Externalism, Internalism, and the Architecture of Justification
Alvin I. Goldman
Rutgers University
1. Internalism versus Externalism: Structuring the Dispute
The term ‘externalism’ was first used in epistemology in connection with theories
of knowledge (Armstrong, 1973). Soon, however, the terms ‘externalism’ and
‘internalism’ came to refer primarily to theories of justification (e.g., Alston, 1986/1989,
1988/1989; BonJour, 1980; Chisholm, 1989; Conee and Feldman, 2004a; Fumerton,
1988; Goldman, 1980/2001; 1999/2002; Kornblith, 1988; Plantinga, 1993). A
justification theory is classified as internalist or externalist as a function of how it
answers the question: “What kinds of states of affairs determine, or make a difference to,
the justificational status of a belief (or other doxastic attitude)?” Are these states of
affairs internal or external in character?
Factors that (help to) fix justificational status are generally called justifiers, or Jfactors. So the central question is whether justifiers, or J-factors, have an internal or
external character. Numerous types of factors, however, might influence justificational
status. What if some are internal and others external? Which side wins: internalism or
externalism? Internalism traditionally holds that justifiedness is exclusively a matter of
internal factors. If we employ this conception, internalism “wins” only if all justifiers are
internal. If any justifiers are external, externalism wins.1 This is probably the most
common way of configuring the internalism/externalism debate. Arguably, however, this
configuration makes it too easy for externalism. To level the playing field a bit, a second
configuration might also be considered, under which externalism wins only if most types
of justifiers are external. Similarly, internalism wins only if most types of justifiers are
internal. Applying this approach hinges on the admittedly knotty question of how
justifiers should be typed and thereby counted. Although this challenge might produce
some bickering, I welcome the second option along with the first. Perhaps it takes an
excess of confidence for an externalist like me to offer the “majority wins” criterion for
the terms of the debate, even as one of two reasonable ways to configure the debate. I’ll
let the reader decide at the end of the paper whether such optimism is indeed excessive.
A further complication in setting up the debate is that epistemologists commonly
admit two conceptions of justifiedness: doxastic and propositional. Under the first
conception, justifiedness and unjustifiedness are properties of doxastic state tokens
(beliefs, suspensions of judgment, degrees of credence, etc.). Under the second
conception, justifiedness and unjustifiedness are properties of ordered triples, consisting
of a subject, a proposition, and a time. Even if person S doesn’t believe proposition P at
time t, nonetheless, the person might be justified in believing P at t (given his/her
evidence). It’s unclear whether the J-factors for doxastic justifiedness coincide with
those for propositional justifiedness. In what follows, my default procedure is to focus on
1
doxastic justifiedness, but propositional justifiedness will also be featured in many
formulations.
Let us turn to the conception of a justifier. Alston (1986/1989: 192)2 defines a
justifier as “anything that affects the justification of a belief, positively or negatively”
(1986/1989: 192).3 Read strictly, the definition applies only to doxastic (not
propositional) justifiedness, and specifically to beliefs. For most of the paper, however, I
will adopt a slightly expanded definition:
(J) A justifier of any belief token or other doxastic attitude token is any property,
condition, or state of affairs (etc.) that is positively or negatively relevant to the
justificational status of that doxastic attitude.
I defer for now the question of how, exactly, “relevance” is to be understood. Obviously,
we are not talking about causal relevance. We are interested in something like
“constitutive” relevance to justificational status. How to characterize this type of
relevance remains to be explored.4
Next let us turn to the two leading conceptions of what it means for a J-factor to
be internal as opposed to external. These conceptions are accessibilism and mentalism.
There is considerable variation in how accessibilism is spelled out by different writers.
The following characterization is intended to convey the core of the idea.
Accessibilist internalism. A factor or condition J is an internal justifier for S to
have doxastic attitude D vis-à-vis p (at t) if and only if
(i)
J obtains (at time t),
(ii)
J is positively or negatively relevant to the justificational status of S’s
holding D vis-à-vis p (at t), and
(iii) J is “directly” accessible to S at t – i.e., S is capable of truly believing or
knowing, at t, that J obtains, and is capable of knowing this ‘directly’.
Mentalist internalism. A factor or condition J is an internal justifier for S to have
doxastic attitude D vis-à-vis p (at t) if and only if
(i)
(ii)
J is a mental state, event, or condition of S, obtaining or occurring at time
t, and
J is positively or negatively relevant to the justificational status of S’s
holding D vis-à-vis p (at time t).
Accessibilism is culled from a variety of authors, including Roderick Chisholm, Carl
Ginet, and Kent Bach.5 Mentalism is primarily derived from Earl Conee and Richard
Feldman.6 For each interpretation of ‘internal’, there is a correlated sense of ‘external’
that simply rejects the internalist constraint (clauses (i) and (iii) of the accessibilist
definition and clause (i) of the mentalist definition) on what it takes to be a J-factor. An
2
externalist J-factor is one that is positively or negatively relevant to a belief’s
justificational status but doesn’t satisfy the additional internalist restriction(s).
A slight emendation of the foregoing characterization of mentalist internalism
might be in order. Williamson (2000) argues that knowledge is a mental state. But
knowledge, being a factive state, is not purely internal. So if mentalists agreed that
knowledge is a mental state and that knowledge is a potential justifier, they would have to
conclude that not all justifiers are internal. This is something mentalists would certainly
wish to resist. The remedy is straightforward, however. In the definition of mentalist
internalism, simply substitute the following clause for clause (i): “J is a non-factive
mental state, event, or condition of S (occurring at time t).” This is the interpretation of
mentalist internalism I shall henceforth adopt – although I won’t bother to insert the nonfactiveness qualification.7
2. The Right-Rule Architecture of Epistemic Justification
Let me now explain the phrase “architecture of justification” that appears in the
title. The architecture I have in mind is inspired by the uncontroversial fact that
‘justification’ is a normative term. Calling a belief justified might be compared to calling
an act right, permitted, or perhaps even obligatory (in the circumstances). When entering
the normative terrain, it’s often appealing to think in terms of rules or principles that
govern the domain in question. In the moral domain, people talk about moral rules or
principles. Arguably, actions are made morally right or wrong by a combination of what
the agent does, what the circumstances are, and what the governing moral rules prescribe,
permit or prohibit. Similarly, actions are legal or illegal because of what those actions
consist in, what the circumstances are, and what the governing legal rules prescribe,
permit or prohibit. Finally, by analogy, perhaps belief tokens (and other doxastic-state
tokens) are epistemically right or wrong -- in the sense of justified or unjustified – in
virtue of the conditions, circumstances, or causes of their occurrence plus the governing
epistemic rules. Numerous epistemologists propound epistemic rules and principles in
precisely this spirit. Rules of epistemic rationality (e.g., Bayesian rules) are more
commonly displayed in the literature than rules of justifiedness, but there is plenty of
precedent for justificational rules as well.8
In all such normative domains, the power to confer (objective) rightness or
wrongness on a piece of “conduct” (behavioral or doxastic) is not possessed by any
random rule. Such power is possessed only by right, or correct, rules. Anyone can
propound a moral or legal rule, for example, one that declares all acts of kind X to be
wrong; but acts of type X won’t be (objectively) wrong unless some correct, authorized
or legitimate rule implies that they are wrong. Why should things be different in
epistemology? So I shall assume that only right epistemic rules make a difference to
genuine justifiedness. This point should be equally acceptable to both internalism and
externalism.
Once this theoretical fact is acknowledged, two important questions belong on the
table. First, which rules of justifiedness (J-rules) are correct or right? Second, what is
3
the criterion, standard, ground, or basis for the rightness of J-rules? Illustrative answers
to such questions will be discussed as we proceed. Staying at the abstract level for now,
however, we can formulate the following linkage principle, which articulates what I have
been calling a right-rule architecture. This linkage principle is also intended to be neutral
as between internalism and externalism.
