To know a proposition, is it necessary that one is able to rule out every possibility of error associated with that proposition? Notoriously, infallibilism about knowledge—as defended, for example, in early work by Peter Unger (1975)—demands just this and argues on this basis for the skeptical conclusion that knowledge is rarely, if ever, possessed. Intuitively, however, the answer to this question is "no," in that in everyday life we only demand that knowers rule out those error-possibilities that are in some sense relevant. For example, to know that the bird before me is a goldfinch, I may be required to be able to rule out that it is not some other bird that could be in the area just now, like a jackdaw, but we would not normally demand (at least not without special reasons) that I be able to rule out the possibility that it is not a mechanical goldfinch made up to be an exact replica of the real thing.
If this line of thought is right, then this prompts a relevant alternatives (RA) theory of knowledge that demands that one only needs to be able to rule out all relevant error-possibilities in order to know, not that one is able to rule out all error-possibilities, even irrelevant ones. (A similar view could be applied to other epistemic notions, like warrant or justification. For simplicity, the focus here is on knowledge.) Such a position would thus be a form of fallibilism, which is directly opposed to infallibilism and which thereby counters those versions of skepticism that are based on infallibilist considerations. The task at hand for the RA theorist is to offer a principled account of what makes an alternative relevant.
Relevant Alternatives and Sensitivity
One can find the beginnings of an RA theory of knowledge in the writings of such figures as Ludwig Wittgenstein and John Austin. The first worked out versions of an RA theory, however, can be found in the works of Fred Dretske (1970) and Robert Nozick (1981), who primarily understand knowledge in terms of the possession of beliefs that are sensitive to the truth in the following manner:
An agent, S, has a sensitive belief in a true contingent proposition, p, if and only if, in the nearest possible worlds in which p is not true, S no longer believes p.
To illustrate this, consider again the example of the goldfinch discussed earlier. Given that the actual world is roughly as we take it to be, gaining a sensitive belief in the proposition, P, that there is a goldfinch before one is relatively straightforward. All one needs is a true belief in this regard and, in the nearest possible worlds where P is no longer true—where, for example, the goldfinch has flown away leaving behind just an empty branch—one no longer believes that there is a goldfinch there, as presumably one does not. Notice that the relevant possible worlds here are limited and concern error-possibilities (e.g., that there is nothing at all on the branch rather than a goldfinch), which are easy to rule out. A theory of knowledge that treats sensitivity as the key requirement on the acquisition of knowledge is thus in a good position to capture the intuition that ordinarily we do not demand that agents are able to eliminate all possibilities of error before we count them as possessing knowledge.
Interestingly, however, the sensitivity-based approach does treat far-off possible worlds, and thus far-fetched error-possibilities, as sometimes being relevant to the possession of knowledge. Consider, for example, the hypothesis Q, that there is a mechanical goldfinch before one, constructed in such a way as to be indistinguishable to the naked eye from the real thing. When one is faced with what seems to be a goldfinch (and circumstances are, apparently, entirely normal), does one know not-Q? According to the sensitivity-based account of knowledge, this is unlikely because it is difficult to have a sensitive belief in not-Q. After all, to have a sensitive belief in this proposition it would be necessary to have a belief that was not only true in the actual world, but that was also no longer held in the nearest possible worlds in which not-Q is false—that is, those worlds in which Q is true, where one is at present looking at a mechanical goldfinch. The problem is, of course, that, ex hypothesi, one would continue to believe that one is looking at a real goldfinch even when one is faced with a mechanical goldfinch, at least unless one conducted special tests (such as capturing the "creature" and cutting it open). So while knowing P is relatively easy, knowing not-Q is hard. And notice that the reason this is the case is because the range of possible worlds, and thus the range of error-possibilities, that is relevant to the determination of one's knowledge is different in each case.
