bigideas

The standard definition of “knowledge” was, for a while, “justified true belief”. Say I claim to know that my cat is sitting on the mat (to use a hoary old example). First, I must actually believe it; I can’t know something but simultaneously hold it false. Second, what I know must be true; it really must be that the cat is on the mat, otherwise we say I hold a false belief rather that I have some knowledge. Finally, my opinion must be justified. If I simply guess that the cat is on the mat then we can’t say I know it until I’ve checked. Even if it turns out I was right all along, without justification the best we can say is that I got lucky and my belief turned out to be correct. Only when it’s justified am I entitled to call it “knowledge”.

In 1963, Edmund Gettier published a very short paper, “Is Justified True Belief Knowledge?”, that upset the epistemological applecart by proposing two cases of justified true belief that we would be unwilling to count as “knowledge”. Cases like these have become known as “Gettier Problems”. Certain solutions were found for his original examples, but later, strengthened ones remains a serious problem for epistemologists who hold with the old theory.

A modern form of the Gettier Problem goes like this. Say I walk into a room and see what appears to be John sitting in a chair. It is John’s room, and the person certainly looks like John, so my belief that John is in the room seems justified. Yet in truth John is hiding somewhere in the room and this person is his twin brother James. My belief was true (since John is in the room) and justified (since I had good evidence for it) but it feels quite wrong to call it “knowledge”. It seems like an accident that we were right, just as it was when we guessed about the whereabouts of the cat.

One popular approach to this proposed by Fred Dretske is that the “justification” part involves “ruling out relevant alternatives“. It’s not reasonable, whenever I meet someone, for me to investigate the possibility that they have a hitherto unknown identical twin, but if later I discover that they do, and that I mistook one for the other, I should adjust my assessment of whether or not I had knowledge. In other words, it’s all about known unknowns and unknown unknowns.

It would work like this. When I walk into the room, I believe I know that John is there. However, my belief is not as justified as I think. When John jumps out of the cupboard I quickly dismiss my belief that I knew John was there as false, although the belief that John was there turned out (by chance) to be true. We’ve moved the justification up a level.

Of course it’s tempting to move it up another level, and another one. Where does it stop? And how many possible alternatives must we admit? Too much inclusivity leaves us open to evil demon and brain-in-a-vat arguments, which lead to radical skepticism. Yet not enough leaves us accepting as knowledge things that turn out to be true only by accident.

These problems, and a proposed solution, were the subject of David Lewis’s 1996 paper “Elusive Knowledge”. I won’t paraphrase what he has to say there since it’s intelligent, often witty, extremely readable and, like Gettier’s paper, available online. I’m attracted to most of what he says, although I wonder whether his — or any — list of criteria is a bit ad hoc.