(L)
S is justified in holding doxastic attitude D toward proposition p at time t if
and only if there are some conditions C and/or some cognitive process(es) R
such that
(i) S is in conditions C at (or before) t and/or S arrives at or retains D(p) at t
via cognitive process(es) R, and
(ii) one or more right J-rules jointly permit the formation or retention of D(p)
when a subject is (or was) in C and/or uses process(es) R to arrive at
D(p).
Given principle (L), the justificational status of a subject’s doxastic attitude toward a
proposition is a matter of whether the attitude conforms to what right J-rules authorize in
light of the subject’s epistemic situation or activity. Conforming to rules is here
contrasted with following rules, where rule following consists in conforming to rules as a
result of mentally representing them and “verifying” that the adopted act or attitude
conforms to them. I shall not consider a rule-following linkage principle at all. Some
theorists might prefer a slightly different type of linkage principle than L (such as a rulefollowing principle), but I will stick with L throughout.9 In the remainder of the paper,
I’ll examine prospects for epistemic internalism versus externalism in the context of the
right-rule architecture articulated by (L).
3. J-Rules and Types of Justifiers
To relate the right-rule architecture to the internalism/externalism dispute, I’ll
proceed as follows. First, I’ll examine the kinds of J-rules that hold initial promise of
having the right form and general content, according to familiar epistemological theories.
The rule schemata to be considered will have, in their antecedents, specifications of types
of conditions that the rules declare to be relevant to the permissiveness of adopting
certain doxastic attitudes toward specified propositions. If the rule schemata are on
target, the material specified in their antecedents will be satisfiable by conditions,
circumstances or processes that qualify as justifiers. As we proceed, however, I shall
argue that certain rule-schemata that can only be satisfied by internalist justifiers have to
be replaced by other rule-schemata that can only be satisfied (in the general case) by
externalist justifiers. The replacements are needed to obtain right J-rules. In addition to
justifiers that satisfy antecedents of (right) J-rules, I shall later identify justifiers that
include right rules themselves plus the criterion, standard, or ground that supports their
rightness.
It might help the reader if I say a word about how I arrived at the present form of
the argument. I originally planned a rather ecumenical paper, one that would espouse a
“synthesis,” or “blend,” of internalism and externalism. It would concede that
4
internalism is right about those justifiers that satisfy the antecedents of right rules; they
are indeed internal. But externalism is right about the contents of the criterion, or
standard, of J-rule rightness. So internalism and externalism is each right about
something important. On further reflection, however, I realized that the first thesis is
incorrect. For the most part, right rules need antecedents that cannot be wholly satisfied
by internal factors. Thus, the position I adopt here is not so ecumenical -- in keeping
with my past tendency to favor externalism over internalism.
I begin by distinguishing non-inferential and inferential J-rules. Non-inferential
rules license new beliefs (etc.) independently of their relations to pre-existing beliefs.
Inferential rules license beliefs based on logical or inductive/probabilistic relations to preexisting beliefs (etc.). Non-inferential rules would presumably appeal to such states or
events as perceptual experiences and ostensible memories, and various features of such
events. This type of rule would take roughly the following form:
“If subject S is in non-doxastic mental conditions X, Y, and Z, then S is permitted
to adopt doxastic attitude D toward proposition p.” [Type-I rule]
The inferential type of rule would take approximately the following form:
“If subject S already holds doxastic attitudes D1, D2, …, Dm toward propositions
q, r, …, s, respectively, and if proposition p bears relation R to the conjunction of
q, r, …, s, then S is permitted to adopt attitude Dn toward proposition p.” [Type-II
rule]
Relation R might be a deducibility relation or some type of inductive, abductive, or
probabilistic relation of support or confirmation.
Of course, these are not the only types of rules that might be right. In fact, they
might not be appropriate rules at all. Elsewhere I argue that appropriate J-rules would
license specified belief-forming processes or operations (Goldman, 1986). But processauthorizing rules might look excessively friendly toward externalism, especially process
reliabilism. At the start, therefore, I’ll ignore rules of that sort, to bend over backwards
not to favor, or even give an appearance of favoring, externalism. I deliberately start with
types of rules that look congenial to internalism. Whether they will be acceptable as right
rules, and will therefore genuinely support internalism, remains to be seen.
4. Non-inferential J-rules
Non-inferential rules look like friendly territory for internalism. A sample of a
type-I rule might have the following content: “If S has a visual appearance as of seeing
an airplane overhead, S is permitted to believe the proposition that an airplane is
currently overhead.” This might be generalized into the following rule: “If S has a
perceptual experience as of p’s being the case, then S is permitted to believe that p.” A
familiar objection is that this is too simplistic. What if someone undergoes an airplaneoverhead visual experience while seated in a planetarium, where all sorts of faux-celestial
5
phenomena are displayed. Doesn’t his awareness of this fact about planetaria defeat the
permissiveness of belief conferred by the visual experience? The rule needs a ceteris
paribus clause, making room for possible doxastic states that are relevant. This
complication is one I won’t pursue, to leave time and space for other complications.
A different problem for the sample rule is more relevant to the
internalism/externalism debate. The generalized version of the sample rule gets one thing
right. It provides a general schema for non-inferential J-rules, one that makes a
systematic connection between properties of a perceptual experience and a propositional
content that would figure in a permitted attitude (e.g., belief). It does this by having
matched propositional content in the antecedent and consequent. Unless something like
this is done, it’s hard to see how to formulate a perception-invoking rule with suitable
generality. At the same time, this strategy gives rise to severe difficulties.
A first difficulty is the Sellarsian dilemma. The phrase “perceptual experience as
of p’s being the case” seems to refer to a perceptual state with propositional content.
That suggests that it’s really some sort of disguised doxastic state. If so, it can plausibly
be argued that such a state cannot confer justification on a belief that p except by
inference. In that case, the rule really belongs in the category of inferential rules rather
than non-inferential rules. Furthermore, it’s doubtful that the state can confer
justification on a belief unless the perceptual state itself is justified. What makes it
justified? We need an additional class of perception-based rules, and it’s far from
transparent what they would look like.
Suppose we try to circumvent the preceding difficulty by holding that a perceptual
experience “as of p’s being the case” is a non-doxastic state that is the basis of a
disposition to believe p. The state’s content is not identical to p; rather, it is some sort of
perceptual, non-conceptual content. Nonetheless, the state tends to cause in the subject a
belief that p. Although this circumvents the preceding problems, it is unclear whether the
J-rule (schema) is right. Is an epistemic subject always justified in believing that p if she
is in a perceptual state that inclines her to believe that p?
I argue to the contrary. Suppose Reginald is in a visual state that inclines him to
believe an identity-ascribing proposition to someone he sees in a crowd. It might be the
proposition, “That’s my long-lost love, Penelope,” or the proposition, “That’s the guy
who has been pursuing me with apparent murderous intent.” In each variant of the case,
Reginald’s vision-based identification is partly caused by emotion. In the first variant it’s
an emotion of desire and longing. Reginald is so anxious to find Penelope that his visual
experience disposes him to believe that it’s Penelope, although his visual cues leading to
the belief are not very dispositive. In the second variant, fear of the pursuer distorts
Reginald’s perception, so disposing him to identify the individual in the crowd as the
pursuer despite the poverty of the visual stimulus. In both cases, I submit, Reginald isn’t
justified in holding the indicated belief.10 This cuts against the rightness of the proposed
rule-schema (under the proffered interpretation). Could it be amended so as to obtain a
right rule-schema?
