Epistemic Conservatism

Daniel and I have been talking a lot about conservatism lately (Daniel's been writing a book chapter on it), and we're considering writing a joint paper on the topic. Here's one of the things we've noticed that we'd like to write about.

A few importantly different kinds of epistemic conservatism seem to be floating around in the literature, not remarked upon nor clearly separated from one another, although it is far from obvious how they are related.

Some versions are about how to update your beliefs (e.g. Quineans, Bayesians), others about how to evaluate beliefs at a time. Let's call these 'update-evaluating conservatism' and 'state-evaluating conservatism' respectively. In the latter category, there are some versions which say that what matters is your belief state at an earlier time than the time which is being evaluated (e.g. Sklar), others which say that what matters is your belief state at that very time (e.g. Chisholm). Let's call these 'diachronic state-evaluating' and 'synchronic state-evaluating' conservatism respectively. Here are some examples from each category:

Update-evaluating (always diachronic):
The best updating strategy involves minimal change to your belief and credence structure.

Synchronic and state-evaluating:
The fact that you believe p at t1 gives a positive boost to the epistemic valuation of your belief in p at t1.

Diachronic and state-evaluating:
The fact that you believe p at t1 gives a positive boost to the epistemic valuation of your belief in p at t2.

Now, the interesting question: does believing one of these principles commit you to any or all of the others? In this paper by McGrath – one of the few I know of that talks about this stuff – it is assumed that the core of conservatism is an update-evaluating kind, but that this is equivalent in truth-value to a corresponding synchronic state-evaluating kind of conservatism.

But here's one reason to doubt things are that simple. Suppose I have a belief at t1 that is so epistemically bad that there is nothing to be said in its favour. Suppose I retain that belief at t2, with no new evidence, purely through inertia. One might wish to approve of the update qua update-evaluating conservative, but not wish to proffer any corresponding (diachronic or synchronic) state-evaluating approval of the belief at t2 – which, after all, is still held for really bad reasons.

Implicit Assumptions

I've just completed an implicit associations test run by a bunch of psychologists at Harvard, Virginia and Washington.

These tests work by assuming that you are faster to sort things into disjunctive categories when the disjuncts are things you (perhaps implicitly) associate with each other. So for instance, in one of their more disturbing tests, it turns out most people are much faster at sorting things into the categories 'African American or bad' vs 'European American or good' than into the categories 'African American or good' vs 'European American or bad', which is taken as evidence of positive associations with the one race and negative associations with the other.

The one I got had to do with TV and books. The test was fun (if a bit predictable), but the result that was given to me at the end seemed to me to be a misreporting of what they could reasonably claim to have discovered. I turned out to be a little bit faster at sorting into 'books or good' vs 'TV or bad' than into 'books or bad' vs 'TV or good', and they concluded that I have a 'slight preference' for books over TV. But 'preference' looks like a really bad word to use here. On a natural reading of what they're claiming, it means that I prefer to spend my time reading books than watching TV. Whereas the most their test has shown is that books have more positive connotations for me than TV. These two things obviously come apart – people could, for instance, enjoy TV much more whilst thinking of it as a guilty pleasure because watching TV is mind-rotting while reading books is worthy and highbrow.

This bothered me just enough to make me send an email ...

I think it would have bothered me a lot more if I'd been given a different test; if, for instance, I had turned out to have more positive associations with European Americans than African Americans and they'd described me as 'preferring' European Americans. This, at least on its natural reading, would imply far more than they could claim to have established.