Causal or Conditional or Explanatory-Relation Accounts

views updated


Edmund Gettier attacked the traditional analysis of knowledge by showing that inferring a true belief from a false but justified belief produces a justified true belief that does not qualify as knowledge. Subsequent analyses of knowledge were motivated in large part by the wish to avoid examples of the type Gettier used. One way to do so is to insist that a belief must be connected in some proper way to the fact that makes it true in order for it to count as knowledge. In Gettier's examples beliefs are only accidentally true since there are no proper connections between them and the facts that make them true. Analyses that require such connections may either retain or drop the justification condition from the traditional analysis. Without it they are thoroughly externalist analyses since they require only that a belief be externally connected with the fact that makes it true, not that the subject be able to specify this connection.

One intuitive way to specify the proper connection is to say that it is "causal": The fact that makes a belief true must help cause the belief in the subject if the subject is to have knowledge. When this causal relation holds, the truth of the belief is nonaccidental. The causal analysis of knowledge therefore excludes standard Gettier-type cases, but it seems on reflection to be both too weak and too strong: too strong in that knowledge of universal propositions, mathematical truths, and logical connections seems to be ruled out if these cannot enter into causal relations; too weak in allowing knowledge when a subject cannot distinguish a fact that causes her belief from relevant alternatives. Suppose, for example, that a subject S cannot tell red expanses from green ones but believes that there is a red expanse before her whenever either a red or a green expanse is there. Then, on an occasion in which a red expanse is before S the usual sort of perceptual causal connection will hold, but knowledge that the expanse is red will be lacking.

A different way to specify the necessary connection that handles the sort of case just cited is provided by the "conditional" account. According to this account, S knows that p only if S would not believe that p if p were not true. In close possible worlds in which p is not true, it must be the case that S does not believe it. This rules out the case of the red and green expanses since, in a close world in which the expanse is not red but green, S continues to believe it is red. A further condition required by this account is that in close worlds in which p continues to be true but other things change, S continues to believe that p.

The conditional account handles both Gettier's cases and those that require the distinction of relevant alternatives. But once again there are examples that seem to show it both too weak and too strong. That the first condition is too strong can be shown by a variation on the color expanse example. Suppose that S cannot tell red from green but is very good at detecting blue. Then, on the basis of seeing a blue expanse S can come to know that there is not a green expanse before her. But if this proposition were false (if there were a green expanse before her), she would still believe it true (she would think she was seeing red). That the second condition is too strong seems clear from the case of a very old person whose mental capacities are still intact but soon will fail him. That there are close worlds in which he does not continue to believe as he does now by exercising those capacities does not mean that he cannot know various facts now through their exercise.

That these conditions are too weak can perhaps be shown by cases in which someone intentionally induces a Gettier-type belief in S. In this case, if the belief were not true, it would not have been induced in S, and yet S does not know. Such a case might or might not be ruled out by the second condition, depending on how it is specified and on how the second condition is interpreted. But there are other cases that seem more certainly to indicate that the conditions are too weak. If S steadfastly believes every mathematical proposition that she entertains, then the conditions will be met, but she will not know all the true mathematical propositions that she entertains.

An analysis of knowledge should not only accommodate various intuitions regarding examples; it should also be useful to the normative epistemologist in reconstructing the structure of knowledge and addressing skeptical challenges. The conditional account, as interpreted by its main proponent, Robert Nozick (1981), has interesting implications regarding skepticism. According to it, I can know various ordinary perceptual truths, such as that I am seated before a fire, even though I cannot know that there is no Cartesian demon always deceiving me. This is because in the closest possible worlds in which I am not before the fire, I do not believe that I am (I am somewhere else with different perceptual evidence). But in the closest world in which there is a Cartesian demon, I do not believe there is one (since all my perceptual evidence remains the same). These implications are welcome to Nozick but are troubling to other philosophers. My knowledge of being before the fire depends on the demon world not being among the closest in which I am not before the fire. But, according to the conditional account, I cannot know that this last clause is true. Hence, I cannot show that my knowledge that I sit before the fire is actual, as opposed to merely being possible, and it seems that I ultimately lack grounds for being convinced that this is so. Furthermore, implications regarding more specific claims to knowledge and skeptical possibilities are counterintuitive as well. For example, according to this account I cannot know that my son is not a robot brilliantly constructed by aliens, although I can know that I do not have a brilliantly constructed robot son.

A third way of specifying the required connection that makes beliefs true is to describe it as "explanatory." If S knows that p, then the fact that p must help to explain S 's belief. To see whether this account handles the sorts of cases cited, we would need to define the notion of explanation being used here. One way to do so is in terms of a certain notion of probability: Roughly, p explains q if the probability of q given p is higher than the probability of q in the relevant reference class (reflecting relevant alternatives); put another way, if the ratio of (close) possible worlds in which q is true is higher in the worlds in which p obtains than in the relevant contrasting set of worlds. Given this interpretation, the analysis handles the perceptual discrimination case. In it S does not know there is a red expanse before him because its being red does not raise the probability of his belief that it is relative to those possible worlds in which this belief is based on its being green. The analysis also allows knowledge in the variation that defeats the conditional account. In it S knows that there is not a green expanse before her since the fact that the expanse is not green (i.e., it is blue) explains her belief that it is not green. Since the account must allow explanatory chains, it can be interpreted so as to include knowledge of mathematical propositions, which do not enter into causal relations. In the usual case in which S has mathematical knowledge that p her belief must be explanatorily linked to p via some proof. The truth of p makes a proof possible, and the ratio of close worlds in which S believes p must be higher in worlds in which there is a proof than in the overall set of worlds.

The explanatory account needs to be filled out further if it is to accommodate cases involving intentionally produced beliefs resembling Gettier's examples since in such cases the fact that p helps to explain why the belief that p is induced in S. As an externalist account, it would also need to provide defense for the claim that S can know that p even when, from his point of view, he has no good reasons for believing p. The analysis does suggest an approach to answering the skeptic different from that suggested by the conditional account. A proponent of this analysis would answer the skeptic by showing that nonskeptical theses provide better explanations of our ordinary beliefs than do skeptical theses.

See also Epistemology; Nozick, Robert.


Goldman, Alan H. Empirical Knowledge. Berkeley, CA.: University of California Press, 1988.

Goldman, Alvin I. "A Causal Theory of Knowing." Journal of Philosophy 64 (1967): 357372.

Goldman, Alvin I. "Discrimination and Perceptual Knowledge." Journal of Philosophy 73 (1976): 771791.

Klein, Peter. "Knowledge, Causality, and Defeasibility." Journal of Philosophy 73 (1976): 792812.

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

Sainsbury, R. M. "Easy Possibilities." Philosophy and Phenomenological Research 57 (4) (1997): 907919.

Alan H. Goldman (1996)

Bibliography updated by Benjamin Fiedor (2005)