
23

24

For example, Goldman [1986] and Alston [1985] both assume that only X's cognitive
state at t can undermine (Goldman) or override (Alston) the justification provided by a
suitably reliable process. By contrast, Casullo [manuscript] has argued that the
reliabilist needs to allow evidence available in a believer's epistemic community at t to
override her justification, in order to account for our ordinary intuitions about
justification. In fact, some may suggest that the Elizabeth example itself shows that
Casullo is correct.
See Alston [1985], p. 82. Note that this is not the same as saying that she believes that
the process is unreliable; of course, that belief may well overrideher justification.

MODESTA RIOR BKNOWLEDGE53

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

concede that the justification conferred by Elizabeth's proof-following is


overridden by her doubts. Even if Elizabeth is objectively in a strong
position epistemically, she justifiably thinks she is not, and so she fails to
be fully justified in believing the theorem,and she does not have knowledge
of it. Thus, this response gets us no furtherthan the initial response given
to the two counter-examples,since it provides no response to the Elizabeth
case as modified.
There is still another way in which the defender of modest a priori
knowledge might respond to the second purported counterexample: She
might ask about the nature of the doubts raised about the process. Elizabeth's proof-following abilities are, in general, reliable, so no doubts are
being raised about her particularabilities to follow a proof. Instead, there
are doubts on the part of the entire community about the validity of a
particularrule of inference. It is tempting to respond as follows: Either a
rule of inference is valid, or it is not. If it is, then, while we may have evidence thatappearsto show thatit is not valid, we may have no evidence that
really shows that it is not valid. So, we might say, if modus ponens is a
valid patternof inference, then we may have no evidence that really shows
that it is not valid. If modus ponens is not a valid patternof inference, then
any evidence we have to that affect will be a priori ratherthan empirical.
Either way, we will fail to have a case in which the degree of support
providedby a process is differentin differentempiricalcircumstances.25
However, I am not altogether happy with this response: First, I would
rathernot hinge my characterizationof a priorijustification on an assumption of necessity, as this response does; second, this response presupposes
that following a flawed proof can provide no warrantat all, and I do not
find that particularly plausible. In the remainder of this section, I shall
explore reasons that might be given for thinking that following a flawed
proof may provide warrant.Doing so will enable us to uncover important
reliabilist intuitions that will be used, in section IV, to motivate
reliabilist responses to the alleged counter-examples.
There are at least two possible reasons for thinking that following a
flawed proof may provide warrant:(1) A proof may be flawed, and yet, if
the proof-follower (X) has done everything she could reasonably be
expected to do in the situation, she may be entirely blameless. (2) X's
proof-following abilities may be extremely reliable, even though in a
particularcase, X may fail to notice a flaw in a proof.
In "Concepts of Epistemic Justification," William Alston
distinguishes deontological concepts of justification from evaluative
25

54

This response is similar to a response Hale [1987] makes to Kitcher [1983], though there
is an important difference: Hale does, but this response does not, suppose that the
process in question must continue to provide support sufficient for knowledge in all
counterfactualcircumstances.

