From: Ben Goertzel (ben@goertzel.org)
Date: Thu May 29 2003 - 05:22:40 MDT
> What I'd like to know is: what do the two Bayesianisms have
> to do with each other? What is the connection between
> an enthusiasm for conditional probabilities, and a belief that
> probabilities are subjective?
Well, the latter is a subset of the former ;-)
Going a little deeper.... In a mundane, data-analysis context, non-Bayesian
probabilists usually wind up making distributional assumptions before
approaching a dataset (assuming things are Gaussian, for example).
Bayesians, instead of making generic distributional assumptions, make
assumptions regarding prior distributions. (Thus Bayesianism involves
"nonparametric statistics") You've got your choice: make a generic
("objective", but maybe wrong) assumption or a particular (subjective)
assumption.
Bayesianism gives the opportunity for deeper nonprobabilistic principles to
be invoked -- e.g. Occam's razor a la algorithmic information theory, where
one assumes that computationally simpler entities have higher a priori
probability. In this sense Bayesianism is more "open" than conventional
uses of probability/statistics. Which is why Novamente's use of probability
theory is Bayesian in nature -- because I believe probability theory as used
in the mind is generally "open" in this way... making use of prior
assumptions derived from nonprobabilistic aspects of mind, rather than
making use of assumptions regarding distributional form. (Although in a
perception/action context, some parts of the human mind/brai probably make
use of distributional assumptions as well).
-- Ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT