Graduate Essay Prize Winning Paper of the 48th Annual Meeting of the Florida Philosophical Association
Elisabeth K.I. Shortsleeve, University of Florida
Introduction
In this essay, I present two thought experiments that respectively cast doubt on the adequacy of the coherentist and reliabilist approaches to epistemic justification. Next, I propose an account of justified belief that accommodates coherentist and reliabilist intuitions and avoids both difficulties. Finally, I test this account by measuring its verdicts against my intuitions concerning wishful thinking and induction and answer three objections to this account.
I should start by setting down the definitions of justified belief that I shall respectively count as coherentist (CJ) and reliabilist (RJ) in the present discussion:1
CJ: S’s belief B is C-justified just in case it is a member of a set of beliefs that combines in an optimal fashion the features of coherence, comprehensiveness and explanatoriness.
RJ: S’s belief B is R-justified just in case B is produced by some token process P such that P is of a type that is reliable and P’s being of that type is causally relevant to the production of B.
Tweedledum and Tweedledee
The following two scenarios, with respect to which CJ and RJ in turn issue intuitively inappropriate verdicts, raise some doubt concerning the adequacy of coherentism and reliabilism with respect to justification.
First, CJ is indicted by the following thought experiment: Suppose Tweedledum, a scientist, has an overall set of beliefs that combines in an optimal fashion the features of coherence, comprehensiveness and explanatoriness. This set contains a subset of beliefs T whose contents are the propositions of a well-tested empirical theory, T, and the belief, BT, that the conjunction of all the propositions in T is true. Now, suppose Tweedledum encounters some appropriately obtained, apparent empirical evidence that is such that, ceteris paribus, if Tweedledum were to believe that this apparent evidence is evidence that a given set of events occurred and BT, his overall set of beliefs would become incoherent, and if Tweedledum were to believe that this apparent evidence is not evidence that this set of events occurred and BT, his overall set of beliefs would remain coherent. Let us call ‘E’ the proposition that the given set of events occurred, ‘BE’ the belief that this set of events occurred, and ‘Bnot-E’ the belief that this set of events did not occur. Suppose also that Tweedledum believes that the apparent evidence was gathered appropriately but does not have any additional beliefs about appropriate scientific procedures that would be inconsistent with either believing or disbelieving E. Tweedledum accepts Bnot-E and does not add any other belief to his overall set. By stipulation, his overall set of beliefs remains coherent. Further, it remains adequately comprehensive because it now includes a belief about E. It remains explanatory because not-E is entailed by T, where the entailment is explanatory in the sense that the truth of the well-tested empirical theory T provides a reason for rejecting E that could be cited as an explanation of the falsity of E. According to CJ, then, Bnot-E is C-justified. Intuitively, however, one would judge that, under such conditions, neither Bnot-E nor BT is justified for Tweedledum. Rather, it seems that he should either accept E at face value and modify BT accordingly, or disbelieve E for some appropriate reason other than his belief in the truth of T. It seems, then, that ‘C-justified’ and ‘justified’ denote distinct properties.
RJ, on the other hand, is indicted by the following thought experiment: Suppose Tweedledee, another scientist, believes that the same theory T is true. Suppose that Tweedledee’s BT is R-justified. Tweedledee encounters the same appropriately obtained apparent empirical evidence for E, whose truth is inconsistent with the truth of BT. Suppose also that, if Tweedledee were to accept E, his belief BE would be produced by some token process PE of a reliable type and PE’s being of that type would be causally relevant to the production of BE in Tweedledee. Tweedledee accepts E and does not effect any further modification to his overall belief system. According to RJ, Tweedledee’s BE is R-justified. Intuitively, however, one would judge that BE is not justified for him because it is inconsistent with his belief BT. Instead, he should either accept BE and replace BT by a belief in some appropriately modified theory T*, or reject BE for some appropriate reason. It seems, then, that ‘R-justified’ and ‘justified’ denote distinct properties.