Relevant Alternatives and Nonclosure
On the face of it, this rendering of the RA theory seems to capture our pretheoretical intuition that in normal circumstances we ought to be able to know that we are looking at a goldfinch even though we are unable to rule out (i.e., know to be false) the hypothesis that we are looking at a mechanical goldfinch. Nevertheless, this view does have a counterintuitive result, one that both Dretske and Nozick are prepared to accept. This is that the highly intuitive principle that knowledge is "closed" under known entailment ("closure") has to be rejected. We can roughly formulate closure as follows:
Closure for Knowledge
If an agent, S, knows a proposition, p, and S knows that p entails a second proposition, q, then S also knows q.
For example, if one knows P, then given that one also knows that P entails not-Q (as surely one does), it follows from closure that one must know not-Q. Conversely, of course, if one fails to know the latter proposition, which is what the sensitivity-based approach predicts, then one fails to know the former.
Closure is highly intuitive and yet, as we have just seen, if it holds it would appear to license a restricted form of infallibilism. For although closure does not demand that it is a precondition on knowledge possession that one is able to rule out all possibilities of error, it does demand that one is able to rule out (i.e., know to be false) all those error-possibilities that are known to be inconsistent with what one knows, and this set of error-possibilities, while smaller, is large enough. This point is important, since if the appeal of infallibilism rests on the appeal of closure, then the view is on far stronger ground that one might have initially supposed because of the obvious appeal of the closure principle.
Nevertheless, Dretske and Nozick argue that recognizing that sensitivity is a necessary condition for knowledge highlights why this principle must go, since there are clearly cases, such as the goldfinch example, where one knows one proposition (and thus has a sensitive belief in this proposition) and knows that this proposition entails a second proposition, and yet one lacks a sensitive belief in the entailed proposition and thus fails to know it.
Relevant Alternatives and Contextualism
Although the sensitivity-based proposal has been influential, it does face the problem that it denies the highly intuitive closure principle for knowledge, and this has led some commentators to try to see if there is a way of accommodating the general intuition behind the RA theory in a way that preserves this principle. One of the guiding considerations behind views that try to offer an RA thesis that is consistent with closure is that the Dretske-Nozick treatment seems to incorporate the idea that closure fails because while sometimes knowledge is hard to attain, sometimes attaining it is relatively straightforward. This tends to suggest that an alternative way of approaching the issue could be to regard knowledge as in some sense context-sensitive, so that one knows both of the target propositions in the closure-based inference relative to one set of epistemic standards (the less demanding ones), but knows neither of them relative to another set of epistemic standards (the more demanding ones). We would thus get a view that incorporates a reading of the RA intuition—because it would remain that not every error-possibility is always relevant to the possession of knowledge—but which was also consistent with closure. This view—known as contextualism about knowledge—is hinted at in an early response to Dretske's denial of closure written by Gail Stine (1976), and has been developed by Stewart Cohen (1991), Keith DeRose (1995), and David Lewis (1996).
Consider again the goldfinch example. On the Dretske-Nozick view the class of possible worlds, and thus the class of error-possibilities, that is relevant to the determination of knowledge can differ depending on the content of the proposition at issue, which is why in this case the agent comes out as knowing P while failing to know not-Q, despite knowing that the former entails the latter. The reason for this is that when it comes to knowing P only nearby possible worlds are relevant, whereas knowing not-Q brings in farther out possible worlds. Suppose instead, however, that one simply treated the class of possible worlds as fixed in each context, so that the epistemic status of all beliefs—whatever their content—were in that context evaluated relative to those possible worlds. In normal contexts, then, only nearby possible worlds would be relevant, while in more demanding contexts far-off possible worlds would become relevant. This way of understanding knowledge would mean dropping sensitivity as a requirement on knowledge, of course, since there may be no nearby possible worlds in which the target proposition is false (this is, indeed, what we would expect to be the case when it comes to the hypothesis that one is at present looking at a mechanical goldfinch). Nevertheless, the guiding thought here is that so long as the agent's belief matches the truth in the relevant possible worlds—that is, where the agent believes that proposition, it is true; and where the proposition is not true, the agent does not believe it—then the agent's belief will be in the market to be counted as an instance of knowledge.
