1 Notes on the analysis of knowledge—a general framework. Davidson Phil 385 I. Epistemology up to the time of Gettier Lore has it that there was a generally-accepted analysis of what it is for a subject to know a proposition. Call this the "JTB analysis" ("JTB" is short for "justified true belief"). So, we have S knows that p iff [which is read "if and only if"] and (i) (ii) (iii) S believes that p p is true S is justified in believing that p. Justification in (iii) is to be understood in terms of evidence or reasons that S could, in principle, be aware of. Then, Edmund Gettier (now at UMass) published a 2-page paper in 1963 showing that justified true belief is not sufficient for knowledge. (Russell actually noted the same sorts of counterexamples 50 years earlier.) Here is an example of a "Gettier-style counterexample" to the JTB analysis. I am a farmer and raise sheep on my farm. I'm driving through the countryside, get out of my car, and note that there is a sheepish-looking animal in the field 10' in front of me. So, I form the belief (1) There is a sheep in the field 10' in front of me is true. I am justified in this belief; I am familiar with sheep and have ample reason to believe this is a sheep. I'm also obsessive about obtaining true beliefs. I realize that if there is a sheep in the field 10' in front of me, it follows logically (and I am justified in believing it follows logically) that there is a sheep in the field somewhere. So I form the belief (2) is true. There is a sheep in the field somewhere 2 I seem to be justified in believing that (2) is true; I inferred it from a justified belief (the belief that (1) ) and I used a form of inference that I am justified in believing is reliable. So the justification seems to transfer from (1) via a justified inference to (2). Unbeknownst to me, the thing 10' in front of me is a cleverly-disguised dog, and not a sheep. So my belief that (1) is true is false. But, suppose that there is a sheep hidden behind a bush. I have no reason to think there is; I can't see the sheep. But then, my belief that (2) is true is true and justified. But do I know that (2) is true? No, it seems. My belief is true, but accidentally so. I got lucky that there was the sheep behind the bush, otherwise, my belief that (2) is true would have been false. Knowledge can't be accidental or lucky in this way. So, we have a case of justified true belief with (2), yet it is not a case of knowledge. So, justified true belief is not sufficient for knowledge. II. Where to go after Gettier? There have been two sorts of reactions to Gettier-style counterexamples. The first, taken by most internalists, is to say that though the JTB account is not sufficient, we can add a fourth condition to the analysis to take care of Gettier cases. (Call this the JTB+ approach.) So, knowledge might be justified true belief along with a fourth condition like if p were false, S wouldn't believe that p. (Note that with the sheep case, this condition wouldn't be satisfied, because I still would believe that (2) is true, even if it were false (presumably via there being no sheep behind the bush).) Many fourth conditions have been proposed; see Robert Shope's An Analysis of Knowing for various fourth conditions on knowledge. The second sort of path to take in response to Gettier is to significantly modify the JTB account. People like Alvin Goldman (Rutgers) Robert Nozick (formerly of Harvard) and Fred Dretske (Stanford, Duke) proposed dropping the justification condition—at least as understood in terms of evidence and reasons that are such that whether the justification condition holds is a matter that is accessible to the subject. So, from Dretske ("Conclusive Reasons", 1971) we get. and S knows that p iff (i) S believes that p (ii) p is true (iii) S has a reason for believing that p such that if p were false, R would be false. Robert Nozick (Philosophical Explanations) gives a condition that is similar to Dretske's: If p were false, S wouldn't believe that p. (Consider what happens here when there is someone waiting in the wings to maintain the truth of the reason or the cause of the belief, even if p is false.) 3 Here's another fourth condition: There is no true proposition q such that if S came to believe q S's justification for p would be defeated. (Consider a situation where q is something like: S believes x is standing in the field. Suppose this is true. But S doesn't know that x has a twin, y, and y has been standing in the field quite a bit lately. This might show this is false.) Here is another fourth condition: None of one's reasons for believing that p are false. This may be overly restrictive; we may want some knowledge to be obtainable on the basis of false reasons, just so long as there is enough reason to stand without the false beliefs (or something like this). Goldman ultimately formulated an analysis of knowledge called Reliabilism (see Goldman "What is Justified Belief?" 1979; and Epistemology and Cognition (1986)). "Generic" reliabilism may be stated thus S knows that p iff (i) S believes that p (ii) p is true (iii) the belief that p is the result of a reliable belief-producing mechanism or process. In response to Gettier, Goldman would say that the belief that (2) is true is not the result of a reliable belief-forming process (e.g. inference from a false proposition). Note that S need not even entertain the question of whether (iii) above holds. So long as a belief is true and the result of a reliable belief-forming process, one can have knowledge. This allows animals and small children to have knowledge (whereas on the JTB+ account, it is argued that the justification condition may put knowledge beyond animals and small children). Now we introduce a new term, warrant. Warrant is whatever together with true belief is sufficient for knowledge. So, S knows that p iff a) S believes that p, b) p is true, and c) S is warranted in believing that p. For the JTB+ theorist, warrant will be justification and satisfaction of some fourth condition on knowledge. Justification is the "key" part of warrant for the JTB+ theorist. For the reliabilist, warrant (so far as we've seen) will consist in a belief's being produced by reliable processes or mechanisms. III. Internalism and Externalism We have two new definitions: 4 x is an internalist analysis of knowledge iff on x, the key component of warrant is inprinciple introspectively accessible to the subject. The following also will be true: x is an internalist analysis of knowledge iff the key component of warrant is not "necessarily" truth-conducive (it could hold in an evil demon world). Evidentialism and deontologism are two sorts of internalist theories. The evidentialist (Feldman and Conee) analyzes justification in terms of