Pur Autre Vie: And If You Don't Love Me Now

Thursday, August 14, 2014

[deleting post, going to have to try again when my thoughts are clearer]

Zed said...: I assume there is some Greco paper one is supposed to have read but that you haven't linked to?

I think it is almost obvious that the chain breaks at the last step. Belief isn't really a discrete variable; there is usually some uncertainty associated with it, and if the belief isn't associated with some imminent action (suppose you're just trying to learn about something to amuse yourself/satisfy your curiosity) you can keep accumulating "justifications"/data and adjusting uncertainties accordingly. The only step at which you need a binary variable rather than a distribution is the action step. Presumably the "right" way to decide on actions is to multiply the distribution by some sort of cost function, but the cost function isn't computable (because e.g. of your regress problem) and, in many cases, probably isn't well-defined -- we often don't know what we want. So this step is self-evidently messy. On the other hand, apart from pathologies like the problem of induction, the steps that allow you to update your beliefs as new justifications come in are clear and well-behaved.; 3:10 PM
James said...: Yeah, I don't know if there is a copy that would be legal for me to share, but on the other hand I'm no stranger to criminality and so I will email you a copy.; 5:21 PM
James said...: On second thought I am increasingly dissatisfied with this framework and I am going to delete it and try again.; 6:44 PM