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

concepts of justification [1985]. To a first approximation,X is justified in


the deontological sense in believing that p in so far as X does not violate
any epistemic obligations, whereas X is justified in the evaluative sense in
believing that p in so far as X's believing (in the way X does) is a good
thing from the epistemic point of view. Someone with a deontological concept of justification may take (1) as a reason for thinking that a flawed
proof may provide warrant, since X may not violate any epistemic
obligations in believing p. However, reliabilists should prefer an
evaluative concept of justification. Furthermore,reliabilists have a more
objective than subjective conception of what makes X's believing a good
thing from an epistemic point of view, since it is the actual reliability of a
process rather than the apparent reliability of a process that provides
warrant. Thus, (1) provides no reason for a reliabilist to think that
following a flawed proof may provide warrant, since X may fulfill all
epistemic obligations and yet her believing p may not be a good thing from
the epistemic point of view (e.g., perhaps the process that produces her
belief is unreliable, though through no fault of her own). Still, (2) above
does provide the reliabilist with a reason for thinking that following a
flawed proof may provide warrant.Even if X is not entirely blameless (e.g.,
she fails through neglect to notice a flaw in the proof), her prooffollowing abilities may be extremely reliable.
Of course, Elizabeth's following of a proof on a particularoccasion may
be taken to instantiate more than one process type: Elizabeth's prooffollowing at t may belong to the type "Elizabeth's proof-following over a
certain field F under conditions C," or it may belong to the type "human
proof-followings that include failing to notice highly dubious inferences."
If we type-individuatein the first way, then, by hypothesis, the process is
quite reliable, since we have supposed that Elizabeth is a highly reliable
proof-follower; if we type-individuatein the second way, then the process
undoubtedly is highly unreliable. This is an instance of the difficult
generality problem, which is a pressing problem for any reliabilist theory
of justification or knowledge.-6 I do not believe it is a problem I need to
solve here, however, since it does not arise exclusively for a priori
knowledge, and my task is not so much to defend reliabilism per se as to
defend the compatibility of a priori justification with reliabilism. Still, let
me say that I think Ernest Sosa's proposal towards a solution is promising
[1988]: Very briefly, Sosa takes justification to be a matter of the
"cognitive or intellectual virtues, faculties, or aptitudes of the subject,"
and that has to do with "the state of being a dependable source of
information over a certain field in certain circumstances." Sosa believes
that the generality problem may be solved or limited by placing
6

See RichardFeldman[1985] andJohn Pollock [1986].

MODESTA PRIORIKNOWLEDGE 55

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

restrictionson F and C that allow them to be usefully generalized upon by


the subject and by the subject's epistemic community. I agree with Sosa
that we should focus on X's intellectual capacities rather than on features
of the particular process that produces X's belief at t. In the case under
discussion, we should focus on the fact that Elizabeth is a highly reliable
proof-follower rather than on the fact that in a particular instance, she
fails to notice a dubious patternof inference. Even so, as Sosa recognizes,
we will have to limit F and C. For example, if the field F is the field of
proofs that contain extremely subtle flaws that will go unnoticed by most
good mathematicians, then even Elizabeth will turn out to be highly
unreliable. However, this way of limiting F is highly unlikely to allow it
to be usefully generalized upon by the subject and by the subject's
epistemic community. If instead we limit the field to proofs that are
within the capacities of good human mathematicians, then Elizabeth will
be reliable. Thus, I am sympathetic with the idea that following a flawed
proof may provide justification.
Recall what prompted these ruminations on flawed proof-following
and the generality problem:We were considering a possible response to the
Elizabeth example that depended on assuming that flawed proofs provide
no warrant. If we reject the assumption that flawed proofs provide no
warrant, as I have urged reliabilists to do, then the purported response
fails. The fact that either a rule of inference is valid or it is not will
provide no reason for thinking that the degree of support provided by a
process cannot be different in different empirical circumstances.Again, we
fall back on the first response given to the two alleged counter-examples.
However, insights gleaned along the way will enable us to develop that
first response in a distinctively reliabilist way. In the case under consideration (the second alleged counterexample), a reliabilist will want to say
that Elizabeth is in a strong epistemic position in the counterfactual
situation, since she is, by hypothesis, a highly reliable proof-follower.
Still, her justification in the counterfactualsituation may be overriddenby
her beliefs that modus ponens occurs in the proof and that modus ponens is
a highly questionable rule of inference (or by the latter belief plus the
evidence available to her that modus ponens occurs in the proof). In other
words, her belief in the counterfactual situation may fail to be fully
justified and/or to be knowledge. Nevertheless, since the process that
produces Elizabeth's belief in the counterfactual situation is the same
process that produces her justified belief in this world, and since, by
hypothesis, it is reliable, a reliabilist should admit that it provides the
same degree of support for her belief as is provided by the process in this
world. In the next and final section, I shall show that this assumption, far
from being a controversial new assumption for which substantialevidence

56

DONNAM. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

must be provided, is simply a condition of making distinctions reliabilists


insist on making. To show this, I will explain how three different
philosophers, given their own assumptions, should respond to the
purported counter-example to my characterization: Alvin Goldman,
William Alston, and Philip Kitcher.
IV. A Priori Justification and Reliabilist Intuitions
In Goldman's terminology [1986], Elizabeth's belief is permitted by a
right system of justification rules, though it is not fully justified, since it
does not meet what he calls the 'non-undermining condition'. In
Epistemology and Cognition, Goldman [1986] endorses the following
framework principle:
(P3) S's believing p at t is justified if and only if
(a) S's believing p at t is permittedby a right system of J-rules,
and
(b) this permission is not underminedby S's cognitive state at t.
Very briefly, Goldman conceives of permissibility in standard reliabilist
terms (i.e., as a function of the reliability of the (type of) process that produces S's belief that p), but he concedes thatreliability alone does not yield
our intuitive notion of (full) justification. So he requires for full
justification that the permissibility not be undermined by S's cognitive
state at t. This additionalconstrainton justification is designed to handle a
set of counter-examples to earlier versions of reliabilism which consist of
cases in which we suppose that S's belief is in fact produced by a
sufficiently reliable process but where p is brought into doubt in a way
that intuitively undermines S's (full) justification for believing it. "In
these cases," Goldman believes, "S's belief, or justification for believing,
that the belief in p is not permittedundermines its permittedness. That is,
although the belief is permitted, this does not intuitively suffice for its
justifiedness" (p. 62).
Notice that there is a significant difference between the two ways in
which Goldman articulatesthis last point: If we say that S's permission is
undermined, it sounds as though that permission is somehow lessened or
eliminated; if we say instead that S's belief is permitted, though not fully
justified, it sounds as though the permission is intact, even though something else is needed to make S's belief fully justified.27These two ways of
talking correspond to the differing intuitions about the cases we've been
27

It should now be obvious why I think that 'override' is preferableto 'undermine' in not
suggesting that the supportprovided by the process that produces (or sustains) a belief is
erroded.

57
A PRIORI
KNOWLEDGE
MODEST

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

discussing, since, as has been pointed out, some people intuitively feel that
doubts about the reliability of a reliable process lessen or eliminate the
supportfor p provided by that process, whereas others intuitively feel that,
if the process is in fact reliable, doubts about its reliability cannot change
that fact. However, the latter intuition is far more suited to reliabilism
than the former. The former intuition, though undoubtedly one that many
people have, simply does not fit well with reliabilism. Since the latter
intuition is central to the defense of my characterization of a priori
justification, the upshot is that reliabilists like Goldman should be
particularlyhappy with my notion of a priorijustification.
Let me explain. Notice that Goldman distinguishes sharply between
that which accounts for a belief's permittedness and the beliefs (etc.)
which may keep that permittedness from being sufficient for full
justification. If we accept the intuition that Elizabeth's support for
believing p is lessened or eliminated by her doubts about the reliability of
the process that produces her belief in p, then we blur the boundary
Goldman wants to draw. We are saying, in effect, that Elizabeth's belief
that p is not really permitted by a right system of justification-rules. We
are treating the permissibility of a belief p as a function of more than the
reliability of the process that produces (or sustains) p. As before,28what
does the warrantingin the actual world? If, on one hand, Elizabeth's belief
in the theorem is permitted in virtue of the reliability of the process that
produces it, and not in virtue of beliefs she has about the reliability of the
process, then it's difficult to see how the absence of those beliefs in
counterfactualcircumstances could affect the permissibility of her belief.
If, on the other hand, Elizabeth's belief in the theorem is permitted in
virtue not only of the reliability of the process that produces it, but also in
virtue of beliefs she has about the reliability of the process, then it's not
difficult to see how the absence of those beliefs in counterfactual
circumstances could affect the permissibility of her belief. However, in
that case we give up the central feature of reliabilism, i.e., that it is the
reliability of the process that produces or sustains a belief that justifies it,
rather than, e.g., coherence relations among beliefs. Admittedly,
reliabilists such as Goldman added the non-underminingcondition to their
theories as a concession to internalist intuitions that a belief may be produced by a reliable process and yet fail to be justified, due to featuresof the
subject's cognitive state. But I submit that reliabilists will be in danger of
giving up reliabilism altogether if they take the further step of smudging
the distinction between that which provides the warrantfor a belief p (that
which accounts for its permittednessor that on which X's belief positively

28

In my initial response to the Claudiaand Elizabethexamples.

