views updated


Reliabilism is an approach to the analysis of either knowledge or justified belief that makes, in some way or another, the reliability of belief-producing faculties or processes the key notion of epistemic assessment. An early version of a reliabilist theory of knowledge was proposed by David M. Armstrong (1973), who thought of knowledge in terms of a reliable thermometer that accurately indicates the correct temperature. A (noninferential) true belief amounts to knowledge, according to Armstrong, if its properties nomically (i.e., via the laws of nature) guarantee its truth. Closely related theories conceive of knowledge as resulting from a counterfactual guarantee of truth. For instance, according to Robert Nozick (1981), knowledge comes about when a subject's belief that p tracks the truth of p, which it does (focusing just on the core of Nozick's theory) if the following condition is met: S would not believe that p if p were false. Alternatively, Fred Dretske (1971, 1981) suggests that a true belief counts as knowledge if the subject possesses a conclusive reason for p. According to this proposal, S knows that p on the basis of a reason, R, if R would not be true unless p were true.

Reliabilist Theories of Justified Belief

Turning to reliabilist theories of justified belief, there are two main versions to consider: reliable indicator theories, and process reliabilism. A version of the former was developed by Marshall Swain (1979): What makes S's belief that p justified is the belief's being a reliable indication (conceived of in terms of objective probability) of p's truth. In contrast, process reliabilism (reliabilismpr), as advocated by Alvin Goldman (1979), focuses not on the belief itself, but its causal history, or the cognitive process by which it was formed. The basic proposal is that a belief that p is justified if and only if the process by which it was formed is (sufficiently) reliable. On the one hand, perception, memory, and introspection are reliable cognitive processes, typically resulting in justified beliefs and indeed knowledge. Hasty generalization, wild hunches, and wishful thinking, on the other hand, are unreliable processes and invariably produce unjustified beliefs and prevent the formation of knowledge. Moving beyond this initial characterization, an account of reliabilismpr requires refinement as to the question of precisely what determines the reliability of a cognitive process. According to one approach, a process's reliability is fixed by the truth ratio of its actual doxastic output: The greater the ratio of true over false beliefs within the set of beliefs that make up a process's actual track record, the higher its degree of reliability. As an alternative to such a track record conception, William Alston (1995) recommends a propensity construal, according to which a process's reliability is determined by, not its actual track record, but what the truth ratio of its output would be in an appropriate range of cases.

According to reliabilismpr, as well as other versions of reliabilism, whether S's belief that p is justified, or an instance of knowledge, does not dependat least not exclusivelyon S's evidence relative to p. Reliabilist theories of knowledge and justification must thus be viewed as intended alternatives to evidentialist theories (Chisholm 1989, Conee and Feldman 1985). Nevertheless, in Goldman's defense of reliabilismpr, evidentialist considerations do not completely drop out of the picture. Suppose the following: (i) S forms the belief that p via process C; (ii) C is in fact reliable; and (iii) on the basis of professional testimony, S has reason to believe that C is unreliable. Intuitively, S's belief is not justified. Cases such as that make it doubtful that origination in a reliable process is by itself sufficient for justification. Accordingly, Goldman (1986) supplements his account with a nonundermining clause to the effect that S must not believe, or be in possession of evidence supporting the belief, that the relevant process is unreliable.

In its most radical and challenging manifestation, reliabilismpr asserts that reliable belief formation is both necessary and sufficient for a belief's justification. Alston (1989) defends a moderate version, according to which reliable belief formation is merely necessary for justification. An approach that may be viewed as an alternative to reliabilismprvirtue epistemologyshifts the focus away from reliable processes to reliable faculties or cognitive virtues, giving rise to the thought that justification and knowledge may be conceived of as resulting from the employment of virtuous faculties (Sosa 1991). A related approach is advocated by Alvin Plantinga (1993). Plantinga's view is that knowledge is generated by properly functioning faculties, where a faculty's proper functioning requires reliability, in addition to adequate design and an orientation towards truth and the avoidance of falsehood.

The Three Major Problems with Reliabilism

Reliabilismpr is confronted with three major problems. The first of these raises the issue of whether production via reliable processes is necessary for justification (Cohen 1984, Ginet 1985). Consider the beliefs of a subject who is deceived by an evil demon. Because the evidential situation of an evil demon victim is not relevantly different from that of a normal person who has (presumably) by and large justified beliefs, it is commonly agreed that such a victim's beliefs are, just like a normal person's, by and large justified. Alas, the victim's beliefs are, unlike those of a normal person, massively false. The challenge for reliabilismpr, then, is this: The beliefs of an evil demon victim are justified although they are the result of unreliable cognitive processes.

