From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Tue Nov 08 2005 - 15:35:15 MST
Michael Vassar wrote:
> GPA is the sum of a modest number of correlated variables. That could
> be approximately Gaussian or could not be depending on the number of
> variables, the range of values they can take, and the strength of the
> correlation. For instance, given a single grade, or given two possible
> values (not implausible in many classes) and a .5 correlation between
> grades in different classes, GPA would be uniformly distributed. I
> wonder what the actual distributions are, but that's not a SL4
> question.
The whole point of picking a prior is to try to score as well as
possible relative to the actual distribution. In real life, the
distribution for this sort of thing is usually gaussian.
> Using good H&B examples could be SL4 relevant however,
> because it's important that people learn normative reasoning and H&B
> properly. Regression, like entropy and evolution is not (as many though
> surely not most posters here think) an active force but rather a
> statistical tendency that arises from logic in many different
> circumstances.
I'm well aware that repeating to myself, "I will not regress to the
mean, I will not regress to the mean" is a meaningless statement.
Nonetheless I think it - or to be more precise, I think to myself, "I
have to do well enough that future observers regard my childhood SAT
scores as regressing to the mean relative to my actual intelligence, not
the other way around."
-- Eliezer S. Yudkowsky http://intelligence.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:23:19 MST