58 DONNA
M.SUMmRIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

depends) and that which may increase the warrant needed for full
justification (that on which X's belief merely negatively depends).
Thus, a reliabilist who distinguishes permission (by a right system of Jrules) from full justification may require for (modest) a priori
justification that what accounts for a belief's permittedness be independent
of experience, even though what accounts for a belief's full justifiedness
need not be entirely independentof experience. Experience and empirical
beliefs may override a priorijustification of a belief that p, so long as they
play no (inappropriate)role in bringing it about that the belief that p is
permittedby a right rule system. If a belief that p (for X) is "permittedby
a right rule system" a priori because it meets certain conditions (e.g., was
produced in an appropriate way by a sufficiently reliable non-empirical
process), then those grounds are good ones, even though X may not be
justified in believing so due to the fact that p's permittednessis overridden
by experience or empiricalbeliefs.
This response shares with the responses rejected above the intuition
that, if Elizabeth is (objectively) in a good epistemic position (whether
because modus ponens is a valid rule of inference or because Elizabeth is a
reliable proof-follower), no doubts she has about that position may alter
that fact. However, whereas both of the rejected responses assumed that
Elizabeth is fully justified even in the counterfactual situation, this
response does not: distinguishing permission from full justification
allows us to maintain the insight that Elizabeth is objectively in a good
epistemic situation without claiming therefore that she is fully justified or
that she has knowledge.
Goldman distinguishes between permissibility and full justification;
Alston makes a similar distinction thatxan be used to make the same point.
In "Concepts of Epistemic Justification" [1985], after distinguishing
deontological concepts of justification from evaluative concepts of
justification, Alston argues that the evaluative sense of justification is
more adequate than the deontological to capturing the core of the concept
of epistemic justification. He then insists that, within that conception, we
must maintain the distinction between the grounds on which X's belief
that p is based and any reasons X may have that override the justification
provided by X's grounds. We cannot ignore "the distinction between what
provides the basis and what doesn't, and make the crucial condition [for
full justification] something like 'The totality of ... [X's] perspective
provides adequate support'." If we do ignore this distinction, given that
X's total perspective provides adequate support for the belief that p, we
will not be able to discriminate cases in which X's belief is based on the
adequate grounds from cases in which X's belief is based on totally
inadequate grounds, though X has adequate grounds available within her

MODESTA PRIORIKNOWLEDGE 59

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

perspective. Again, it seems to me, to adopt the intuition that the degree of
supportElizabeth has for believing the theorem p is lessened or erased by
her doubts is tantamount to bluffing the distinction Alston wants to
maintain.
If we say that Elizabeth's belief is based on (and thereforejustified by)
the non-empiricalprocess of proving and not on any beliefs she has about
the reliability of her proving abilities, then it is difficult to see how a
change in her beliefs about the reliability of her proving abilities could
alter the groundson which her belief is based (by eliminating or erasing the
support they provide). If we say that Elizabeth's belief is justified by the
non-empirical process of proving plus the beliefs she has about the
reliability of her proving abilities, then it is easy to see how a change in her
beliefs about the reliability of those abilities could alter the degree of
support she has in this world for her belief. But in the latter case we have
blurred the distinction Alston thinks we must maintain.
Goldman and Alston insist, each in his own way, on making a distinction between that which provides the warrantfor a belief that p (in my terminology, that on which p positively depends for its justification) from
that which is needed in addition to insure full justification (in my
terminology, that on which p negatively depends for its justification). I
have been suggesting that my assumptionabout the constancy of the degree
of support provided by a process in counterfactual conditions, far from
being a controversial new assumption for which substantialevidence must
be provided, is really just a condition of making distinctions that
philosophers such as Goldman and Alston insist on making. Thus, I
maintain, philosophers sympathetic to reliabilism should welcome modest
apriorism.
Let me illustrate my point with respect to a third philosopher-Philip
Kitcher. Whereas Alston clearly distinguishes between the basis of a belief
p and the factors that may override the justification provided by that basis,
and whereas Goldman clearly distinguishes between the permissibility of a
belief p and the factors that may "undermine"that permissibility, Kitcher
wants both to maintain and to blur the distinction, by insisting that the
process that produces a belief distinguishes knowledge from true belief and
by insisting that beliefs are not justified independentlyof their relations to
"undermining"beliefs. I shall argue that he cannot have it both ways.
Kitcher advocates a psychologistic conception of knowledge, according
to which "the difference between an item of knowledge and mere true
belief turns on the factors which produced the belief-thus the issue
revolves around the way in which a particularmental state was generated"
[1983], p. 13. Kitcherintroducesthe term 'warrant'to refer to the processes
which produce belief "in the right way," though he remains neutral on the