having sufficient evidence. The deontologist analyzes justification in terms of not violating epistemic duties (e.g. the duty not to form beliefs based on processes one knows to be unreliable). x is an externalist analysis of knowledge iff on x, the key component of warrant need not be introspectively accessible to the subject. The following also will be true: x is an externalist analysis of knowledge iff the key component of warrant is "necessarily" tied to truth (it wouldn't hold in a demon world). So, for instance, though one might have justification for a sensory belief in a demon world, sense perception would not be a reliable belief-forming process in such a world. So, the key component of warrant for a reliabilist does not hold in a demon world. Reliabilism is the main externalist theory. Dretske's theory (above) is externalist, as well. There are many sorts of reliabilisms; we've looked at a very bare-bones version of it. IV. Arguments for Internalism and Arguments for Externalism Arguments for Internalism A. The Demon World argument (Bonjour, Cohen, others): Were we in a Cartesian situation where we were systematically deceived by an evil demon, the key component of warrant surely could be present—we'd still be reasonable in holding or entitled to hold certain beliefs, even if most of our beliefs were false. The externalist—and reliabilist in particular—can't allow for this. B. The clairvoyance/tea leaves argument (BonJour The Structure of Empirical Knowledge, elsewhere; others). This is an argument against reliabilism. Suppose I read the arrangement of tea leaves at the bottom of a tea cup after I finish my tea, and I form beliefs about the weather in Moscow based on this. I have no reason to think that the tea leaves will tell me anything about the weather in Moscow. Indeed, I have reason to think that beliefs about Moscow weather are likely to be false. But, it turns out (unbeknownst to me) that I form accurate beliefs about Moscow weather based on the tea leaves. Then, this mechanism can turn a true belief into knowledge. Yet, surely it shouldn't do so. I have reason to think that the belief-forming process is unreliable. In response, the reliabilist often will add a "no-defeater" clause—her own fourth clause. So, the clause will be something like (iv.) it's false that S has a defeater for her belief that p where we have the following (rough) analysis of a defeater 5 A belief b is a defeater for a belief b' if b reduces the justification for b' such that S no longer ought to hold b'. So, let b' be my belief that it is sunny in Moscow and let b be my belief that the method by which I arrive at Moscow weather beliefs is unreliable. (Things are much more complicated than I suggest here—see Plantinga Warranted Christian Belief for a good discussion of defeaters.) C. The Generality Problem for reliabilism (Feldman and Conee): We will talk more about this in class, but for each belief, there are all sorts of processes that may (accurately) be said to have produced the belief. If I believe it is sunny outside the following processes all are appropriate descriptions of processes that went into the formation of the belief: a) sense perception, b) vision, c) vision on a sunny day, d) vision on a sunny day on Saturday, April 16th 2005 at 2:05 pm, and so on. The more specific we get in specifying the process, the closer we get to having the process instanced only a few times—or one time—and having the process be reliable so far as it produces a true belief on that one occasion. For instance, suppose we say that the process that gave rise to my belief that it's sunny outside is d (above). Then, so long as the belief formed by the process is true, the process is 100% reliable and I have an instance of knowledge if my belief is true. This becomes problematic when add significant amounts of detail to what ought to be unreliable processes (reading tea leaves), and the process happens to be reliable during the specified time. Then I could have knowledge (based on reading tea leaves at 4pm on Saturday April 16th 2005 while sitting down drinking coffee wearing a white t-shirt, etc.) If this very detailed process gives rise to a true belief, it's 100% reliable—it is used just once. Arguments for Externalism A. Internalism makes knowledge too difficult to obtain. Lots of individuals don't have the sort of introspective access to reasons that an internalist requires (we'll see in class how complicated things get). Dogs certainly don't have this sort of capability. Yet dogs know when it's dinner time or when a stranger is approaching. The reliabilist, for instance, can have knowledge with true belief that is reliably produced, whether or not the subject ever has thought of how the belief was produced. B. Externalism gives an answer to skepticism On our basic analysis of reliabilism, we don't need to have any sort of reason to think that we're not being deceived by an evil demon. We don't need any reason to think our belief 6 is true at all, in fact (except when reasons are a part of the belief-producing process). If our belief is produced by a reliable process and is true, we will have knowledge— whether or not we ever think about how it was produced. So, the skeptic who puts very stringent requirements on knowledge—that one be able to rule out all sorts of possibilities that we've systematically false beliefs is off-the-mark. Descartes worried about this sort thing, but for the externalist-reliabilist, there is an easy answer: Descartes knows he has a body if his belief that he has a body is true and produced by reliable faculties or processes. V. Some reflections If the argument that the reliabilist (and externalist more generally) needs a no-defeater clause in her analysis of knowledge is a good one, then the reliabilist has an "internalist" sort of clause in the analysis of knowledge—the no-defeater clause has to do with evidence, and evidence is something to which we have introspective access. Arguably, any fourth condition the JTB+ theorist adds will be "externalist"—that it holds will not be introspectively accessible. If it were introspectively accessible, then one could "look inside" and see if one is being Gettiered. But one can't do this. Knowledge turns out to be lucky in Gettier cases precisely because it is beyond one's ken introspectively that Gettier conditions obtain. So an internalist theory (JTB+) will have an externalist component to warrant, though the main component (justification) will be internalist. And, our paradigm externalist theory-reliabilism has an internalist component to warrant, though the main component (reliability of belief-forming processes) will be externalist. But, we see that prima facie, there is some overlap between internalist and externalist theories.