What is infallibility? I realized, last time I taught epistemology, that simple explanations of the idea don’t work. I’m working on a paper where I need to be more careful, and am excerpting the relevant section here, to see what others think. I’ll put the section below the fold.
As a first pass, we might try this: infallibility amounts to, as Lewis maintains, being able to rule out every possibility of error:
It seems as if knowledge must be by definition infallible. It you claim that S knows that p, and yet you grant that S cannot eliminate a certain possibility in which not-p, it certainly seems as if you have granted that S does not after all know that p. To speak of fallible knowledge, of knowledge despite uneliminated possiblities of error, just sounds contradictory. (Lewis, “Elusive Knowledge”)
Lewis’s view, however, remains murky, because of its reliance on the language of “ruling out.” A natural interpretation of this phrase is in terms of knowledge itself: one rules out a possibility when and only when one knows, or is in a position to know, that it does not obtain. Yet, once we endorse a plausible closure principle about knowledge–that one can come to know the logical consequences of what is already known by competently deducing them–infallibility, on this construal of it, can be a consequence of epistemic closure alone, which is too weak an interpretation.
One might try to avoid this weakness by referring to the evidential basis of the knowledge in question, and require that the ruling out be done by it alone. Fallibilists, however, can still trivialize the view: evidence e rules out all ~p possibilities when and only when e provides an adequate epistemic basis for believing p. To avoid such trivialization of infallibilism, a retreat into the refuge of logic seems fortuitous: e rules out all ~p possibilities when and only when e entails p. Thus, it might seem, the heart of infallibilism is found in the demand for logical guarantees of truth in one’s evidence base.
This characterization can only work if insists on excluding p itself from e, but this restriction is appropriate: we don’t want an account of knowledge that allows knowledge of p to arise in virtue of the fact that p entails p. Even so, this account of infallibility remains too easy to satisfy. Suppose e is adequate evidence for p. Then the epistemic conditional “if e,p” is contained in the relevant system of information for beings who are logically and epistemically omniscient, and even for the rest of us, reflection can make this conditional available as part of our total evidence. Once present, however, we have evidence available that entails p, and thus generates infallibility by the above standard.
Even though one’s evidence, including epistemic conditionals, entails the belief in question, it needn’t soundly entail it, if we allow false claims to be in the set of evidence (as I believe a sensible epistemology should). There is no refuge here, however, for defining infallibilism in terms of sound entailments, for we should prefer an account of infallibilism that doesn’t depend on whether I’m right about the nature of evidence. If nothing false is ever evidence for anything, then the above argument shows that bodies of evidence too easily generate infallible beliefs.
One might try to save the evidential approach to infallibility by adopting a restrictive condition on proper basing. In the case above, one might insist that it isn’t the total system of evidence that matters, but rather only the part that confers justification on p–namely, e itself.
This approach cannot succeed, however, for there simply is no legitimate restrictive requirement on proper basing. Note first how this point holds with respect to ordinary, fallible justification. Recall that, underlying instances of epistemic support are general epistemic principles. A good example of such is Chisholmian: if S is appeared to F-ly and lacks grounds for doubt that something is F, then it is reasonable for S to believe that something is F. I’m not endorsing this principle, but only using it for purposes of illustration, since its antecedent makes reference to both a conferrer of justification–the appearance state in question–and an enabler–the lack of grounds for doubt. Given such a principle, the above restriction on proper basing insists that, relative to such a principle, one can only base a belief properly on the appearance state in question. That restriction is mistaken, however. S might, for example, reflect on whether there are grounds for doubt concerning the claim in question, and might come to believe that there are no such grounds for doubt, and might then come to believe that something is F, basing that belief on the appearance state in question together with the belief that there are no grounds for doubt concerning it.
S needn’t so base the belief after reflection, but may do so legitimately. S might reflect and come to believe that there are no grounds for doubt, and still might rationally believe that something is F by basing it only on the appearance state in question. The additional reflective belief is allowed to be part of the basis of the belief that something is F but it is not required to be part of the basis even when it exists. The proper conclusion to draw here is that proper basing comes in a variety of flavors, and any account of infallibility that restricts these options is bound to fail.
Not only can a belief be properly based on more than a minimal conferrer of justification, it can also be based on a minimal basis when something stronger is available. Consider the Cartesian project of restoring confidence in our standard belief system in the face of the skeptical doubts in the Meditations. Descartes’ desire is to provide an argument for the conclusion that mistakes in our system of beliefs occur when and only when our will outruns our understanding–put colloquially, when our non-alethic interests, desires, and motivations could our purely intellectual judgment. In attempting to restore confidence, Descartes provides an argument for this conclusion. The argument fails, but let’s ignore that issue for the moment and suppose it had succeeded. Moreover, suppose it were transparent to us whether we were motivated by non-alethic factors. If these assumptions held, the Cartesian project of defending our systems of belief against skepticism would succeed. Our present system of beliefs (properly adjusted to include nothing ill-motivated) would be a system in which we could place full confidence, and it would meet the most stringent infallibilist strictures. But nothing in this story requires that we base our beliefs on the Cartesian system itself in order to satisfy infallibilist strictures. The system must withstand scrutiny, and perhaps those with the secure knowledge undergirded by this system must be aware of the arguments involved in it, but there is no reason to insist that ordinary beliefs, formed in the ordinary way, are not infallibly known just because they are not also based on the system itself. The reflective knowledge about how the system works and why it is adequate is enough when one bases, say, a perceptual belief on an appearance state while recognizing that the belief is motivated by a concern for understanding alone.
