The generality problem for reliabilism

Philosophical Studies

Volumn 89, Issue 1, 1998, Pages 1-29

  Conee, E ^a   Feldman, R ^a  

^a UNIVERSITY OF ROCHESTER   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 54649084656     PISSN: 00318116     EISSN: 15730883     Source Type: Journal    
DOI: 10.1023/A:1004243308503     Document Type: Article

References (46)

1. Some authors discuss process reliability accounts of knowledge rather than accounts of epistemic justification. No point will be made below that turns on the differences between knowledge and justification.
2. Alvin Goldman in "What is Justified Belief?" in G.S. Pappas (ed.) Justification and Knowledge (Dordrecht, Holland, 1979)
3. and Epistemology and Cognition (Cambridge, MA, Harvard University Press, 1986) defends process reliabilist accounts of epistemic justification. In those works he recognizes the existence of the generality problem.
4. See especially "What is Justified Belief?", p. 11
5. and Epistemology and Cognition, pp. 49-51.
6. The problem is emphasized in Richard Feldman's "Reliability and Justification", The Monist 68 (1985): 159-174.
7. It is also discussed by John Pollock in "Reliability and Justified Belief", Canadian Journal of Philosophy 14 (1984): 103-114. For responses to the problem, see the works of William Alston, Ralph Baergen, Mark Heller, Frederick Schmitt, Ernest Sosa, and Charles Wallis cited and discussed below.
8. It is possible to construct a version of process reliabilism which is only about process tokens and does not confront the generality problem. It faces a considerable problem in making sense of the claim that a token sequence of events has some tendency toward producing beliefs whose truth-ratio would constitute its "reliability". Furthermore, the problems that affect (NS3) below, in virtue of types having just one belief content in their outputs, also affect reliability theories that locate a sort of reliability in process tokens.
9. There may not always be a fact of the matter. In the examples used here the belief is either definitely justified or definitely unjustified. The reliability of relevant types for process tokens that lead to beliefs whose epistemic status is unclear will be of less value to present concerns, since such cases are less useful in assessing epistemological theories.
10. "How to Think About Reliability", Philosophical Topics (Spring 1995): 1-29.
11. The proposal mentioned here appears on p. 10. If a satisfactory solution to the generality problem existed, it would be worth addressing difficulties with details of this proposal. For one thing, it is not clear who "we" are supposed to be: all of humanity, or all sentient life on earth, or sentient life everywhere in the universe, or etc. And for another thing, it is unclear which belief-forming situations are "typical". Presumably, bizarre psychology lab situations are atypical. But is perception during space travel atypical, no matter how common it becomes? Are situations of fatigue, intoxication, and excitement atypical? Another difficulty is that we may be specially perceptive during rare emergency conditions. If these are atypical situations, then the justified beliefs from these perceptions might turn out not to be of generally reliable types. In any event, if the present work is correct in its main thesis, then these difficulties are not worth pursuing because the generality problem is insoluble.
12. In some passages in "How to Think about Reliability" Alston seems to construe the generality problem somewhat differently. For his purposes, a solution need only show that there are "objective, psychological facts of the matter that pick out a unique type as the one of which a particular process is a token", (p. 5) Thus, he is content to identify relevant types, leaving as a different matter the question of the acceptability of the resulting reliabilist theory. The problem discussed here is that of getting the theory stated and getting it right. Any rule of relevance that selects one type for each token will generate some reliabilist theory or other, most of them preposterous.
13. Reliabilist theories that make use of the reliability of indicators or mechanisms of belief-formation are thus not our topic. But the problems for the theory of relevance (NS3) below carry over straightforwardly to many reliable indicator theories. Also, there is a problem similar to the generality problem concerning "the mechanism" that produces a given belief. For instance, when a visual judgment relies on only black-and-white discrimination, is the person's whole visual apparatus the relevant mechanism, or is it the black-and-white sensitive portion of that apparatus, or is it only the active part of that portion? Does "the mechanism" for remembered beliefs include parts of the brain active in forming the belief, or just parts active in storing it and recovering it? These questions may have answers that are attractive to reliabilists, but as with the generality problem, the challenge is to identify a principle that implies all and only the correct answers to such questions.
