cole_haus comments on Potential downsides of using explicit probabilities

• Some re­lated things that come to mind:

• Challenges to Bayesian Con­fir­ma­tion The­ory out­lines some con­cep­tual po­ten­tial is­sues aris­ing from the use of ex­plicit prob­a­bil­ities in a Bayesian frame­work.

• Gerd Gigeren­zer likes to claim that “fast and fru­gal” heuris­tics of­ten just perform bet­ter than more for­mal, quan­ti­ta­tive mod­els. Th­ese claims can be linked to the bias-var­i­ance trade­off and ex­treme pri­ors.

• The op­ti­mizer’s curse can be gen­er­al­ized to the satis­ficer’s curse. This gen­er­al­iza­tion doesn’t ob­vi­ously seem to differ­en­tially af­fect ex­plicit prob­a­bil­ities though.

• Thanks for these links. I know a lit­tle about the satis­ficer’s curse, and share the view that “This gen­er­al­iza­tion doesn’t ob­vi­ously seem to differ­en­tially af­fect ex­plicit prob­a­bil­ities though.” Hope­fully I’ll have time to look into the other two things you men­tion at some point.

(My knee­jerk re­ac­tion to “”fast and fru­gal” heuris­tics of­ten just perform bet­ter than more for­mal, quan­ti­ta­tive mod­els” is that if it’s pre­dictable that a heuris­tic would re­sult in more ac­cu­rate an­swers, even if we imag­ine we could have un­limited time for com­pu­ta­tions or what­ever, then that fact, and ideally what­ever causes it, can just be in­cor­po­rated into the ex­plicit model. But that’s just a knee­jerk re­ac­tion. And in any case, if he’s just say­ing that in prac­tice heuris­tics are of­ten bet­ter, then I to­tally agree.)