6
What seems needed to obtain an acceptable rule or rule-schema is a
characterization of the perceptual cues or operations that license belief in such a
proposition. What might examples of such cues or operations be? What perceptual cues
would provide suitable perceptual evidence for what is believed? Here is an example,
taken from a simpler, well-studied domain, viz., depth perception. The visual system has
multiple types of cues by which it estimates an object’s distance from a subject. Two
such cues are binocular disparity (the difference in images presented to the two eyes
because of their different perspectives on the object) and eye convergence (the angle
between the two eyes’ lines of sight, i.e., the extent to which the two eyes are turned
toward each other to fixate an object, especially a nearby object). These conditions
provide evidence to the visual system in its computation of an object’s distance. Perhaps
a subject whose visual system undergoes the right sorts of binocular disparity cues or eye
convergence cues is thereby justified in believing that the observed stimulus is a certain
distance from him. Borrowing pertinent details from vision science, we might be able to
construct a J-rule that fits the circumstances in which subjects make accurate distance
judgments based on binocular disparity or eye convergence.11
Rules of this sort, however, would not be congenial to internalism; neither to
accessibilism nor to mentalism. The problem for accessibilism is that the cues associated
with binocular disparity and eye convergence aren’t conscious. So presumably they
aren’t directly accessible to the subject. What about mentalism? It isn’t clear whether
this is equally severe a problem for mentalism because it isn’t clear what qualifies as
“mental.” It is highly plausible, however, that being the mental state of a person requires
more than being a representational state of some cognitive module in that person’s brain.
In most cases concerning the cues or operations of the visual system that enable people to
form accurate visual beliefs, these cues are only states or operations of one or more
modules of the visual system, not ascribable to the person, and hence not mental states of
the person.12 If this is right, it seems impossible to construct any plausible perceptionbased J-rules with antecedents that meet internalist strictures.
5. Inferential J-Rules
Next let’s consider inferential rules, type-II rules. Our initial sample of a type-II
rule schema has a conjunctive antecedent. The first conjunct specifies antecedent
doxastic states that the subject is in; the second conjunct specifies some sort of deductive,
inductive, or probabilistic relation R that holds between the contents of the antecedent
doxastic states and the target proposition p. In considering whether this rule-schema
would meet internalist criteria, and whether such a rule-schema would be right, I’ll start
with the second conjunct in the antecedent, the one that invokes some relation R.
The fact that a certain logical, inductive, or probabilistic relation holds between
specified propositions is a fact independent of any individual’s mind (Comesana, 2005;
Goldman 1999a/2002). The relation of deducibility between ‘Socrates is mortal’ and the
conjunction of ‘All men are mortal’ and ‘Socrates is a man’ is not a mental fact or
condition. It is independent of any and all minds. Thus, factors or states of affairs that
instantiate an antecedent of a type-II schema, under our initial proposal, will include
7
states of affairs that aren’t mental states of subject S. These factors clearly won’t qualify
as internal according to mentalism. Would they qualify as internal according to
accessibilism? Much depends on what accessibilists mean by “direct” access. But surely
there are many relations of the intended type that aren’t directly accessible to most
epistemic agents. Extremely complex deducibility relations, holding between a
conclusion and a huge set of internally complex premisses, will not be “directly
recognizable,” or recognizable at all, for most epistemic agents. So even under the
accessibilist notion of ‘internal’, many conditions or states of affairs that could instantiate
part of an antecedent of a type-II rule would not qualify as internalist J-factors.
Internalists sometimes respond to this kind of problem by trying to “internalize”
the external facts that are initial candidates for justifiers. Conee and Feldman
(2001/2004) present an example about a logic teaching assistant (TA) and a beginning
logic student who are looking over a homework assignment together. One question in the
assignment features a sentence that both know to express a truth and asks whether certain
other sentences are true as well. The student is clueless, whereas the TA can easily tell
that some of the other sentences are logical consequences of the original one; so the TA is
justified in believing that they are true as well. Conee and Feldman comment as follows:
Again there is an internal difference between the two [the TA and the beginning
student]. The difference is that the TA has justification for her beliefs to the
effect that certain propositions validly follow from the original ones. She is
expert enough to “see” that the conclusions follow without performing any
computations…. [J]ust as in example 3, relevant internal differences make the
difference. (2001/2004: 60)
The moral they draw from this and similar cases is the following:
It is reasonable to generalize from these examples to the conclusion that every
variety of change that brings about or enhances justification either internalizes an
external fact or makes a purely internal difference. It appears that there is no need
to appeal to anything extramental to explain any justificatory difference. These
considerations argue for the general internalist thesis that these epistemic
differences have an entirely mental origin. (2001/2004: 61; emphasis added)
What do Conee and Feldman mean by “internalizes an external fact”? A natural
interpretation is that S internalizes an external fact that p just in case S believes a true
proposition p. Pursuing this hint, Conee and Feldman might ask us to rewrite our schema
for type-II rules as follows, where the new version adds a belief operator to the second
conjunct of the antecedent:
“If subject S already holds doxastic attitudes D1, D2, … Dm toward propositions q,
r, … s, respectively, and if S believes that proposition p bears relation R to the
conjunction of q, r, … s, then S is permitted to adopt doxastic attitude Dn toward
proposition p.” (Type II rule, first revision)
8
This rewriting of the rule-schema eliminates reference to any external fact that posed a
problem for internalism. I shall argue, however, the new formulation is both too strong
and too weak.
It is too strong because it’s implausible to require people to have beliefs about
logical, inductive or probabilifying relationships as a precondition of having justified
belief in conclusions that “rely” on such relationships. To get justified beliefs in those
conclusions, it’s sufficient to use psychological operations or computations that “realize”
or “incorporate” corresponding principles. Ordinary cognizers rarely have a theoretical
grip on these principles, or even have the concepts (e.g., validity) in which the principles
or relations could be expressed. Hence, unlike the logic TA described by Conee and
Feldman, they lack beliefs in the appropriate principles (e.g., principles about validity).
Nonetheless, their inferred beliefs can be justified. So the new rule-schema implicitly
places too stringent a demand on inference-based justifiedness.
The schema is too weak in a different respect. By requiring mere belief in the
holding of the relevant relation, rules that fall under the schema would allow a conclusion
to be justifiedly inferred even though belief in the authorizing relationship is either false
or entirely unjustified. Either of these scenarios is inappropriate. Conee and Feldman
tacitly acknowledge the point about justifiedness. In their discussion of the TA example,
they describe the TA (in part) as follows: “the TA has justification for her beliefs to the
effect that certain propositions validly follow from the original one.” This sentence
implies (or implicates) that unjustifiedly believing that certain propositions validly follow
from the original one would not guarantee a justified belief in the conclusion (“such-andsuch a sentence is true”).
The revision is also too weak because the revised rule-schema no longer requires
an objective, “external” relation of validity or strong support to hold between the contents
of the antecedent beliefs and a newly inferred conclusion. That makes the schema too lax
or undemanding. Suppose someone believes that affirming the consequent is a valid
form of inference, and derives a belief in some proposition based on affirming the
consequent. Clearly, his newly formed belief isn’t justified. But it would be justified
according to the type-II rule, first revision. If there is no (“external”) logical, inductive,
or probabilistic fact of an appropriate sort, simply believing that there is one doesn’t
confer justifiedness on a resulting conclusion. Thus, the problem cannot be solved by
formulating J-rules that “internalize” external facts, as Conee and Feldman express it.