There are four ways in which reliabilists can respond. First, they canimplausiblydeny that evil demon victims have justified beliefs. Second, they can deny the relevance of the counterexample, as Alston (1995) does. He argues that the cognitive processes to consider are to be restricted to those that would yield a high truth ratio over a wide range of situations of the kind one typically encounters. Third, advocates of reliabilismpr can try to accommodate the counterexample by modifying reliabilismpr. For example, Goldman (1986) introduces normal worlds reliabilism, the basic idea of which is that a belief (in any possible world) is justified if and only if the process by which it was formed is reliable in normal worlds: worlds that correspond to our general beliefs about the actual world. Because perception is reliable in normal worlds, normal worlds reliabilism arguably yields the result that the perceptual beliefs of evil demon victims are justified.

Whether this response succeeds depends on whether the processes by which an evil demon victim forms beliefs can properly be characterized as perceptual, memorial, and so on. After all, at the end of the causal chains from which a deceived subject's beliefs originate, there is the evil demon: not exactly the kind of creature one finds in any normal worlds. So if the belief-generating processes are considered in their entirety, it is hard to tell what their truth ratios in normal worlds would be because such processes are not part of any normal worlds to begin with. Fourth, defenders of reliabilismpr can introduce different concepts of justification. For example, Goldman (1988) distinguishes between strong and weak justification. According to this proposal, the beliefs of evil demon victims are justified in the weak sense, whereas reliabilism is intended to be an analysis of strong justification, the kind of justification that is needed if a true belief is to count as an instance of knowledge.

the second problem

The second problem raises the issue of whether origination in a reliable process is sufficient for justification (BonJour 1985, ch. 4). Suppose the following: (i) Norman's belief that p is the result of clairvoyance; (ii) Norman's faculty of clairvoyance is reliable; and (iii) Norman has no reason to believe that his belief that p originated in, or is sustained by, a reliable faculty. Reliabilismpr implies that Norman is justified in believing that p, whereas intuitively he seems to be unjustified. In response, reliabilists can deny the intuition underlying the objectionthat is, insist that Norman's belief is justified. Alternatively, they can, once again, attempt to accommodate the example by devising a suitable modification. For example, they might consider ignoring the absence of evidential support an unreliable cognitive process, and suggest that beliefs whose causal origin includes that process are unjustified. The problem with this suggestion is that it threatens to rob reliabilismpr of its identity by letting it collapse into a disguised version of evidentialism.

the third problem

The third problem, known as the "generality problem," raises the issue of how to individuate the cognitive processes the reliability of which is supposed to determine whether a belief is, or fails to be, justified (Feldman 1985). Suppose a person sees, and thus believes, that the cat is lying on the couch. The process by which this belief is formed could plausibly be classified as perception. More specifically, it could be viewed as an instance of visual perception. Further specification yields further choices: visual perception at a distance of (say) eight feet; visual perception of a medium sized object at a distance of eight feet; visual perception of a medium sized object at a distance of eight feet under daylight illumination. In general terms, the point is that a particular token of a cognitive process instantiates many different process types. Some of them are reliable, some of them are not. Whether a belief whose justificational status is at issue comes out justified or unjustified will depend on which process type is made the basis of the assessment. The challenge advocates of reliabilismpr face is to give a principled account of how to select the right process type.