60

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

question of what are the appropriateconditions on warrants.Accordingly,


the first condition of Kitcher's analysis of a priori knowledge is what
Kitcher calls a "simple normal form for a psychologistic account of
knowledge":
(1)

X knows thatp if and only if p and X believes that p and X's


belief that p was producedby a process which is a warrantfor it.

Following Hilary Kornblith [1980], however, he insists that even basic


warrants29need not succeed in warrantingbeliefs independently of other
beliefs. To show this, he distinguishes between "the explanation of why a
person has a particularbelief' and "the explanation of why the belief is
warranted":
Given any case of knowledge, we explain the presence of the state of belief (which is the state
of knowledge) by describing the process which produced it, a process which is in fact a
warrant.However, to explain the state of knowledge as a state of knowledge, to show that
the process is a warrantand that the subject knows, we shall typically have to do more. Our
task will be to demonstrate that, in the particular situation in which the process produced
belief, it was able to serve the function of warrantingbelief. The reason for this is quite
straightforward.Processes which can warrantbelief given favorable background conditions
may be unableto warrantbelief given unfavorablebackgroundconditions [1983], p. 19.

Kitcher thinks that this distinction, between warrantsas "processes which


engender belief' and warrants as "processes which engender warranted
belief' enables him to escape some of the troubles that have plagued
traditional foundationalism. Foundationalists have often claimed that
foundational beliefs are justified independently of any relations to other
beliefs. Given the distinction above, however, Kitcher can say that some
beliefs are basic or "foundational"in the sense that the processes that produce them involve no other beliefs as causal factors in the productionof the
belief; nevertheless, the ability of the process that produces a basic belief
to warrantthat belief may well depend on other beliefs. In short, processes
that count as basic warrantsare independentof other beliefs as "processes
which engender belief" but often not as "processes which engender
warrantedbelief." Thus, the beliefs they produceare (in a sense) basic, even
though they are not justified independently of their relations to other
beliefs.
Notice that, in any ordinarycase of knowledge, the psychologistic conception, as Kitcher understandsit, allows that there are factors extrinsic to
29

"A process which warrants belief counts as a basic warrant if no prior beliefs are
involved in it, that is, if no prior belief is causally efficacious in producingthe resultant
belief' [1983], p. 18.