Right J-rules must feature objective, external (logical or confirmational) relations, a
requirement that cuts against (strong) internalism of all stripes.13
In response, it might be argued that right inferential rules don’t need to feature
any conjunct of the second kind in their antecedents.14 The right sort of formal
relationship between antecedent beliefs (premises) and conclusion beliefs can be
specified in the relations between the antecedent and consequent of the J-rule schema
itself. For example, there could be a J-rule schema of the following sort: “If S already
holds pairs of beliefs of the forms q and (if q then p), then S is permitted to form a new
belief vis-à-vis proposition p.” This would implicitly capture the pertinent formal
9
relationship and eliminate the need to specify any relation R in the antecedent. There are
two problems with this proposal. First, formal relationships of the relevant kind are
known to exist only in the deductive sphere, not in the inductive or abductive domains.
So it isn’t clear how this would work in the latter cases. Second, and perhaps more
importantly, there is no known way to borrow even principles of logic and incorporate
them directly into correct epistemic rules. For example, if you believe q and believe (if q
then p), it’s not obvious that an epistemically permissible move to make is to believe p.
Maybe you should instead abandon belief in either q or (if q then p), rather than to adopt
p (see Harman, 1986; Christensen, 2004). Nonetheless, whatever correct inferential rules
would look like, logical and inductive connections among propositions will be important.
So it looks very much as if a second conjunct is needed in the antecedent that specifies
types of (“objective”) relations among propositions.
Let us turn now from the second to the first conjunct of the antecedent of the typeII schema. The first conjunct also has an important inadequacy, which is also hinted at
by Conee and Feldman’s logic TA case, as well as by a familiar strand of epistemological
theorizing. Epistemologists of justification generally agree that inferences only transmit
justifiedness from prior beliefs to new beliefs. If the initial beliefs themselves lack
justificational “juice,” no such juice can be transmitted to a new belief. Subjects starting
with unjustified beliefs as premises cannot arrive at a justified conclusion belief by any
form of inference, no matter how impeccable. Thus the type-II schema must be revised
once again if it is to generate genuinely correct J-rules. It should read as follows:
“If subject S already holds justified doxastic attitudes D1, D2, … Dm toward
propositions q, r, … s, respectively, and if proposition p bears relation R to the
conjunction of q, r, … s, then S is permitted to adopt doxastic attitude Dn toward
proposition p.” (Type-II rule, second revision)15
This emendation poses another problem for the thesis that all justifiers are
internal. Uncontroversially, believing a proposition is an internal matter (at least under a
mentalistic criterion of the internal). But it’s doubtful that justifiedly believing a
proposition is equally an internal matter. Under our latest revision, however, this type of
condition will be the most prevalent type of J-factor under type-II rules. If all such Jfactors are external J-factors, this would be a big blow to internalism. However, I
contend that all such J-factors are external.
On what grounds do I base the suggestion that justifiedly believing a proposition
is an external rather than an internal matter? Doesn’t this beg the central question
between internalism and externalism? Internalists like Conee and Feldman (2001/2004)
claim that justifiedness supervenes on mental states that occur at the time of belief.
Doesn’t this imply that justifiedly believing a proposition is an internal matter? No. To
say that something supervenes on states of a certain kind doesn’t imply that it is a state of
that kind. So even on the Conee-Feldman approach, justifiedly believing isn’t a mental
state. Nonetheless, if Conee and Feldman are right in holding that justifiedness
supervenes on current mental states, doesn’t that suffice for present purposes? If
justifiedness supervenes on current mental states, then (borrowing the notation from the
10
rule-schema), the current mental states on which justified attitudes D1, D2, …, Dm toward
q, r, …, s (respectively) supervene will be the justifiers of doxastic attitude Dn vis-à-vis p.
I shall now argue against the thesis that justifiedness supervenes on the mental
states obtaining at the time of belief. Specifically, I’ll argue that the J-status of a belief
held at time t partly depends on its psychological history, on what transpired in the
subject’s cognitive history prior to t. The historicity of justifiedness is a major problem
for internalism.
6. Preservative Memory and the Historicity of Justifiedness
On the basis of testimony, Ursula acquires an initially justified belief in a certain
generalization, G. This occurs at time t1. She subsequently encounters a mountain of
counterexamples to G, each of which is very obvious. The joint effect of these
counterexamples, at time t5, is the total defeat of Ursula’s original justification for G.
Now she is justified in rejecting G, not believing it; hence she is unjustified in believing
it. Nonetheless, Ursula continues to believe it. Each reader may imagine details of the
defeat to suit his/her own epistemological taste. This should be no problem for any
internalist, because internalists routinely appeal to evidential defeat to account for an
attitude’s unjustifiedness. The rough idea is that Ursula either fails to consider how the
counterexamples she observes bear on generalization G (an epistemically culpable
omission), or she considers their bearing each time but fails to appreciate their
falsification of G. Much later, at time t20, Ursula still retains the same high level of
credence in G as she held originally. This seems to be a clear case in which Ursula’s
belief in G at t20 is still unjustified. This conclusion holds up even if we further assume,
as I now stipulate, that at time t20 Ursula has forgotten the defeating evidence she
encountered at time t5.
The Ursula case poses a serious problem for internalism. Her being unjustified in
believing G at t20 is incompatible with mentalism because Ursula has no mental states at
t20 that jointly constitute defeaters to her original justification for G. All such defeating
mental states – her observations of the counterexamples -- lie in the forgotten past.
Similarly, it is incompatible with accessibilism, because at time t20 those past defeating
states are not directly accessible to her. So the unjustifiedness of Ursula’s belief in G (at
t20) cannot be accommodated by either brand of internalism.
Let me amplify my analysis of the Ursula case with the help of a broader
treatment. The transmission of justifiedness over time is underpinned by what Burge
(1997) calls “preservative memory.”16 As Burge writes, “a person clearly can be entitled
to believe a theorem she believes because of preservative memory even if she cannot
remember the proof she gave long ago, and even if she cannot remember that she gave a
proof. Most of what one is entitled to believe from past reading, past interlocution, past
reasoning, or past empirical learning, derives from sources and warrants that one has
forgotten” (1997: 38). Returning to our J-rule framework, we might formulate a right
rule of justification preservation through memory:
11
(MP) If S has a justified attitude D toward proposition p at time t, and if, via
preservative memory, S retains D toward p at the later time t’, then, ceteris
paribus, D is (still) the permitted attitude for S to take toward p at t’.
The ceteris paribus phrase is intended to accommodate the possible cancellation of
justification transmission if new evidence arises between times t and t’, as in the Ursula
case (between t1 and t5). Rule (MP) straightforwardly explains why Ursula is unjustified
in believing G at the later time. Her being justified in rejecting G at t5 is a principal
determinant of her being justified at t20 in rejecting it. Hence she is unjustified at t20 in
believing it. However, being justified in rejecting G at t5 is only a defeasible determiner
of her being justified in rejecting it at t20. Had she encountered evidence at t20 that
defeats her earlier defeaters of G, she might might have been justified in believing G at
t20. I would say that both Ursula’s observation of the counterexamples at t5 and her being
justified in rejecting G at t5 are relevant to her J-status at time t20. This implies that both
of these states of affairs are J-factors vis-à-vis this J-status. Our definition of “J-factors,”
it will be recalled, requires only positive or negative relevance, nothing stronger.
Given the importance of the Ursula case to my overall argument, internalists will
surely want to resist a principle like (MP). I shall argue that resisting any such principle
will deliver a major victory for knowledge-skeptics, a gift to skepticism we have no good
reason to bestow.