A New and Improved Account of Justification
Our discussion of these two thought experiments suggests an account of justification that is more in keeping with our intuitive responses to each case. It seems that in both cases we find that it is appropriate to reject (or modify) an empirical theory because it conflicts with apparent empirical evidence, while it is inappropriate to reject apparent empirical evidence for the mere reason that it conflicts with an accepted empirical theory. CJ is not well suited for motivating the discrepancy between our intuitive assessment of the two options available to Tweedledum in the first thought experiment. Although the coherentist demands for comprehensiveness and explanatoriness seem tailored for this very sort of case, they ultimately fail to yield the appropriate verdict. Indeed, it is conceivable that if Tweedledum were unable to explain E away, he would—and should—replace T by T*, a theory of far lesser explanatory scope that is consistent with E, thus reducing the overall comprehensiveness and explanatoriness of his beliefs. In such a case, however, we should have to conclude that believing E and T* is not C-justified for Tweedledum. RJ, on the other hand, offers not only a criterion that enables us to reach an intuitively appropriate verdict, but also explains this verdict: Tweedledum’s belief Bnot-E is not R-justified because taking our acceptance of empirical evidence as a basis on which to ground our epistemic economy regarding empirical matters is presumably generally reliable, while taking our acceptance of empirical theory as a basis on which to ground our epistemic economy regarding empirical matters is presumably generally unreliable. This, of course, is not to say that our theoretical beliefs should not be involved in our interpretation of empirical evidence. The claim is merely that our theoretical beliefs alone do not provide a sufficient justifying reason to deny apparent empirical evidence that conflict with them because doing so is generally epistemically unreliable.
However, in both cases we also find that merely accepting an empirical belief that is produced by a token of a causally relevant and reliable cognitive process-type does not guarantee that this belief is justified. In addition, such a belief should cohere with other beliefs that are accepted by the believer (either all of her beliefs, or some appropriate portion of them). While CJ captures this intuition, RJ does not. Indeed, in an attempt to accommodate this requirement, reliabilists would have to show that process-types that maximize coherence are generally reliable. In particular, they might argue that in the case of independently well-supported empirical theories facing apparent recalcitrant empirical evidence, process-types that result either in revising one’s theory in the light of the evidence, or in explaining the evidence in a manner that is consistent with one’s theoretical beliefs, are generally more reliable than process-types that result in holding inconsistent beliefs about the matter at hand. That this is so, however, is far from obvious. Indeed, it seems that process-types requiring that the coherence among one’s beliefs be maximized are reliable provided that one adjusts one’s beliefs so that they cohere with one’s true beliefs, and unreliable otherwise. Such a reliabilist account of a positive coherence requirement is clearly much stronger than our intuitive notion, which merely asks that a justified belief cohere with some relevant subclass of one’s beliefs, rather than some relevant class of one’s true beliefs. Accordingly, our revised account of justification will consider coherence as separate from the reliability requirement.
In an attempt to reconcile CJ and RJ with the intuitions discussed above, I propose the following hybrid definition of justification:
RCJ: S’s belief B is RC-justified just in case
B is produced by some token process P such that P is of a type that is reliable and P’s being of that type is causally relevant to the production of B, and
B is a member of a set of beliefs that combines in an optimal fashion the features of coherence, comprehensiveness and explanatoriness.
RCJ’s verdicts line up with our intuitions concerning the two cases discussed above: We judge that Tweedledum’s belief Bnot-E is not justified because it seems to have been produced by a token of the causally relevant and unreliable process-type “if some empirical evidence that P disagrees with the predictions of some believed well-tested empirical theory, simply deny that P.” If Tweedledum’s belief Bnot-E was indeed so produced, then it is not RC-justified. On the other hand, we judge that Tweedledee’s belief BE is not justified because, although it seems to have been produced by a token of the causally relevant and reliable process-type “if empirical evidence that P is appropriately obtained, believe that P,” his continued belief in BT results in the incoherence of his overall set of beliefs, and thus in both BE and BT being unjustified for him. According to RCJ, these beliefs also fail to be RC-justified for him. The courses of action that were intuitively suggested to both Tweedledum and Tweedledee were the following: Each should either believe Bnot-E for some appropriate reason other than BT, or replace BT with BT*, where T* is some appropriate modification of T that is consistent with BE. According to RCJ, both courses of action are also RC-justified, provided of course that the beliefs in question are produced by tokens of some causally relevant and reliable process type. According to RCJ, then, the recommendation is that believers acquire their beliefs via tokens of causally relevant and reliable process-types that ultimately maximize the coherence, comprehensiveness and explanatoriness of the believer’s overall set of beliefs.
RCJ Tested: Wishful Thinking
In “What is Justified Belief2,” A. I. Goldman discusses various cases in which wishful thinking is a reliable cognitive process and yet results in beliefs that one would intuitively judge unjustified. The last case presents a scenario in which a benevolent lazy demon, who has so far been inactive, becomes involved in our world and starts arranging things so that all of our wishes come true. Although in this case, wishful thinking has become a perfectly reliable cognitive process in our own world, why are we tempted nonetheless to call it unjustified? In order to answer this question, we need to understand wherein wishful thinking differs from appropriate thinking.