MODESTA PRIORIKNOWLEDGE 61

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

the particularprocesswhichproducesthe belief thatcan effect the status


of the belief as knowledge.And yet, Kitcherinsists thatit is the process
(and not the backgroundconditions)that distinguishesknowledgefrom
truebelief. (See (1) above, Kitcher's"normalform for a psychologistic
account of knowledge.")If Kitcherallows that it is the process that
producesa belief thatp thatdistinguishesbetweenknowledgeand mere
true belief, even thoughbackgroundconditionsaffect the status of the
belief as knowledge,then he ought to allow an analogousmove to the
apriorist:it is the (e.g.) non-empiricalprocessthatproduces(or sustains)
Elizabeth'sbelief in the theoremp thatmakesElizabeth'sknowledgeof p
a priori,even thoughempiricalbackgroundconditionsaffect the statusof
the belief as knowledge.In otherwords,if Kitchercan acknowledgethe
presenceof two factorsin the productionof knowledge,e.g., the particular
processthatproducesthebelief andthebackground
conditionsthathaveto
be met for the belief producedby the process to have the status of
knowledge,withoutgiving up the claim thatknowledgeis distinguished
from true belief by the process that producesit, then the defenderof
modestapriorismoughtto be allowedto acknowledgethe presenceof two
factorsin the productionof a prioriknowledge,e.g., the particularnonempiricalprocessthatproducesthe belief and the backgroundconditions
thathaveto be metfor thebeliefproducedby theprocessto havethe status
of knowledge, withoutgiving up the claim that a priori knowledge is
distinguishedfroma posterioriknowledgeby theprocessthatproducesit.
Of course, Kitcherwould have a responseto make to this objection.
Whereasthe mere possibilityof underminingfactorsdoes not normally
affect the ability of a process to warrantbelief and thus produce
knowledge,it does affect the abilityof a processto produceknowledgea
priori,i.e., completelyindependently
of experience.Kitcherwouldconcede
thatnon-empirical
processescan succeed,as well as empiricalprocesses,in
producingknowledge,and that,wherethey succeed,it is the processthat
produced the belief that distinguishes knowledge from true belief.
However, according to Kitcher, it is a mistake to confuse a priori
knowledge with "knowledge obtained by following a non-empirical
process." Kitcher quotes, and then responds to, an example of Saul
Kripke's:
Something can be known, or at least rationallybelieved, a priori, without being quite certain.
You've read a proof in the math book; and, thoughyou think it's correct,maybe you've made a
mistake. You often do make mistakes of this kind. You've made a computation,perhapswith
an error[end of Kripkequotation,from Naming and Necessity, p. 39]
In this case, a nonempirical process engenders belief. However, the statement known
(believed) is not known (rationallybelieved) a priori. The process producing it does not meet

62 DONNAM. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

condition (3b) on a priori warrants.Experiences which cast doubt on the accuracy of the book
(by appearing to expose errors in many "theorems," let us say), and in which eminent
mathematicians denied the conclusion, would interfere with the ability of the process to
warrantthe belief [1983], p. 43.

I think that this responsewill not do, and that Kitcherhas to face a
dilemma.On one hand, given only the distinctionbetween "processes
belief,"
whichengenderbelief' and"processeswhichengenderwarranted
it seems to me that,contraKitcher,the merepossibilityof undermining
factorsdoes affect the ability of a basic processto warrantbelief, even
thoughit does not normallyaffect the statusof the belief as knowledge.
Afterall, whatdoes the warranting,on Kitcher'saccount?We aretoldthat
a processis a basic warrantsimplyby virtueof causinga belief in a way
thatdoes not involveany otherbeliefsin the causalprocess,but the belief
that is so-caused is only justified by virtue of its relations to other
backgroundbeliefs. It seems that Kitcher'sview can more plausiblybe
construedas a versionof coherentismthanas a versionof foundationalism,
since the epistemicstatusof the so-called'basicbeliefs' derivesfromtheir
relationsto otherbeliefs.30Butif so, thenit seemsmisleadingto say thatit
is the process thatproducesa belief that distinguishesknowledgefrom
true belief; rather,Kitchershould say that coherencerelationsbetween
If Kitcherrespondsthatthe causalgenesisof
beliefs do the distinguishing.
so-called 'basic beliefs' is necessary, though not sufficient, for their
epistemicstatus,thenhe can still say only thatit is in part the processthat
producesa beliefthatdoes theepistemicwork.
On the otherhand,if Kitcherdistinguishespositivefromnegativeindependence,he can holdon to a strongformof thepsychologisticconception,
but then the modest aprioristcan make use of the same distinction.
Kitcher's distinctionbetween "processes which engenderbelief" and
belief' enableshim to say thatthere
"processeswhichengenderwarranted
of otherbeliefs,butnot that
are somebeliefs thatarecausallyindependent
thereare any beliefs that are epistemicallyindependentof otherbeliefs.
But if thereare to be basic beliefs in any epistemicallyinterestingsense,
this is not enough.Kitcherwouldrespondthatthis mustbe enough,if we
acknowledgethat even basic beliefs can be underminedby background
since thenbasic beliefs are not warbeliefsand/orexternalcircumstances,
ranted independentlyof their relations to other beliefs. Given the
distinctionbetweenpositiveand negativeindependence,however,we can
say thatbasic beliefs are warrantedindependentlyof their relationsto
otherbeliefs, in the sense thatX's warrantfor believingp does not, at t,
30

For an intricatediscussion of various versions of coherentism and foundationalism, see


Audi [1978].