14. The results of this theory may be implausible in "demon worlds" in which a demon sees to it that believing in accord with one's evidence does not reliably lead to truths. Whether this is a decisive objection to our evidentialist pseudoreliabilism depends in part on how reliability is measured. The objection as it is often described makes the challengeable assumption that a process is reliable in a world only if it regularly leads to truths in that world. In contrast, see for instance William Alston's proposal, stated as (R) above. It does not imply that unreliability in a demon world entails a lack of justification. What (R) makes decisive is roughly the truth-ratio of belief-production in more typical situations.
15. "What is Justified Belief?" pp. 10-11.
16. "How to Think about Reliability", p. 13f.
17. Charles Wallis, "Truth-Ratios, Process, Task, and Knowledge", Synthese 98 (1994): 243-269.
18. See especially p. 266. Wallis relies on belief-forming strategies as part of his response to problems that he discusses for reliability theories of knowledge. It is not clear that he is attempting to solve the generality problem that is the topic of this essay. One reason for this unclarity is that Wallis is working on a concept of knowledge that is relativized to the specification of a task, unlike the traditional concept which is our topic. In any case, we do not intend to attribute to him a simple reliance on strategies as a full solution.
19. What follows is a possible solution to the generality problem, suggested by some of Alston's language, that merits a brief look. It is not what Alston proposes. His proposals will be taken up shortly.
20. In "What is Justified Belief?" Goldman introduced a distinction between belief-dependent belief-forming processes and belief-independent belief-forming processes. The former processes take beliefs, as well as other factors, as inputs and yield new beliefs as outputs. The latter processes do not take prior beliefs as inputs. Belief-dependent processes are reliable when, over a suitable range of cases, they yield true beliefs if their input beliefs are true. Furthermore, a belief resulting from a belief-dependent process is justified only if the input beliefs are themselves justified. One might hope to appeal to this distinction to help deal with some of the examples discussed in this section and elsewhere in this paper. For example, if one's background beliefs are part of the cause of one's animal classifying beliefs, then the differences in the degree of justification for the beliefs mentioned here might be attributable to differences in the degree of justification of the beliefs upon which they depend. One might therefore be able to maintain the claim that one relevant type is responsible for all the species classifying beliefs. Defenders of reliabilism have not made significant use of the belief-dependent/belief-independent distinction in their efforts to solve the generality problem. There are good reasons for this. First, it is likely that virtually all beliefs that adult humans form are partially caused by other beliefs. Hence, virtually all our beliefs result from belief-dependent processes. It is therefore doubtful that there is any acceptable way for reliabilists to account for the differing epistemic status of the background beliefs in the examples under discussion. Furthermore, some account of the reliable types for belief-dependent processes is needed. If they are identified in terms of, say, patterns of inference, then process reliabilism turns out to be equivalent to the view that a belief is justified if it results from an inference that is likely to be truth preserving from justified beliefs. This familiar view violates the spirit of process reliabilism since it uses processes only as an indirect way to refer to inferential relations. Finally, it is difficult to see just how to make use of the belief-dependent/belief-independent distinction in conjunction with the specific proposals discussed here. Consider, for example, (G). According to (G), the relevant type is determined by the level of generality of the resulting belief. Thus, according to (G), if two people end up believing that there is a giraffe nearby, they have used processes of the same relevant type. None of the details of the routes by which they got to that belief play any role in determining which type they used. One could be making an invalid inference from justified premises while the other is making an accurate classification based on background knowledge. A theory employing (G) incorrectly evaluates the two beliefs the same way. As the solutions proposed in the existing literature are discussed below, the reader is invited to note that, like (G), they do not give any role the difference between belief-dependent and belief-independent processes.
21. Alston's and Baergen's implementations of this idea are discussed below. Goldman mentions this sort of approach in Epistemology and Cognition, p. 50.
22. "How to Think about Reliability", p. 11.
23. "How to Think about Reliability", section vi.