Let us first consider wishful thinking in the context of a world that is much like the one we normally take ourselves to inhabit—in particular, benevolent demon free—in order to explain what motivates our intuitive reaction to Goldman’s case. The epistemic blameworthiness of wishful thinking obviously does not arise merely from letting one’s wishes guide one’s thinking. Indeed, if such were the case we should find ourselves compelled to criticize the thinker who lets her wish to discover truth guide her thinking. Rather, wishful thinking consists in letting one’s thinking be guided by her wishes in a manner that either (i) leads her to believe that P when she does not possess sufficient evidence that P is true or (ii) leads her to believe that P when she possesses evidence that P is false. Goldman correctly stresses that under normal circumstances, wishful thinking is an inappropriate cognitive process because it is unreliable. However, I believe that in addition to being unreliable, wishful thinking is epistemically blameworthy because it tends to promote incoherence in our overall belief system.
That this is so is obvious in cases of type (ii) above. Indeed, in such cases the believer presumably has beliefs concerning the evidence that P is false which are incoherent with her belief that P. Should a believer engage in wishful thinking of type (ii) and modify her beliefs in a manner that maximizes the coherence, comprehensiveness and explanatoriness of her overall belief system and that is consistent with her believing P, however, it is no longer so clear that we should intuitively deem her belief that P unjustified. In particular, let us imagine that a believer does so in Goldman’s benevolent demon-infested world in which wishful thinking is a reliable cognitive process. It is certainly no longer intuitively obvious that her belief that P is epistemically unjustified in such circumstances. If it is the case that adding a clause which safeguards coherence changes our intuitive response to the case at hand, then it seems that coherence is indeed a concern that motivates our original judgment. It is less obvious that wishful thinking of type (i) above always promotes incoherence in one’s overall belief system. A threat to coherence does arise, however, in that evidence to the contrary of the culpable belief might come to be believed in the future, at which time the believer’s overall belief system will be incoherent unless she revises her beliefs so as to maintain coherence. Generally, since there is no rational prohibition against incoherent wishing, letting one’s wishing guide one’s thinking is liable to threaten the overall coherence of one’s belief system unless a special effort is made to maintain it. Let us imagine, once again, that a believer engages in wishful thinking of type (i) in a world where wishful thinking is a reliable cognitive process. If we add that all beliefs she acquires via such wishful thinking maximize the coherence, comprehensiveness and explanatoriness of her overall belief system, it is no longer so clear that we should intuitively deem that the beliefs in question are not justified.
This, in my opinion, is the reason why we intuitively judge that wishful thinking is inappropriate even in a case when it is a perfectly reliable cognitive process. Even if a benevolent demon were to become involved in our world, wishful thinking would tend to promote incoherence in our overall set of beliefs, once again most obviously so in cases of type (ii) above. Further, as noted above, the rational permissibility of incoherent wishing jeopardizes the overall coherence of our beliefs even in cases of type (i) above. For these reasons, wishful thinking is an inappropriate cognitive process even when it is reliable.
To conclude our reflections on wishful thinking, let us consider the following case: One can imagine that the very meddling of the benevolent demon might cause our world to be such that a believer who had a sufficiently large number of true beliefs about our world would have an incoherent overall belief system, so that coherence among one’s beliefs could be maintained only through deliberate disbelief of discrepant empirical evidence, which is a species of wishful thinking. Some of the beliefs thus obtained could meet both requirements for RC-justification. Does this verdict accord with our intuitions? The case is a little confusing, partly because it seems inappropriate to say that the propositional attitudes at issue are “beliefs” in the usual sense of the term. It seems to me that an essential ingredient of believing that P consists in assigning to P the truth-value that one perceives P as having. In other words, it seems to me that a necessary condition for something to count as a belief is that it purports to represent something that is perceived as independent from our representing it. In such a benevolent demon-infested world, the propositional attitude resulting from deliberately denying discrepant empirical evidence seems to create rather than represent what is perceived as true, and for this reason seems not to count as a belief in the first place, let alone a justified belief. If so, the attitude in question is neither justified, nor RC-justified.
RCJ Tested Again: Induction
I now propose to test RCJ against my intuitions concerning induction. The well-known problem with induction is that for any appropriate inductive argument such as:
GREEN:
P(1) All emeralds so far observed are green C(1) All emeralds are green,one can construct an inappropriate counterpart such as: GRUE:
P(2) All emeralds so far observed are grue
C(2) All emeralds are grue,
where ‘grue’ means green and observed on or before 12/31/2003 or blue and observed after 12/31/2003. The availability of a GRUE-like counterpart to any appropriate inductive argument is thought to show that induction is not a reliable cognitive process.