MODESTA PRIORIKNOWLEDGE 63

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

positivelydependon, in the sense thatit derivesfrom,the warrantof any


otherbelief(s)S has at t.
If Kitcheracceptsthis distinction,thenhe can continueto say that"the
differencebetweenan itemof knowledgeandmeretruebelief turnson the
factors which produced3"
the belief-thus the issue revolves aroundthe
way in whicha particularmentalstate was generated."But if he accepts
this distinction,he shouldalso grantthatthe differencebetweena priori
justificationandempiricaljustificationturnson the factorswhichproduce
(or sustain)thebelief, even thoughbothtypesof justificationmaybe overriddenby empiricalbeliefs.
Consideran exampleKitchergives to show how backgroundbeliefs
may affectwarranting
power:
...Suppose that I am looking at some flowers on a table and that the circumstances of this
inspection are perfectly normal.I come to believe that there are flowers before me, and ... I do
so as the result of undergoinga process which is a basic warrantfor the belief.... Imagine that
... all the flowers on the table are genuine, but that, priorto my inspection of the table, I have,
for whatever reason, acquiredthe belief that my eyes are not functioning properly and that I
am liable to mistake ordinaryobjects for quite different things. However, this belief does not
influenceme as I standbefore the table. Behaving as I would if I did not have the belief, I look
at the table and form the belief that there are flowers before me. The process which engenders
my belief is the same as that which producedbelief in the everyday situation. But it no longer
warrantsbelief. Because I have perversely ignored my backgroundbelief about my perceptual
powers, I do not know that there are flowers before me. Thus a process which, in standard
circumstances,is a basic warrantfor belief can be deprived of its power to warrantthat belief
by circumstancesin which backgroundbeliefs are different[1983], p. 20.

Let's applythis point to the Elizabethcase. Kitcherwantsto say that,in


the actual world, the process that producesElizabeth's belief in the
theoremp has warrantingpower, even though, in the counter-factual
situationin whichElizabethhas doubtsaboutthe reliabilityof the process,
the process would be deprived of its warrantingpower. I maintain,
however,thatKitchercannotconsistentlysay bothof these things.If it is
the processalone thatdoes the epistemicworkof warranting
in the actual
situation,then, the additionof doubtsaboutthe reliabilityof the process
will not changethe (degreeof) warrantprovidedby the reliableprocess.If
it is the processplus Elizabeth'sbeliefsaboutthe reliabilityof the process
thatdoes the epistemicworkof warranting
in the actualsituation,then,a
changein herbeliefs will changethe (degreeof) warrantprovidedby the
process-plus-beliefs.
But in thatcase, it is not theprocessthatproducesthe
belief p thathas the warranting
powerandthusdistinguishesbetweentrue

31

64

He should have said "producedor sustained."

DONNAM. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

belief and knowledge; it is, at best, the process plus the background
conditions that has the warrantingpower and does the distinguishing.
I conclude that reliabilists, whether they realize it or not, need to make
distinctions that should make them particularlyhappy to accept my characterization of modest a priori justification as justification (of p, for X, at t)
that depends positively on no experience or empirical beliefs and as
justificationwhose degree of supportcannot be diminishedby experience or
empirical beliefs. Although philosophers who are not reliabilists are
equally welcome to the characterization of a priori justification offered
here, still, since reliabilists are among those who have insisted most loudly
that there can be no a priori knowledge, it is perhaps especially surprising
that, given central tenets of reliabilism, they ought to accept this
characterization.32

WORKS CITED
Alston, William [1985]: "Concepts of Epistemic Justification." The
Monist 62, pp. 57-89.