Stored beliefs are beliefs lodged in memory and not currently retrieved or
activated. They reside outside the sphere of consciousness. Nonetheless, stored beliefs
can have a J-status. Indeed, if we assume a justificationist approach to knowledge (i.e.,
knowing entails being justified), it is essential for stored beliefs to have a J-status. It is
widely assumed by both ordinary people and epistemologists that normal adults have a
vast storehouse of commonsense knowledge. Only a tiny fraction of what they know,
however, is occurrent or conscious at a given time; the remainder of their knowledge
consists of stored items of knowledge. If an epistemologist insists that very few stored
beliefs are justified, this would imply a major diminution in people’s stock of knowledge
at any given moment (as compared with our commonsense assumption)
Let’s see how the absence of (MP) and any similar principle from the list of right
J-rules would produce a major diminution in justified stored beliefs. If (MP) is rejected,
there will be a phenomenon that I’ll call recurrent clearing of the justificational slate
(RCJS). There will be no carry-over of a belief’s J-status from moment to moment. If a
justified stored belief is in Alfred’s head at time t, a memory-generated continuation of
that belief in Alfred’s head in the next moment won’t inherit justification from the prior
moment’s belief.
This doesn’t yet prove that omission of (MP) would drastically reduce the number
of stored justified beliefs (at any given time). Perhaps many stored beliefs are
evidentially (i.e., inferentially) supported by other simultaneous stored beliefs. They
would receive justification from a non-mnemonic source. (Evidential support from
occurrent beliefs at the time is also possible, but occurrent beliefs at a given moment are
12
usually vastly fewer in number and not usually content-congruent with the wide
assortment of stored beliefs we learned long ago, so they wouldn’t help very much.)
Assistance from other stored beliefs, however, runs into familiar problems when (MP)
has been excluded. Other stored beliefs can be helpful only if they themselves are
justified. How many of them will be justified if they cannot inherit justificational juice
via memory preservation? The RCJS phenomenon rears its ugly head. The other stored
beliefs, then, must also receive evidential support from still further stored beliefs. If this
regress is to be stopped, along familiar foundationist lines, it presumably stops with
perceptual beliefs. But no stored beliefs are perceptual beliefs in the normal sense. The
only perception-related beliefs that might help will be beliefs immediately generated by
perception and then retained by memory. These are not immune, however, to the familiar
obstacle that RCJS poses. Finally, as a contingent matter of fact, people aren’t very good
at retaining specific perceptual observations made in the past, especially the remote past.
What pages of print did I observe when I read sentences expressing facts about Julius
Caesar or the Roman Empire that I still believe? I haven’t a clue.
In discussing the epistemology of memory internalists usually take a rather
different tack from the one I have been pursuing. Memory’s contribution to justification
is usually taken to rest on what is called “apparent memories”, where an apparent
memory is a conscious event understood on a perceptual model. In other words,
seeming-to-remember that p is conceived of as a non-doxastic conscious event analogous
to seeming-to-perceive that p, which confers (prima facie) justifiedness on the indicated
proposition for analogous reasons. In this fashion, a certain kind of conscious memory
event affects a subject’s J-status toward a proposition without appeal to events or states of
affairs in the subject’s past.17
Internalists of this stripe are likely to respond to the Ursula case as follows.
Suppose that Ursula, at t20, consciously considers whether G is true and has an apparent
memory of its being true. Then surely she is justified in believing it. The apparent
memory will confer prima facie justifiedness on her belief in G, and this will be bolstered
by a background (justified) belief that her apparent memories are usually reliable.
Finally, no other apparent memories will defeat her prima justifiedness with respect to
believing G, since she has forgotten the counterexamples to G she once observed. So our
earlier verdict on the Ursula case, internalists will urge, was a mistake.
Before considering this response, we should first notice that the internalist here
substitutes a new version of the Ursula case in place of my own. My preferred version of
the case – call it “Ursula-A” – is one in which Ursula doesn’t have an apparent memory
of G at t20. In Ursula-A, the belief in G remains in storage at t20. What does the
internalist say about Ursula-A? The internalist has no basis, from his own theory, to say
that Ursula is justified at t20 in believing G. The internalist must concede that the belief
in G in Ursula-A has no J-status at all, because of the RCJS problem, which leads to a
substantial skeptical upshot. This is too a high price to pay for rejecting principle (MP).
So we should keep (MP), with its straightforward implication that justifiedness doesn’t
supervene on a subject’s current mental states.
13
What shall we say, however, about the internalist’s version of the Ursula story
(call it “Ursula-B”) where Ursula consciously thinks about G and experiences an apparent
memory that G? Consider an analogue in the domain of inference. Suppose Leonardo is
interested in establishing a complex scientific or mathematical proposition w, the
argument or proof of which is quite intricate. Leonardo starts with propositions a, b, and
c, all of which he believes justifiably. From these he infers proposition d, which he then
combines with e to infer proposition f. So it goes until he arrives at w several weeks
later. Leonardo has total confidence in his memory and therefore doesn’t record his
inferences in any form that can be consulted. At the later time (call it t20, in parallel with
Ursula) he has forgotten the details of his early inferences. Now, however, he asks
himself whether w is the case and experiences a vivid apparent memory that it is.
According to internalists, his apparent memory should confer prima facie justifiedness on
w, and there is no other evidence in Leonardo’s current possession that defeats this prima
facie justifiedness. So Leonardo’s belief in w is ultima facie justified according to
internalism. The full truth of the matter, however, is that Leonardo used very bad
inference processes in the early stages of his reasoning. His final few steps of inference
(or proof) were fine, but his earlier inferences were quite improper. What shall we say,
then, about the J-status of Leonardo’s belief in w at t20? Clearly, Leonardo isn’t justified
in believing w at t20.18 This is despite the fact that his epistemic situation precisely
parallels that of Ursula-B. I conclude that the internalist account just isn’t right. In both
the Leonardo case and the Ursula-B case prior states are justificationally relevant to
subsequent J-status. This solidifies the argument for the historicity of justifiedness.
Once it is recognized that J-status doesn’t supervene on current mental states, the
problem ramifies. As we now recognize, the revised right-rules schema of the type-II
kind involves a justification condition in its antecedent that doesn’t satisfy internalist
criteria. So all states of affairs that instantiate that type of condition are externalist
justifiers rather than internalist justifiers.
One option for (mentalist) internalism is to abandon their original claim that
justification supervenes on current mental states. If internalists admit the justificational
relevance of past states, none of the problems raised in this section will apply. This
would amount, however, to an abandonment of traditional internalism. Traditional
internalism, the kind inspired by Descartes, aspired to show how ordinary beliefs could
be justified from the starting-point of one’s current mental states and nothing more.
“Solipsism of the moment” is the classical epistemic predicament that internalism has
sought to address and resolve. This is clearly enunciated by Chisholm, perhaps the
leading internalist of the preceding generation:
A consequence of our “internalistic” theory of knowledge is that, if one is subject
to an epistemic requirement at any time, then this requirement is imposed by the
conscious state in which one happens to find oneself at that time. (1989, pp. 5960; second emphasis added)
To abandon the currency requirement (“at that time”) would be a major departure from
traditional internalism and a major concession to externalism.
14
7. Other Externalist J-Factors: Right J-Rules and the Standard of J-Rule Rightness
Let’s review where our argument stands. Two types of J-rule schemata have been
considered, one for non-inferential rules and one for inferential rules. In the case of noninferential rules, it initially looked promising that right rules might invoke J-factors that
are internalist in character. But serious troubles arose, which I won’t now review. Even
in this favorable-seeming territory, internalism sailed into rough waters. Turning to
inferential J-rules, prospects for internalist J-factors ran into two serious problems. First
was the problem of the R-relations. Suitable R-relations seem to be non-internal, and no
attempt to “internalize” them looks promising. The second problem concerns the input
conditions for right J-rules. Right inferential J-rules require input conditions specifying
justified doxastic states as inputs. This (re-) raises the question of whether the
justifiedness of doxastic states is an internal state of affairs. We then showed that past
evidential states or inferential processes can be positively or negatively relevant to a
subsequent state’s J-status. This was shown both for preservative memory and for chains
of inference or proof. The relevance of past evidential states implies that the J-status of a
current doxastic state does not, in general, supervene on current mental states (or
accessible states and processes). Thus, states of affairs that qualify as J-factors under
right J-rules are not purely internalist in character.
In short, there is bad news for internalism, a verdict that holds both under the
demanding terms of debate in which internalism wins only if all (types of) J-factors are
internal and under the less demanding terms of debate in which internalism wins only if
most (types of) J-factors are internal. Externalist types of J-factors predominate in the
category of inferential rules, and the same may hold for non-inferential rules. In this
section the bad news gets worse. I present two further categories in which externalist Jfactors are likely to predominate. The two categories are right J-rules and the ground, or
standard, for J-rule rightness.
Under the right-rule architecture, right J-rules qualify as justifiers. After all, a
right J-rule is positively or negatively relevant to the justificational status of any token
attitude to which it applies. Given that right J-rules are justifiers, is their being right an
internalist or externalist matter? That depends on what makes a right J-rule right. J-rule
rightness is something conferred by the criterion, or ground, of rightness. If a J-rule
meets the criterion, or ground, it’s right; if it doesn’t, it isn’t. Is meeting such a criterion
an internalist matter? Let’s examine this first under the mentalist approach and then
under the accessibilist approach.
Is meeting a rightness criterion a matter of an individual subject’s mental states?
Presumably not. For any selected criterion, whatever its content, meeting or failing to
meet the criterion is an objective matter of fact, not a matter of somebody’s belief or
other mental state.
15
Is meeting a rightness criterion directly accessible to any pertinent subject?
Direct accessibility consists of being able to determine correctly, though directly, that the
state of affairs in question obtains. Is the rightness of a J-rule determinable correctly and
directly? The question should be clarified: determinable by whom? I assume that the
answer is: by the average epistemic subject, a person on the street. Surely the average
person on the street cannot decide correctly which J-rules are right. Ordinary people
never consider such specialized issues, deep within the recesses of epistemology. The
average person on the street never dreams of these questions, much less their answers.
He or she has no clue as to what a standard or criterion of J-rule rightness is; nor would
he or she be in a position to determine which J-rules meet such a standard. Hence, on
neither approach to what constitutes internality does the rightness of a J-rule qualify as an
internal state of affairs. In this category, like most if not all of those examined earlier, Jfactors turn out to be external rather than internal.
Finally, what about the criterion, standard, or ground of J-rule rightness? Is this
criterion (whatever it turns out to be) itself a justifier of beliefs and other doxastic
attitudes? Again, the answer seems to be affirmative. After all, the justificational status
of each token attitude is a function, positively or negatively, of a specific right J-rule, or
the membership of the total set of right J-rules. But the membership of the set of right Jrules is a function of the standard, ground, or criterion of J-rule rightness. So, by
transitivity, the justificational status of each token attitude depends on the standard,
ground, or criterion of J-rule rightness. Hence, this standard falls in the category of a
justifier.
Is this justifier internalist or externalist? That is, does the content of the rightnessstandard invoke internalist or externalist properties or states of affairs? An obvious
externalist candidate for the rightness standard would invoke truth-conduciveness -- for
example, conduciveness to a high truth-ratio (or veritistic value19) among the beliefs that
would result from conforming to a rule. The standard might say, for example, that a Jrule U is right if and only if, among the J-rules that compete for a certain category of Jrules, conformity with U would maximize the truth-ratio (or veritistic value) of the
doxastic attitudes that would be generated if epistemic subjects conformed their doxastic
“conduct” to U (as compared with the truth-ratio that would be generated by conforming
their doxastic conduct to alternative rules).
At this point I am not trying to argue that the (correct) standard of J-rule rightness
is externalist. I am just trying to clarify the issue and sustain the thesis that the standard
itself (i.e., its content) is among the justifiers of token beliefs. The content of the
standard is one thing that indirectly helps to fix the justificational status of doxastic
attitudes.
No doubt other candidates for the standard of J-rule rightness would feature
internalist contents. However, it might take more than a little delicacy to formulate some
plausible ones. There is reason to suspect that certain internalists – viz., mentalists –
would resist the idea that a rightness criterion is needed. Conee and Feldman
(2001/2004) defend internalism primarily through the vehicle of a supervenience
16
principle (their principle “S”). Although they don’t explicitly say that this supervenience
principle provides a full theory of justifiedness, they strongly imply this. It’s important to
indicate, therefore, that this is incorrect. A supervenience principle like “S” doesn’t
provide everything a theorist wants to know about justifiedness.
To explain this important point, consider an analogy. Suppose a moral
consequentialist holds that the moral rightness of actions supervenes on their
consequences. Is this a (theoretically) complete story of moral rightness? Hardly. It
leaves open which (types of) consequences confer rightness and which ones confer
wrongness. Is rightness fixed by the pleasure and pain that actions cause, or by other
types of consequences (e.g., the surprise that they occasion)? Even being told that the
only relevant consequences are hedonic consequences leaves a crucial question
unanswered: Does greater pleasure conduce to rightness or to wrongness? Supervenience
only guarantees that if two items share the same base properties, they also share the same
supervening properties. But supervenience per se doesn’t tell us which properties
supervene on which base properties.
It is similarly under-informative to be told that epistemic justifiedness supervenes
on a subject’s mental states. Even if this is true, it wouldn’t tell us everything relevant.
We also need to be told what combinations of mental states give rise to positive
justifiedness and what combinations give rise to negative justifiedness. Knowing that
mental states M1, …, Mk jointly fix the J-status of an attitude toward q wouldn’t indicate
whether the resulting J-status is positive or negative. This is why right J-rules are
needed, and, at a higher level, why a criterion of J-rule rightness is theoretically desirable.
Specifying a supervenience base (e.g., the class of mental states) leaves a great deal
indeterminate. This indeterminacy is what a truth-conduciveness criterion might resolve.
I do not mean to imply that no internalist rightness criterion could possibly be up to the
challenge. But it must be shown that an internalist rightness criterion is up to the
challenge. It won’t do to dismiss the challenge on the ground that no rightness criterion
is needed.
Let me now sketch an argument in support of an externalist, truth-conduciveness
approach as the right approach to the job. The argument is simply that such a criterion
would provide a satisfying explanation of why epistemologists find certain candidate Jrules right (or roughly right) and others wrong. I’ll illustrate the point first with respect
to non-inferential J-rules and then with respect to inferential J-rules.
Why do so many examples of non-inferential J-principles center on perceptual
experience, especially where the epistemic subject is in “good” perceptual
circumstances? Because these are cases in which beliefs formed in accordance with these
experiences are usually true. Or -- of equal relevance from an explanatory perspective -because these are cases in which epistemologists assume, and assume that others assume,
that believing in accordance with the recommended principles leads predominantly to
truths. What I say is intended to apply not only to reliabilists, who explicitly invoke a
truth-conduciveness story. I also speak of epistemologists who offer an entirely different
theory, but still adduce similar principles. I claim that the underlying appeal of these J-
17
principles is a tacit recognition that they are truth-conducive, even when this isn’t the
official doctrine being endorsed.
