January 26, 2008

The Strevens talk - a Michel reaction

Though I had volunteered to write up a quick review/reaction to the Strevens talk yesterday, Michel has beaten me to it (to my delight I must say). Michel has not entered the blog world as of yet, so I have posted his review below. Enjoy.

I thought I'd type up my reaction to yesterday's talk, focusing unsurprisingly on the exchange Antigone and I had with the speaker in Q&A. Here is a handout I have used in my COI class to deep-six instance confirmation. That may be helpful to those who weren't familiar already with the ravens. In what follows I assume everybody is.

So Strevens avoids some of Hempel's problems by recasting all ravens are black as a causal claim. That's a steep price to get out of trouble. As John Norton never tires of pointing out, it's not much help to cast your account of confirmation in terms of concepts that are even more poorly understood such as explanation or causation. But since IBE (inference to the best explanation) is my favorite account of confirmation, that objection doesn't carry too much weight with me (though bringing in causation would presumably set Hempel spinning in his grave).

The main problem with the account, it seems to me, is that the sliver of confirmation that he's looking at is just not where the action is in real science. Antigone put it very nicely in Q&A. In real science, you're not interested in confirming that all ravens are black, but in confirming some causal story about how raven-ness causes black-ness. And that claim is only indirectly confirmed on Strevens' account. He conceded when I asked him that that's not instance confirmation but something like IBE. In that case, IBE does all the interesting work. I think, however, that he was too quick conceding that point and later in discussion he actually had a better line of defense. Suppose you have two competing causal accounts of how raven-ness causes blackness. As long as they both entail that all ravens are black, the observation of yet another black raven indirectly confirms both of these accounts equally well. Other confirmation mechanisms (Bayes, IBE), it seems, need to be brought to distinguish between them. But, as Strevens pointed out, that's not necessarily true. These causal accounts correspond in real science to high-level theories which entail other laws subject to instance confirmation besides the one about ravens. So even though it may be a draw as far as the ravens are concerned, one may beat the other on some other entailed law that is subject to instance confirmation. So instance confirmation plus the indirect confirmation of higher level theories can capture more than one would think just focusing on the ravens. Still, the typical situation in science is probably NOT the one where you have two perfectly good competing accounts, but one in which you barely have the beginnings of one reasonable account. In that situation, the real work is done by IBE.

It was a strange experience listening to this talk about instance confirmation. It was a bit like someone telling me, after I thought I'd decisively overcome it at the age of 15, that there's something to Catholicism after all. So, I was very focused on what Strevens would say about the problems that killed instance confirmation for me. To me the white shoe is really indicative of two much more general problems with Hempel's account. The first is that it shows that it's hopeless to have a purely syntactic criterion of confirmation (of course, the grue paradox drives that point home even more forcefully). Strevens avoids that one by bringing in causation. But there's another problem that Strevens doesn't address. He's still assuming that the empirical domain of the theory can be carved up into instances independently of the high-level theory that is being tested. The white shoe illustrates that problem very nicely. The mechanism through which the white shoe seems to confirm that all ravens are black is what at Pitt back in the days we called "content cutting." If you spell out the set of instances in advance, the white shoe confirms because it eliminates a potential falsifier (an instance that could have been the dreaded non-black raven turns out to be another innocuous non-raven). Strevens' response to that one was that the hypothesis subjected to instance confirmation does the individuating of the instances. I don't believe that. It's the high-level theory that does that. That seems obvious in physics (but maybe that's because I'm a diehard Kantian), so I asked Ken Waters whether he could give me an example from the biological sciences. He mentioned ecology, which sounds like a great example to me. For these reasons I remain skeptical about instance confirmation.

Respectfully submitted,

No comments: