[Theory] Probabilities over What?

Jun 1, 2018

Commentary

Free Access

Probabilities over What?

Commentary on Tourmen

Bickhard M.H.

….

A crucial assumption in Bayesian models is that the space of relevant hypotheses (or other choice alternatives) is available. This is simply an aspect of the fact that a probability distribution is distributed over something already given. Bayesian procedures, then, modify prior probability distributions over such spaces into posterior distributions over such spaces, based on current (relevant) data. They do not modify, nor generate, the spaces over which the probability distributions are distributed. Modifying the probability distribution over a space of hypotheses is not equivalent to Piagetian equilibration, for example producing new hypotheses, or new representations out of which new hypotheses could be constructed. In that sense, Bayesian models are models of confirmation and disconfirmation, not of the learning of new cognition or representation, or new hypotheses [as noted for learning theories in general by, for example, Fodor, 1975]. This is in contrast to Piaget, who attempted a non-foundationalist model of representation - one that was neither empiricist nor nativist. Bayes presupposes representation; Piaget did not.

…

Choe: I think the above part it key!

← Back to all articles         Quick Navigation:    Next:[ j ] – Prev:[ k ] – List:[ l ]