A similar example can be drawn from inferential J-rules. What R-relations do
epistemologists regularly choose to incorporate into inferential rules? Relations such as
deducibility, strong confirmational support, or inference-to-the-best-explanation. We
don’t choose epistemic rules in which the evidence-hypothesis relation features weak
probabilistic support, or inference-to-the-worst-explanation! Why? Because they would
yield a much lower rate of true-belief formation than the preferred principles. Or so it is
assumed. What apparently underpins the choice of congenial J-rules is their tendency to
produce a fairly high ratio of true beliefs. So that is the appropriate choice of a standard
of J-rule rightness. Even if some of these rules are criticizable for one reason or another,
the best unifying explanation of their appeal to epistemologists is that epistemologists
expect good epistemic practice to be the kind of practice that is comparatively truthconducive.20
If all this is right, the J-factor at the level of the rightness criterion is an externalist
factor. This adds yet another type of externalist J-factor to our list, covering many
categories of justification determiners. So it appears that externalism wins the dispute
with internalism even when the playing field is made more congenial to internalism than
it normally is.
However, we haven’t examined possible rightness criteria from the internalist
camp. One such candidate comes from Foley (1987).21 According to Foley a belief is
rational if and only if it conforms to the believer’s most deeply held principles about
promoting the epistemic goal of now believing the truth. Translating this into a proposal
for justification rather than rationality, this suggests that a J-principle is right just in case
it is one of the believer’s most deeply held principles about getting the truth. Being
deeply held certainly looks like an internalist matter. It takes little reflection, however, to
notice that Foley’s criterion is highly relativistic. A principle deeply held by one
epistemic subject is not necessarily deeply held by another. This makes Foley’s approach
ill suited to the objectivist, non-relativistic spirit of our entire framework. As explained
at the outset, our right-rule architecture links justification to conformity with right rules,
and it looks as if Foley’s approach gives no sense to person-independent rightness of
rules. A given rule may be right-for-Smith but not right-for Jones, with no sense to made
of right simpliciter. It’s not even clear, then, that one can treat Foley’s criterion as a
rightness criterion. Instead it’s a relative-rightness criterion, a beast of another kind
altogether.
8.
Can Internalism be Salvaged by a New Way of Defining Justifiers?
After reviewing my case for externalism, readers might suspect that I have been
unfair to internalism from the outset, unfair in my characterization of justifiers. I adopted
Alston’s characterization of a justifier as any state of affairs that positively or negatively
affects justification. This is too broad, it might be objected. Alston himself describes
justifiers rather differently in another paper. In “An Internalist Externalism” (Alston,
18
1988/1989) he introduces the notion of a “ground” as a state that an epistemic subject
“goes on” in deciding (as it were) whether to believe a proposition. Prime candidates for
grounds are the subject’s other beliefs and experiences. Suppose now that we restrict
“justifiers,” or “J-factors,” to grounds (in Alston’s sense). Won’t that substantially
change the terms of the debate, and won’t internalism win the debate so defined? Surely
experiences and beliefs are internal factors.22 If the debate’s outcome is so easily
changed, shouldn’t we reconsider our original definition of justifiers as unduly
prejudicial?
I respond by saying that this proposal would itself be prejudicial. Since grounds
as conceived by Alston are only the current beliefs and experiences on which a subject
might base a belief, that already excludes important externalist factors that, by the
arguments of section 6, are highly relevant to justification. Furthermore, it excludes the
causal processes, including inferential processes, that subjects might use to arrive at new
beliefs based on antecedent beliefs. This is, arguably, another important externalist
factor. Finally, this itself is a terminological ploy that might make us forget about right Jrules and the criterion of J-rule rightness, but should not be allowed to do so. They are
too important to the nature of justifiedness.
Fortunately, all is not lost. Even if we adopt the new definition of “J-factors” as
grounds, the original debate can be refigured in the following terms. (For simplicity, I
don’t enumerate all relevant categories in addition to grounds; I focus on the criterion of
J-rule rightness. Readers can supply other appropriate elements.) The proposal would be
to re-frame the debate, not as a debate over the nature of justifiers, or J-factors, but as a
debate over theories or approaches to justification. It would begin by saying that any
theory of justification within a right-rule architecture requires at least two components:
(1) an account of the character of justifiers, and
(2) a criterion, ground, or standard of J-rule rightness.
The question, then, is two-fold: Are justifiers internalist or externalist, and is the criterion
of J-rule rightness internalist or externalist? Similarly for other categories that this brief
list omits. Using this way of posing the central questions, the debate would proceed
essentially as before, and the expected answers should not differ. Thus, a mere
redefinition of the term ‘justifier’ or ‘J-factor’ makes no substantive difference to the
debate. There was nothing genuinely prejudicial in my original way of framing the issue.
Externalism is likely to win under either choice of framing.
19
References
Alston, W. P. (1986/1989). “Internalism and Externalism in Epistemology,”
Philosophical Topics 14: 179-221. Reprinted in Epistemic Justification (1989). Ithaca,
NY: Cornell University Press.
Alston, W. P. (1988/1989). “An Internalist Externalism,” Synthese 74: 265-283.
Reprinted in Epistemic Justification. Ithaca, NY: Cornell University Press.
Armstrong, D. M. (1973). Belief, Truth and Knowledge. Cambridge: Cambridge
University Press.
Bach, Kent (1985). “A Rationale for Reliabioism,” The Monist 68: …..
BonJour, Laurence (1980). “Externalist Theories of Empirical Knowledge,” Midwest
Studies in Philosophy, vol. 5, Studies in Epistemology. Minneapolis, MN: University of
Minnesota Press (pp. 53-73).
Chisholm, R. M. (1977). Theory of Knowledge, 2nd ed. Englewood-Cliffs, NJ: PrenticeHall.
Chisholm, R. M. (1989). Theory of Knowledge, 3rd ed. Englewood-Cliffs, NJ: PrenticeHall.
Christensen, David (2004). Putting Logic in Its Place. Oxford: Oxford University Press.
Comesana, Juan (2005). “We Are (Almost) All Externalists Now,” in J. Hawthorne, ed.,
Philosophical Perspectives 19, Epistemology. Malden, MA: Blackwell (pp. 59-76).
Conee, Earl and Feldman, Richard (2004a). Evidentialism: Essays in Epitemology.
Oxford: Oxford University Press.
Conee, Earl and Feldman, Richard (2001/2004a). “Internalism Defended,” in H.
Konrblith, ed., Epistemology: Internalism and Externalism. Oxford: Blackwell.
Reprinted in Conee and Feldman (2004a). Evidentialism: Essays in Epistemology.
Oxford: Oxford University Press.
Feldman, Richard and Conee, Earl (2004b). “Afterward to ‘Evidentialism’,” in Conee
and Feldman, Evidentialism: Essays in Epistemology. Oxford: Oxford University Press.
Foley, Richard (1987). A Theory of Epistemic Rationality. Cambridge, MA: Harvard
University Press.
Fumerton, Richard (1988). “The Internalism/Externalism Controversy,” Philosophical
Perspectives vol. 2, Epistemology. Atascadero, CA: Ridgeview (pp. 443-461).
20
Ginet, Carl (1975). Knowledge, Perception, and Memory. Dordrecht: D. Reidel.
Goldman, Alvin (1980). “The Internalist Conception of Justification.” Midwest Studies
in Philosophy, vol. 5, Studies in Epistemology. Minneapolis, MN: University of
Minnesota Press (pp. 27-52).
Goldman, Alvin (1986). Epistemology and Cognition. Cambridge, MA: Harvard
University Press.
Goldman, Alvin (1999a/2002). “Internalism Exposed,” Journal of Philosophy 96: 271293. Reprinted in Pathways to Knowledge: Private and Public. New York: Oxford
University Press.
Goldman, Alvin (1999b). Knowledge in a Social World. Oxford: Oxford University
Press.
Goldman, Alvin (1999c). “A Priori Warrant and Naturalistic Epistemology,” in J.