By contextualist lights, then, in contexts where the epistemic standards are low (and thus only nearby possible worlds count as relevant) one will tend to know both P and not-Q, since even one's belief in not-Q will tend to match the truth (i.e., one believes it in all nearby possible worlds and it is true in all nearby possible worlds). In contrast, in contexts where the epistemic standards are more demanding, and thus where farther out possible worlds become relevant, it will now no longer be the case that one will tend to know either of these propositions. After all, there will be possible worlds, such as the far-off world in which there is a sophisticated plot to deceive people about the presence of goldfinches, in which one's beliefs in P and not-Q no longer match the truth. Thus, as long as one consistently sticks to a specific epistemic standard then this construal of the RA intuition is not in conflict with closure since, depending on the context at issue, either one has knowledge of both of the target propositions or one has knowledge of neither of them.
Relevant Alternatives and Safety
In more recent work, however, a third rendering of the RA thesis has come to the fore, one that is neither contextualist nor results in the denial of closure. This position—defended, for example, by Ernest Sosa (1999)—holds that far-off possible worlds are always irrelevant to knowledge, whatever the content of the target proposition or the context at issue. Accordingly, one is able to know, for example, both P and not-Q, whatever the context.
This view tends to hold that the key condition that a belief must meet if it is to count as knowledge is that it be safe. Safety can be roughly formulated as follows:
An agent, S, has a safe belief in a true contingent proposition, p, if and only if, in all nearby possible worlds in which S believes p, p is true.
Notice that contextualists will have to appeal to something like safety to explain how agents can know a proposition like not-Q in epistemically undemanding contexts where there are no nearby worlds in which what is believed is false. The point will be that while such beliefs are not sensitive, since there is no relevant Q-world for them to be sensitive to, they are safe, in that the agent's belief in not-Q is always true across the relevant possible worlds. What is important about safety for our purposes is that it simply specifies the class of possible worlds that is relevant and leaves the matter at that—there is no room here for a shift in context that would in turn alter the class of possible worlds, and thus the class of error-possibilities, that is relevant to the determination of knowledge. Accordingly, it will not matter which context one is in. Just so long as one's beliefs that P and not-Q are both safe—as presumably they will be—then one is in a position to know both of these propositions and thus there is no tension with closure.
There are thus three competing conceptions of the RA intuition in the literature. The first view treats relevance as being determined by the content of the proposition known, and as a result maintains that the closure principle for knowledge fails. The second view treats relevance as being determined by context, and thereby retains closure. Finally, the third view also retains closure, but does so by maintaining an invariant standard of relevance, regardless of the content of the target proposition or of the context at issue.
Austin, J. L. "Other Minds." In Philosophical Papers. Oxford, U.K.: Clarendon Press, 1961.
Cohen, Stewart. "Skepticism, Relevance, and Relativity." In Dretske and His Critics, edited by Brian P. McLaughlin. Cambridge, MA: Basil Blackwell, 1991.
DeRose, K. "Solving the Sceptical Problem." Philosophical Review 104 (1995): 1–52.
Dretske, Fred. "Epistemic Operators." Journal of Philosophy 67 (1970): 1007–1023.
Lewis, D. "Elusive Knowledge." Australian Journal of Philosophy 74 (1996): 549–567.
Nozick, Robert. Philosophical Explanations. Cambridge, MA: Harvard University Press, 1981.
Sosa, Ernest. "How to Defeat Opposition to Moore." Philosophical Perspectives 13 (1999): 141–154.
Stine, Gail. "Skepticism, Relevant Alternatives, and Deductive Closure." Philosophical Studies 29 (1976): 249–261.
Unger, Peter. Ignorance: A Case for Scepticism. Oxford, U.K.: Clarendon Press, 1975.
Wittgenstein, Ludwig. On Certainty. Translated by Denis Paul and G. E. M. Anscombe, edited by G. E. M. Anscombe and G. H. von Wright. Oxford, U.K.: Blackwell, 1969.
Duncan Pritchard (2005)