One might hold out hope that infallibility could be defined in terms of entailing minimal conferrers of justification. Instead of thinking in terms of a proper basis of justification, one might think of all the proper bases, and select the minimal one available, and insist that it entail the truth of the belief in question. Or, if one is concerned that there may be no unique minimal one available, we can speak instead of one’s basic evidence, and insist that it be in virtue of one’s basic evidence that infallibility obtains.
Such a restriction has implausible consequences. One’s basic evidence will consist either of beliefs or something else, such as experience and its contents. If infallibility is a product of entailment from the set of basic evidence, then any belief included in the set of basic evidence will itself be infallible. Such a conclusion once again makes certain instances of infallible belief too easy to come by: mere basicality of belief shouldn’t ensure infallibility. So one’s basic evidence can’t include any beliefs, or any other item that is itself capable of being justified or unjustified.
Moreover, there is no good reason to restrict the evidence to which an account of infallibility can appeal in a way that will avoid this problem. For human beings in ordinary circumstances, basic evidence will involve some collection of beliefs and experiences, and if beliefs are excluded from the set of basic evidence, infallibility would then depend solely on the justificatory power of experience itself. Note that in such a case, the Cartesian project was hopelessly confused from the beginning. Had that project succeeded, our ordinary conception of the world would have been shown to be infallible, but nothing in the Cartesian strategy hoped to demonstrate this point from experiential inputs alone. So it is an objectionable account of infallibility that it must result solely from some logical connection between experience and belief.
I suggest that the lesson here is that we shouldn’t expect to find an adequate account of infallibility in terms of evidence and what it entails. Instead, I think we should exploit modal dimensions of knowledge, looking for an account of infallibility in terms of some strengthening of the epistemic modal notions of safety and sensitivity. A belief is safe when it would normally be true when held, and a belief is sensitive when it would not be held when false. These notions need to be made modally stronger, however, requiring not just a counterfactual connection between truth and belief but a necessary one: in some sense, the belief couldn’t be held while false and couldn’t be absent when true.
Pursuing this approach will lead us to consider whether a person lands in error on the issue in question in other possible worlds. We can’t proceed in a way that requires existence in all other worlds, since there is no reason to think that only necessary beings can be infallible. Nor should we include worlds where an actual person exists but is incapable of thought. Moreover, even in worlds where actual cognizers retain their cognitive powers, there is the issue of whether they ever attend to or consider the claim in question. One would think that considering a claim is often cognitively optional, and that failure to consider the claim in question in some other world say anything at all about one’s strength of epistemic position in the actual world. Even so, there is a danger with restricting the worlds that are relevant to those in which the person in question considers the proposition in question. One’s concern about p might be so fragile that only in the actual world does one consider whether p and in the actual world one merely follows a hunch in believing p. By luck, one is correct, but having such a fragile interest in p shouldn’t allow one to be infallible concerning p just because there is no world in which one considers p and is mistaken about it.
We can refer to this last issue as the issue of “optional cognitive interests”. It is a psychological issue, and it is not easy to see how to find a middle way between two mistakes. One mistake is to disbar a belief from being infallible simply because there is some world where the individual never bothers to check out the issue in question. The other mistake is to let a belief move from the set of fallible beliefs to infallible simply because the person’s interest in the issue in question is maximally fragile.
Even if we found some middle way through the issue of optional cognitive interests, such an account will have difficulty in the face of the Cartesian project. We want an account of infallibility so that if the project had succeeded, we would all be in a position to have a wide range of infallible beliefs: avoiding error would be totally up to us, requiring only that we don’t let our wills outrun our understanding when it comes to belief. If it were transparent to us whether our will were outrunning our understanding, then we could know infallibly that our conception of the world is correct, but we would not thereby satisfy the definition above. For we could have such Cartesian certainty even though there are worlds where we exhibit intellectual weakness of the will and allow our will to influence our beliefs.
This latter problem shows that if we are going to pursue a modal approach to infallibility, it will be useful to combine it with some reference to evidence or total informational system in some way. The motivation for relativizing to a total system of information is that it appears that it will be needed to address the problem arising from the Cartesian product, and it also offers hope of helping with the problem of fragile concerns for certain propositions, since the system of information would show in such a case why the belief in question is fallible even if true. One approach along such lines is:
S’s belief with content p, relative to and based on S’s informational state s, is infallible iff there is no possible world in which S is in s and in which (i) p is false and (ii) S believes ~p on the basis of s.
P.S. I do not believe that this account is fully adequate, but it is adequate enough for the purposes of my paper. It is similar to an approach recommended by Julien Dutant, in “The Case for Infallibilism.” Dutant, too, distinguishes between evidential and modal approaches to infallibilism, but his modal approach differs slightly from the present one, focusing on the basis of belief rather than the informational system in question. Because of the points raised earlier about the optionality involved in basing, I think the latter approach has more hope of success here.
One worry about it is that the connection between the belief and the information state might be too fragile to sustain the status of infallibility. Suppose there is a hard-to-see feature of the tree that guarantees it has a certain disease. You’ve examined the tree multiple times, trying to decipher what is wrong with it. You’ve failed every time, but today you finally notice the feature. Your friend has believed all along what you’ve now come to believe, but he doesn’t notice the feature. You are much more careful about belief, so careful that there is no world in which you believe ~p in the presence of this new information. So your belief is infallible, since the hard-to-detect feature is a decisive sign of the disease.
To solve this problem, we might try revising recursively, insisting that infallibility depends not only on the modal feature in question but on infallibility with respect to the information system in question as well. Some form of formal foundationalism would require defense in order to provide a suitable base clause for such an account, but I won’t pursue these issues here, since my primary concern in this post is to lay out a problem and a possible direction for solving it to see what others think.