Audi,Robert[1988]:Belief,Justification,andKnowledge:An Introduction
to Epistemology. Belmont, California:Wadsworth.
_
[1983]: "Foundationalism, Epistemic Dependence, and
Defeasibility." Synthese 55, pp. 119-39.
[1978]: "Psychological Foundationalism." .TheMonist61, pp.
592-610.
Benacerraf,Paul [1973]: "MathematicalTruth."Journal of Philosophy 70,
pp. 661-79.
Casullo, Albert [manuscript]: "Causality, Reliability, and Mathematical
Knowledge."
[1988]: "Revisability, Reliabilism, and A Priori Knowledge."

PhilosophyandPhenomenological
Research49, pp. 187-213
Chihara,C. and Fodor, J. [1965]: "Operationalismand OrdinaryLanguage."

AmericanPhilosophicalQuarterly2, pp. 1-15.


Detlefsen, Michael [1980]: "The Four-Color Theorem and Mathematical
Proof."TheJournal of Philosophy, 77, pp. 803-19.
Edidin, Aron [1984]: "A Priori Knowledge for Fallibilists."
Philosophical Studies46, pp. 189-97.
32

I am indebted to Al Casullo, Penelope Maddy, Pat Manfredi, Eric Watkins, and two
anonymous referees from this journal for helpful comments on earlier drafts of this
paper, and to Robert Audi, Charles Chihara, Joel Friedman, and Dirk Pereboom for
helpful comments on "Against Kitcher on the A Priori,"a paper delivered to a helpful
audience at the Pacific Division Meetings of the APA, April 1988. Work on these issues
was initiated during an NEH Summer Seminar, "Reasons, Justification, and
Rationality," in the summer of 1987, and thanks are due to the NEH for its financial
supportand to Robert Audi for his inspirationand support.

MODESTA PRIORIKNOWLEDGE 65

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

[1986]: "Language-Learning and A Priori Knowledge."

AmericanPhilosophicalQuarterly23, pp.383-91.
Feldman, Richard [1985]: "Reliability and Justification."The Monist, pp.
159-74.
Goldman, Alvin I. [1986]: Epistemology and Cognition. Cambridge: Harvard University Press.
Hale, Bob [1987]: Abstract Objects. New York: Basil Blackwell.
Kim, Jaegwon [1981]: "The Role of Perception in A Priori Knowledge."
Philosophical Studies40, pp. 339-54.
Kitcher, Philip [1980]: "A Priori Knowledge." The Philosophical Review
89, pp. 3-23.
[1979]: "Frege's Epistemology." ThePhilosophicalReview88,
pp. 235-62.
[1982]: "How Kant Almost Wrote 'Two Dogmas'." Philosophical Topics 12, pp. 217-49.
[1983]: The Nature of Mathematical Knowledge. New York:
Oxford University Press.
Kornblith, Hilary [1980]: "Beyond Foundationalism and the Coherence
Theory."TheJournal of Philosophy 77, pp. 597-712.
Quine, W. V. 0. [1961]: From a Logical Point of View, 2nd ed., Cambridge,
Massachusetts:HarvardUniversity Press. [1st ed., 1953.]
[1963]: "Carnapand Logical Truth."In Schilpp, P., ed. ThePhilosophy of Rudolf Carnap. LaSalle, Illinois: Open Court.
Pollock, John L. ContemporaryTheories of Knowledge. Totowa, New Jersey: Rowman & Littlefield.
Rorty, Richard [1973]: "Criteriaand Necessity." Nous 7, pp. 313-29.
[1979]: Philosophy and the Mirror of Nature. Princeton:Princeton University Press.
Sellars, Wilfrid [1963]: "Empiricism and the Philosophy of Mind."

Reprinted in Science, Perception, and Reality. New York:


HumanitiesPress; London:Routledge & Kegan Paul.
Sosa, Ernest [1988]: "Beyond Scepticism, to the Best of our Knowledge."

Mind47, pp. 153-88.

66

DONNA M. SUMMERFIELD

This content downloaded from 82.21.15.145 on Sat, 22 Mar 2014 13:10:05 PM


All use subject to JSTOR Terms and Conditions

You might also like