24. Philosophers often invoke examples in which beliefs result from blows to the head or tumors. It may be that such beliefs do not result from any psychological belief-forming process type. Perhaps the explanations of such beliefs must come from a different science or perhaps psychology must be inclusive enough to account for them too, simply because they are mental effects. If some beliefs lack any psychological cause, that would present a problem for (NS2), since even these beliefs can be assessed for justification, and hence they must have a relevant type.
25. "How to Think about Reliability", p. 26.
26. Throughout this section, when we speak of maximally specific functions or types, we mean the maximally specific psychological functions or types.
27. Epistemology and Cognition, p. 50.
28. Theories can differ over exactly what counts as the input. The process type could begin at the surface of the skin, or farther in at some point where conscious experience begins, or farther out in an external cause of the experience. Alston favors perceptual experiences as the initial step (pp. 12f). He does not defend this selection. No point made here depends on any particular beginning for the causal sequence that constitutes the process.
29. Strictly speaking, the assumption may imply only that the "narrow" content of the beliefs resulting from a given relevant type will be the same. No point made here depends on the difference between narrow and broad content. Also, see note 17 above concerning the completeness of psychological explanation.
30. It is safe to assume that many of our clear vivid experiences of complex ordinary things like trees are produced only by these same ordinary things in all situations of the sort we typically encounter. Holograms, hallucinations, and perfect pictures are, at most, highly atypical.
31. One might think that the fact that Jones relies on unjustified background beliefs has some bearing on this example. That thought seems right. But (NS3) ignores this fact and suggests nothing about how to make use of it in defending a process reliabilist theory. See note 13.
32. "Reliability and Justification", The Monist 68 (1985): 159-174. The example is discussed on pp. 164f.
33. Ralph Baergen, Contemporary Epistemology, (Harcourt Brace, Fort Worth, 1995), p. 99. Contrary to what Baergen says, Feldman does not assert that the processes are of the same type. He merely points out the undesirable consequence of the proposition that they are of the same type. It is notable that this sort of example shows that common sense process types, like the visual belief forming process, do not produce beliefs of equal justification even when relativized to a fully detailed specification of the external circumstances.
34. Contemporary Epistemology, p. 100.
35. Contemporary Epistemology, p. 100.
36. Compare the water boiling example above. There seems to be no reason to think that the explanation at one level of generality is necessarily better than an explanation at any other level.
37. Knowledge and Belief, (Routledge, Chapman, and Hall, New York, 1992), Chapter VI.
38. Knowledge and Belief, p. 169.
39. Knowledge and Belief, pp. 141-2.
40. For example, Schmitt says about an example that one constraint, which favors a broad relevant type, outweighs two others that favor a narrower type (p. 171). In another case, the existence of two constraints favoring a narrower type is said to outweigh one pointing in a different direction (p. 157).
41. Schmitt does say that relevance is a "messy, more contextual affair" than some might think (p. 159).
42. "The Simple Solution to the Problem of Generality",Nous 29 (1995): 501-515. The quotation is from p. 502.
43. "The Simple Solution to the Problem of Generality", p. 502.
44. Ernest Sosa suggests a contextualist response to the generality problem in Knowledge in Perspective: Selected Essays in Epistemology, (Cambridge, Cambridge University Press, 1991). Sosa suggests in a programmatic way that relevant types are ones that can "be usefully generalized upon by us as the epistemic community of the" believer (p. 284). Sosa does not elaborate upon this idea, which is a small part of a complex theory. What he does say does not seem to identify a unique type, since multiple types may be "usefully generalized upon".
45. Although the following thesis is suggested by much of what Heller writes, it goes beyond the explicit proposals in Heller's paper. Also, it makes no use of passages suggesting that a relevant alternatives approach to a theory of knowledge solves the generality problem. We see no plausibility in this latter suggestion on its own, and no way incorporate into it the central theme of Heller's paper concerning the importance for solving the generality problem of the context-sensitivity of 'reliable'.
46. An earlier version of this paper was presented at a symposium at the American Philosophical Association in December, 1996. We are grateful to William Alston for his comments. We are also grateful to Ralph Baergen and John Bennett for comments on earlier drafts.