Tomberlin, ed., Philosophical Perspectives 13, Epistemology (pp. 1-28). Malden, MA:
Blackwell.
Goldman, Alvin (in press). “Epistemic Relativism and Peer Disagreement,” in R.
Feldman and F. Warfield, eds., Disagreement. New York: Oxford University Press.
Harman, Gilbert (1986). Change in View. Cambridge, MA: MIT Press.
Kornblith, Hilary (1988/2001). “How Internal Can You Get?” Synthese 74. Reprinted in
Kornblith, ed., Epistemology: Internalism and Externalism. Oxford: Blackwell.
Lyons, Jack (in press). Perception and Basic Beliefs. New York: Oxford University
Press.
Plantinga, Alvin (1993). Warrant: The Current Debate. New York: Oxford University
Press.
Siegel, Susanna (2008). “Cognitive Penetrability and Perceptual Justification,” draft,
Harvard University.
Williamson, Timothy (2000). Knowledge and Its Limits. Oxford: Oxford University
Press.
21
1
In other words, externalism is simply the denial that that an internalist constraint on justifiers is
appropriate (see Alston, 1986/1989: 185).
2
Unless otherwise indicated, all page references to articles with two publication dates refer to the second,
reprinted version.
More precisely, the definiendum of the quoted definition was called an “epistemizer.” Alston was using
“epistemizer” as a synonym of “justifier” or “J-factor.” However, since other philosophers use
“epistemizer” to denote an entirely different concept, I shall avoid this terminology entirely.
3
In saying that we don’t want to construe “relevance” as causal relevance, I am not, of course, excluding
causal factors of beliefs as candidates for being justifiers. It’s just that such causal factors are causally
relevant to the beliefs (or other doxastic states), not causally relevant to the justificational status of the
beliefs.
4
Chisholm (1977) writes: “the things we know are justified for us in the following sense: we can know
what it is, on any occasion, that constitutes our grounds, or reason, or evidence for thinking that we know”
(252). Ginet (1975) writes: “Every one of every set of facts about S’s position that minimally suffices to
make S, at a given time, justified in being confident that p must be directly recognizable to S at that time.
By ‘directly recognizable’, I mean this: if a certain fact obtains, then it is directly recognizable to S if and
only if , provided that S at that time has the concept of that sort of fact, S needs at that time only to reflect
clear-headedly on the question of whether or not that fact obtains in order to know that is does obtain” (34).
Kent Bach (1985) writes: “Internalism … treats justifiedness as a purely internal matter: if p is justified for
S, then S must be aware (or at least be immediately capable of being aware) of what makes it justified and
why” (250).
5
Conee and Feldman write that according to mentalist internalism, “internalism is the view that a person’s
beliefs are justified only by things that are internal to the person’s mental life…. A mentalist theory may
assert that justification is determined entirely by occurrent mental factors, or by dispositional ones as well.
As long as the things that are said to contribute to justification are in the person’s mind, the view qualifies
as a version of mentalism” (2001/2004a: 55). Elsewhere Feldman and Conee indicate that only the mental
states and conditions that obtain at the time of belief (or other doxastic attitude) are relevant to its
justificational status: “The epistemic justification of anyone’s doxastic attitude toward any proposition
strongly supervenes on the evidence that the person has at the time” (Feldman and Conee, 2004b: 101,
emphasis added). Note that this quotation is from the afterward to their early article “Evidentialism,”
written for their 2004 volume of collected papers, Evidentialism: Essays in Epistemology, and is therefore a
recent statement of their position.
6
7
There is also the question of whether an externalist approach to mental content renders mental states noninternal. I won’t try to tackle this problem because it’s not my problem; it’s a problem for internalists.
Nonetheless, I won’t press this problem on them.
8
Chisholm propounded principles of justifiedness (or epistemic statuses akin to justifiedness) in all three
editions of his Theory of Knowledge (1966, 1977, 1989). I have advanced a right-rule framework for
thinking about justifiedness in two previous works (Goldman, 1986; in press).
9
In a paper presented at the 2007 Rutgers Epistemology Conference, Paul Boghossian argues that a
vicious regress infects a rule-following conception of epistemic justification. Some possible reservations
about (L) emerge in a forthcoming paper of my own (Goldman, in press b). But these reservations are set
aside for present purposes.
22
10
Not long after incorporating these examples into a draft of this paper and readying it for a conference
presentation, I heard Susanna Siegel present a paper with very similar examples, many of which she
similarly considered to feature unjustified belief. See Siegel (2008).
11
Thanks to Jack Lyons for suggesting these examples. His forthcoming book (Lyons, in press) provides a
highly illuminating and subtle analysis of closely related issues.
12
For further discussion of this point, see Lyons (in press, chap. 3).
Subjective Bayesianism is surely a form of internalism, so doesn’t it internalize the R-relation? No, I
would respond. Subjective Bayesianism retains the “external” requirement that all rational or justified
changes in degrees of belief must conform to objective truths of the probability calculus, especially Bayes’
Theorem.
13
14
This was suggested by Alan Sidelle, commenting on a version of this paper delivered at University of
Wisconsin (Madison).
15
Some might prefer a revision that replaces the objective relation R in the second conjunct of the
antecedent with the condition that S justifiedly believes such a relation to obtain. I have already indicated
my discontent with this sort of fix, because it doesn’t include the objective fact. But it wouldn’t help with
the problems I am about to present (in the text) with the second revision of the type-II schema.
Burge’s special interest in preservative memory (and preservative interlocution) is to broaden the ambit
of a particular species of justification, or entitlement, namely, a priori entitlement. This aspect of his
argument is no part of what I mean to endorse here. I agree that memory can preserve the justificational
status of a belief, but I would hold that it puts its own stamp on the kind of justifiedness the agent thereby
has. In any event, I am not interested here in a sustained discussion of types of justifiedness (for related
discussion, see Goldman, 1986, pp. 299-302; Goldman 1999c, pp. 23-24).
16
17
For a clear example of this approach, see Pollock and Cruz (1999, pp. 18-19, 46-49).
If Leonardo heard this verdict on his belief, he might reply as follows: “This verdict is unfair. As far as
I can tell, I am perfectly justified in believing w.” We can concede this point to Leonardo: as far as he can
tell, he is justified in believing w. But, we should add, Leonardo cannot tell enough. He no longer
remembers the inferential operations he performed in the early stages of his argument or proof. So why
should we pay much attention to what he thinks about his justificational status? What I am saying, in short,
is that we should resist a KJ principle that says: “if one has a certain J-status with respect to a certain
attitude (at time t), then one is in a position to know (at t) that one has that J-status.” Rejecting KK
principles is a standard move among externalists, recently exemplified in depth by Williamson (2000).
Here I urge analogous abandonment of a KJ principle.
18
19
What is meant by “veritistic value” is explained in (Goldman, 1999b). See note 20 below.
20
The explanation in the text cites only the propensity of the exhibited rules to produce true categorical
beliefs. This cannot be the entire story, because many prominent epistemologies feature graded beliefs as
appropriate outputs of good J-rules. I have sought to capture the value of graded beliefs in a truthconduciveness framework by introducing the broader notion of veritistic value, in which even graded
degrees of belief assigned to truths is a good epistemic consequence (as compared with lower degrees of
belief assigned to the same truths). For a systematic development of this approach, see (Goldman, 1999b)
21
Thanks to Dennis Whitcomb for urging me to consider Foley’s approach in this context.
Of course, some internalists – Alston (1988/89) calls them “consciousness internalists” – would insist on
restricting justifiers to conscious mental states. By their lights, not every belief would qualify as a justifier.
I set this extreme brand of internalism aside.
22
23