Thanks, I mostly agree with you.
Sorites
[...]
I think it's noticeably less controversial if you imagine this representing a population rather than an individual. — Srap Tasmaner
Yes, here the interpretation of the model is clear and the model may be a good fit (to that interpretation). Or is it? If the middle section is where people are genuinely uncertain about their choice, the actual distribution of answers may break down into random noise.
As above, we could graph her uncertainty about her answer instead, and we'd expect a normal distribution, wouldn't we? — Srap Tasmaner
Does the statistics (if there is in fact a consistent statistics) of individual choice represent one's degree of confidence/uncertainty? If we define it behaviorally, as you say later, then it does so, by definition. But then reporting observed behavior
as the degree of uncertainty is merely tautological: despite the use of an ostensibly psychological term, this does not shed any light on our inner world. But if the assertion is that the graph represents phenomenal uncertainty (which is, after all, the central thesis of epistemic/Bayesian probability interpretation), flattening out a mess of thoughts, feelings and subconscious processes into one number, then it is much less certain (as it were).
I guess my uneasiness goes back to bridging the gap between probability and a single case. Unlike mathematical probability, real-world probability is
always single-case (we don't deal with infinite ensembles!) Defining probability as a frequency is unsound for that reason, while defining it epistemically threatens to oversimplify a complex psychological phenomenon. The moral, I think, is to treat Bayesian models of behavior with caution. When you blow up a detail of a curve and ask about its physical meaning, always keep in mind the possibility that it may not have one: it may just be a modeling artefact.
One thing this curve could represent is an individual striving for consistency under conditions of irreducible uncertainty. — Srap Tasmaner
Yes, that's the Dutch Book argument, and I do find it rather compelling. (And notice how you have switched from behavior to phenomenology, after all!) Don't get me wrong, I like Bayesianism. I like it for its mathematical elegance, consistency, and (when used correctly), instrumental usefulness. When it comes to modeling uncertain beliefs and decisions, it is probably the best game in town.
I'm just interested in how partial belief works, and I keep finding reasons to expect individuals and populations to be homologous. — Srap Tasmaner
Well, one reason for that may just be that the curve, assuming it is the error function, is closely related to the normal distribution, which is ubiquitous whenever you deal with (or assume) random variables.
There may also be an evo-psych story here: the reason individual is homologous with population is because cognition is an evolved feature, and evolution works on populations. The behavioral strategies that statistically increased the population fitness were the ones that were fixed in our genes. This may also serve to explain away the problem of induction: the reason we intuitively trust induction is that our environment does have certain regularities (how could it not? we wouldn't be here if it didn't), and we have adapted to recognize and exploit those regularities.