Consider perception, an obviously reliable process. But not all perceptual beliefs are justified. Nor are, for that matter, all memorial beliefs, or all visual beliefs, or all auditory beliefs. Sometimes, perceivers fail to take into account undermining evidence, and then beliefs produced by reliable processes fail to be justified. Hence individuating process types using broad categories such as perception, vision, or memory will often yield the wrong results. More specification is clearly required. But too much specification also yields the wrong results. At the extreme end of specification are process types instantiated by one and only one process token; one then encounters what Richard Feldman (1985) calls the "single case problem." If such a token results in a false belief, the result will be total unreliability, for the process type's output is false in all cases. If the process token in question results in a true belief, the result will be perfect reliability, for the process type's output is true in all cases. In the former case, the belief will be unjustified no matter what; in the latter case, it will be justified no matter what. This will result in clearly counterintuitive results. Suppose S is a paranoid schizophrenic. While riding on the bus, S's paranoia leads him to believe that the bus will blow up. Suppose further that that is in fact true. Let P* stand for a process type described in such a way that P* has one and only one instantiation: the process token that caused S's belief about the bus. (Such a description can easily be achieved by making reference to properties that uniquely pick out S and the circumstances under which S formed that belief.) Because P* is a (perfectly) reliable process, reliabilismpr impliesimplausiblythat S is justified in believing that the bus will blow up. The problem for advocates of reliabilismpr is this: On the basis of what principled grounds can they claim that P* is not the process type the reliability of which determines the justificational status of S's belief?

Alston (1995) claims that there is a solution to the generality problem. Regarding the single case problem, he suggests that it does not arise when the track-record conception of reliability is replaced with a propensity conception. Consider again the belief of the paranoid subject that the bus will blow up. On the propensity conception of generality, the process type in question, having precisely one instantiation that led in fact to a true belief, nevertheless counts as unreliable when it is taken into account that beliefs resulting from paranoia tend to be false. Moreover, Alston argues that there are objective, psychological facts of the matter that determine, for each process token leading to a particular belief, which process type this token instantiates. According to Alston, every process token instantiates an input-output function. Each time a belief is formed, Alston claims, there is one and only one input-output pair that is psychological real. However, Earl Conee and Richard Feldman (1998) respond that, even if one accepts the constraints Alston places on the selection of legitimate process types, there will still be a wide range of process types going from narrow to broad characterizations. As a result, there will be cases of belief formation for which reliabilismpr will not yield a determinable implication about the belief's justificational status.

Internalist and Externalist Theories

It is common practice to distinguish between internalist and externalist theories of knowledge and justification. According to internalism about justification, the factors that determine a belief's justificational status (call them "J-factors") must be internal to the subject's mind. Typically, such internality is defined epistemically: An item x is internal to S if and only if S can, merely by reflecting on it, determine whether x is present or absent. Internalist theories of justification, then, usually demand that J-factors must be such that their presence or absence is always on reflection recognizable by the subject. As a result of this constraint, justification itself turns into something the presence or absence of which can be recognized upon reflection. According to externalism about justification, J-factors are not subject to any internality constraint. Reliabilism about justification, in its various manifestations, is an externalist theory, for the obvious reason that subjects are not always in a position to determine, solely on the basis of reflection, whether their beliefs are the result of reliable cognitive processes. Consider, again, the victim of an evil demon. Upon reflection, such a victim will think that her perceptual beliefs originate in reliable cognitive processes, when in fact they do not. However, the classification of reliabilism as an externalist theory should not be misunderstood to mean that, according to reliabilism, the reliability of our cognitive processes is completely beyond our ken. To the contrary, there is no reason why reliabilists should deny that, in typical situations when a subject forms, for example, perceptual or memorial beliefs, it should be knowable to the subject, on the basis of presently available evidence, that the beliefs in question have their origin in reliable processes or faculties.

The internalism/externalism issue presents itself in a different form when the object of the dispute is not justification but knowledge. There is broad agreement that knowledge is not, and indeed cannot be, internal in the way in which, according to some, justification is internal. Suppose the following: (i) S has a body of excellent evidence, E, in support of p; (ii) E is misleading: p is in fact false. Reflecting on whether she knows that p, S will of course conclude, mistakenly, that she does. Clearly, then, whether or not one knows cannot always be determined upon reflection. Thus it is beyond dispute that knowledge is external. Nevertheless, it would not be inaccurate to say that evidentialists defend an internalist conception of knowledge. According to evidentialists, S knows that p only if S has a good reason for p. But whether or not one has a good reason for p is something that is internal to the subject; it is something that can be determined merely by reflecting on one's evidence. Evidentialists, therefore, hold that one of the necessary conditions of knowledge is internal, and thus may be considered internalists about knowledge. Reliabilists, however, are externalists about knowledge, for they typically claim that reliable belief production, suitably qualified, is sufficient for making a true belief an instance of knowledge, thus advocating an account of knowledge without any internalist condition.

The reliabilist, externalist view that the employment of reliable processes or faculties is sufficient for making a true belief thus produced an instance of knowledge that can be supported by citing that very young children and animals possess knowledge, for neither the former nor the latter would seem to be capable of having good reasons in support of their beliefs. The evidentialist, internalist view that one knows only if one possesses a good reason can be defended by pointing out that, upon discovering that a person believes that p without having a good reason for p, we tend to judge that that person does not know that p. Reliabilists, on the one hand, need to come to terms with one kind of fact about our ordinary cognitive practice: people are reluctant to attribute knowledge in the absence of good reasons. Evidentialists, on the other hand, need to come to terms with another kind of fact about our ordinary cognitive practice: people do not hesitate to attribute knowledge to very young children and even such animals as cats and dogs.


Even though reliabilist theories are properly classified as externalist, there is no reason in principle why internalists should not acknowledge the relevance, or even fundamental importance of reliability. To begin with, internalists might agree that a true belief counts as knowledge only if it originates in a reliable faculty. Furthermore, Matthias Steup (2004) proposes internalist reliabilism as an answer to the question of why sense experience is a source of justification. According to Steup's proposal, what makes sense experience a source of justification is not de facto reliability, but evidence of reliability. Perception is a source of justification for a subject, S, if and only if S has, on the basis of track record memories, reason to believe that her perceptual faculties are reliable. According to this proposal, perception is a source of justification even in worlds in which it is in fact unreliable, as long as, on the basis of adequate evidence, it appears to be reliable. According to such an approach, issues of reliability lie indeed at the heart of epistemology, for S acquires knowledge only if S (i) employs a faculty that is in fact reliable, and (ii) possesses evidence of that faculty's reliability.

See also Alston, William P.; Armstrong, David M.; Dretske, Fred; Epistemology; Epistemology, History of; Evidentialism; Goldman, Alvin; Nozick, Robert; Plantinga, Alvin; Virtue Epistemology.


Alston, William. "How to Think about Reliability." Philosophical Topics 23 (1995): 129.

Alston, William. "An Internalist Externalism." In Epistemic Justification: Essays in the Theory of Knowledge. Ithaca, NY: Cornell University Press, 1989.

Armstrong, D. M. Belief, Truth, and Knowledge. Cambridge, U.K.: Cambridge University Press, 1973.

BonJour, Laurence. The Structure of Empirical Knowledge. Cambridge, MA: Harvard University Press, 1985.

Chisholm, Roderick. The Theory of Knowledge. Englewood Cliffs, NJ: Prentice Hall, 1989.

Cohen, Stewart. "Justification and Truth." Philosophical Studies 46 (1984): 27995.

Conee, Earl, and Richard Feldman. "Evidentialism." Philosophical Studies 48 (1985): 1544.

Conee, Earl, and Richard Feldman. "The Generality Problem for Reliabilism." Philosophical Studies 89 (1998): 129.

Dretske, Fred. "Conclusive Reasons." Australasian Journal of Philosophy 49 (1971): 122.

Dretske, Fred. Knowledge and the Flow of Information. Cambridge, MA: MIT Press, 1981.

Feldman, Richard. "Reliability and Justification." Monist 68 (1985): 15974.

Ginet, Carl. "Contra Reliabilism." Monist 68 (1985): 17587.

Goldman, Alvin. "Epistemic Folkways and Scientific Epistemology." In Liaisons: Philosophy Meets the Cognitive and Social Sciences. Cambridge, MA: MIT Press, 1991.

Goldman, Alvin. Epistemology and Cognition. Cambridge, MA: Harvard University Press, 1986.

Goldman, Alvin. "Strong and Weak Justification." Philosophical Perspectives 2 (1988): 5169.

Goldman, Alvin. "What Is Justified Belief?" In Justification and Knowledge, edited by George Pappas. Dordrecht, Netherlands: Reidel, 1979.

Nozick, Robert. Philosophical Explanations. Cambridge, MA: Harvard University Press, 1981.

Plantinga, Alvin. Warrant: The Current Debate. Oxford: Oxford University Press, 1993.

Sosa, Ernest. Knowledge in Perspective. Cambridge, U.K.: Cambridge University Press, 1991.

Steup, Matthias. "Internalist Reliabilism." Philosophical Issues 14 (2004).

Swain, Marshall. Reasons and Knowledge. Ithaca, NY: Cornell University Press, 1979.

Matthias Steup (2005)