Let us start by consulting our intuitions regarding this matter. If there were no GRUE-like counterparts, when would belief in the conclusion of an inductive argument be justified? Would it be justified merely in virtue of being about the conclusion of an inductive argument of the form of GREEN? Consider the following inductive argument:
P(3) All rational creatures so far observed are humanoid
C(3) All rational creatures are humanoid
Clearly, belief in C(3) is not justified. I think that the reasons that belief in C(3) is not justified are (i) P(3)’s report concerns a sample that is too homogeneous to be an appropriate basis for induction and, more importantly in this context, (ii) there is no independent reason to accept C(3). Suppose now that the following were available to us: first, a powerful scanning device capable of detecting rational life and humanoid life within a range of a few thousand light-years whose result are reported in P(3); and, second, a justified (i.e., RC-justified) scientific theory of rationality of broad explanatory power that indicates that the material ingredients required for rationality are sufficient for humanoidness. If such were the case, I believe that one would intuitively judge that belief in C(3) is justified because it is a generalization based on an appropriately varied sample and independently supported by theoretical beliefs. Reflection concerning the reliability of induction seems to confirm these preliminary results. Indeed, although mere induction is clearly an unreliable cognitive procedure, induction that is based on an appropriately varied and appropriately sized sample and that is independently supported by a justified theory of appropriate explanatory power is likely to be quite reliable—provided, of course, that we spell out what we mean by appropriate. Let us call the cognitive procedure just outlined “reliable induction,” to be contrasted with “mere induction.”
Do GRUE-like counterparts throw doubt on the justifiability of beliefs acquired via reliable induction? It is clear that mere GRUE-like counterparts will not do. What we should need is a beefed up version whose conclusion is independently supported by some theory of appropriate explanatory power. The kind of theory required would need to be such that belief in it can be RC- justified, that is, belief in it can be a member of a set of beliefs that maximizes coherence, comprehensiveness and explanatoriness and can be produced by some token of a reliable and causally relevant type of cognitive process. Now, it seems to me that GRUE-like terms are not the kinds of word that can be used in explanatorily powerful theories for the same reason that makes them unsuited for induction. Nonetheless, suppose that a theory about which an RC-justified belief can be formed and which contained the term ‘grue’ is available and is suitable for predicting the color of emeralds (let us call this the grue theory). Then, it is highly unlikely that such a theory would offer independent support to C(2) above (All emeralds are grue). If it did offer independent support to C(2), the grue theory would be inconsistent with the beliefs “All emeralds are green” and “No emeralds are blue.” This, of course, can be taken to incriminate the beliefs that all emeralds are green and that no emeralds are blue just as much as it incriminates the belief that the grue theory is true. If the belief system of the individual in question includes the former two beliefs, it is most likely that accepting them rather than the grue theory would be optimal for her in terms of the coherence, comprehensiveness and explanatoriness of her overall belief system. If so, the belief that the grue theory is true is not RC-justified for her after all. If her belief system does not include these two beliefs, then perhaps the belief that all emeralds are grue is RC-justifed for her. In either case, then, it does not seem that we have found a GRUE-like counterpart that incriminates reliable induction. It is likely, then, that finding GRUE-like counterparts to reliable inductive arguments is far from easy. At this point, it seems appropriate to lay the burden of providing such a counterpart on the shoulders of the critic of reliable induction who claims that such counterparts can be constructed.
If our account is correct, then, it seems that our definition of RC-justification enables us to discriminate between appropriate and inappropriate induction in a manner that agrees with our intuitions and avoids GRUE-like traps. Since RC-justification has been shown to agree well with our intuitive verdicts, I propose to adopt the following as a definition of justification tout court:
S’s belief B is justified just in case:
B is produced by some token process P such that P is of a type that is reliable and P’s being of that type is causally relevant to the production of B, and
B is a member of a set of beliefs that combines in an optimal fashion the features of coherence, comprehensiveness and explanatoriness.
RCJ Defended: Three Objections
Appeal to reliability can do it all
I must now defend my proposal against obvious objections. First, someone might protest that one of Goldman’s reformulated reliabilist definitions of justified belief can accomplish as much as our RCJ above, namely,
(10) If S’s belief in p at t results from a reliable cognitive process, and there is no reliable or conditionally reliable process available to S which, had it been used by S in addition to the process actually used by S, would have resulted in S’s not believing p at t, then S’s belief in p at t is justified,3
(11) where a process is conditionally reliable when a sufficient proportion of its output-beliefs are true given that its input-beliefs are true.4
It seems that Goldman’s (10) is likely to yield verdicts that agree with RCJ in many cases. Indeed, the process through which one would satisfy RCJ’s second condition seems to fall within the type of conditionally reliable processes covered by the second clause in Golman’s (10). However, I think that the verdicts of Goldman’s (10) and RCJ would diverge in some cases. Indeed, in the process of making one’s overall beliefs meet RCJ’s second condition, the relevant set of input beliefs is her current overall set of beliefs other than p (let’s call this set Z). Her set of output beliefs satisfies RCJ’s second condition if that set is:
(i) p, if p given Z meets RCJ’s second condition, or
if p given Z does not meet RCJ’s second condition, either
(ii) not-p, or
(iii) p and Z*, where Z* is a set of beliefs such that, if Z* were substituted for Z, p would meet RCJ’s second condition.
