"Reliability" in Epistemology
There are two ways of being unreliable. Something, or someone, might be unreliable because it often goes wrong or yields the wrong result, or it might be unreliable because it fails to do anything or yield any result at all. A secretary is unreliable in one way if he fouls up the job, unreliable in another if he simply doesn't do it. A program for delivering stock prices is unreliable in one way if it tends to misquote, unreliable in another if it crashes. Either way, they can't be depended on to do what they ought.
Contemporary epistemologists tend to classify only the first sort of failure as a failure in reliability. Here's Alvin Goldman, probably the world's leading "reliabilist":
An object (a process, method, system, or what have you) is reliable if and only if (1) it is a sort of thing that tends to produce beliefs, and (2) the proportion of true beliefs among the beliefs it produces meets some threshold, or criterion, value (1986, p. 26).
We can easily liberalize this definition to accommodate the stock quote program: The stock quote program is "reliable" on this definition if most of its quotes are right, no matter how much it crashes or how rarely it successfully delivers a quote when asked.
This peculiarity serves a purpose: For reliabilists, knowledge and justification require (something like) reliability -- and what matters in knowing or being justified, it seems, is that you're not likely to err. Regardless how glitchy the stock quote program is, if whenever it does happen to give a quote it gives the right quote, you can have knowledge and justification from it (setting aside some complexities).
Yet I wonder if epistemologists haven't lost something valuable in giving up on the ordinary notion of reliability. In cognition -- for example in introspection -- the difference between failing to reach a judgment about whether you have (e.g.) very detailed current imagery or not and reaching the wrong judgment about that is sometimes vague and cognitively minor. Either way, introspection (like the secretary or stock quote program) has failed to deliver what one might reasonably hope it should. There's often no firm line between guesses, conjectures, impressions, and definite opinions, and whether one expresses oneself hesitantly, unhesitantly, or not at all -- to oneself or aloud -- may depend more on context and temperament than anything else. These different ways of failing must of course be distinguished -- yet drawing too sharp a distinction between them, and giving them vastly different roles in our epistemology, is artificial and misses something important.
So: When I say introspection is unreliable, I mean that in the broad, ordinary sense of "unreliable". It is no objection to my pessimism, but rather supports it, if the reader or general population fails to reach introspective judgments about their experience -- as long as it's a case where it seems like introspection should be able to deliver results, like a basic and pervasive aspect of currently ongoing conscious experience patiently considered.