Goldman’s second clause is inappropriately more stringent than RCJ because it rules out (iii) above. Indeed, according to Goldman’s second clause, the reliability of the process in question is conditional on the truth of Z. Accordingly, if Z ∪ {p} is incoherent, one can only come to believe p if she fails to employ any conditionally reliable process that would lead her to reject p.
It is possible that some reformulation of Goldman’s (10) could yield verdicts that agree with RCJ. For example, one that changes the second clause of (10) to something like
. . . and there is no reliable or conditionally reliable process available to S which is more reliable than all other reliable or conditionally reliable processes available to S and which, had it been used by S in addition to the process actually used by S, would have resulted in S’s not believing p at t.
It seems that, thus reformulated, Goldman’s (10) will yield verdicts that are in agreement both with RCJ and our intuitions. If so, RCJ would nonetheless retain one advantage over the reformulated Goldman’s (10), namely, the advantage of greater simplicity.
Unholy internalist/externalist marriage
Second, it might be objected that even if RCJ does present the advantage of greater simplicity over Goldman’s reformulated (10), the former is unacceptable because it strives to forge an unholy marriage between internalist and externalist requirements for justification. In defense of RCJ, I shall maintain that the marriage is not unholy and that it is routinely consummated in our intuitive judgments concerning justification. It seems to me that even if one were able to show that, in most cases, beliefs that meet the two conditions for RC-justification were produced by a token of a causally relevant reliable process-type, the coherentist clause in our definition of RC-justification is not reducible to reliabilist concerns. As our earlier discussion of the benevolent demon-infested world shows, a reliably produced belief that fails to meet the coherentist condition for RC- justification is intuitively judged unjustified. I think that our intuitive judgments about the justificatory status of beliefs are motivated precisely by the two very different concerns that are expressed in the two requirements for RCJ-justification. That is, they are motivated by an internalist concern with whether the belief in question bears the appropriate relationship to other beliefs that are accepted by the believer and by an externalist concern with whether the belief was acquired in an appropriate way, i.e., in a way that is truth-conducive under normal circumstances. It is true that, given that a belief that is acquired in a way that maximizes appropriate relations among one’s beliefs can be expected to be truth-conducive under normal circumstances, the coherentist condition can often be satisfied by merely satisfying the reliabilist condition. However, it seems to me that the coherentist concern is not reducible to the reliabilist concern because the former is interested primarily in assessing whether the cognitive methods used by the believer in acquiring the belief in question tend to generate beliefs that are appropriately related to the world while the latter is interested primarily in assessing how acquiring this belief affects the overall cognitive state of the believer. I hope that the foregoing discussion was successful in showing that both concerns routinely motivate our intuitive evaluations of the justificatory status of beliefs.
RCJ is too stringent
The third and last objection I wish to discuss is that, since most of us have at least some incoherent beliefs, most of our beliefs do not meet the first condition for RC-justification. Accordingly, RCJ is too stringent. I agree that the coherentist condition for RC-justification requires a more careful formulation in order to circumvent this last objection. An improved formulation of RCJ should meet the following two objectives. First, it should prevent the incoherence of a single belief from infecting all other beliefs. This could be achieved by circumscribing the requirement for coherence to a subset of beliefs that are more closely related to the belief whose justificatory status is at issue. Second, however, it should require some coherence among various subsets of beliefs. I am not certain how one can best meet these two seemingly conflicting desiderata. It might be that in most cases, the requirement that the belief in question maximize not only the coherence, but also the comprehensiveness and explanatoriness of the believer’s belief system works well enough to bypass this last objection. I do not doubt, however, that pointed thought experiments can be constructed in which RCJ fails to satisfy one or the other of these desiderata. If it can be reformulated in a manner that circumvents this last objection, I believe that RCJ would offer a very satisfactory account of justification.
Work Cited
Goldman, A. I. “What is Justified Belief?” Epistemology. Ed. Ernest Sosa and Jaegwon Kim. Malden, Mass.: Blackwell, 2000.