Computer Chess Club Archives


Search

Terms

Messages

Subject: Measuring Elo Drift

Author: Milton Zucker

Date: 06:33:42 09/26/98


    One of the things people like to do within any field of competition is
speculate
on how individuals of one era compare in skill with individuals of another era.
Relating this concept to chess, one might ask questions like "How much better
are average chess masters today than masters of 100 years ago?" or "What would
be the  likely outcome of a (hypothetical :-)) match between Paul Morphy and
Garry Kasparov?"
     I would love to see a long-term study done with the help of chess programs
which could objectively answer questions like these by measuring elo drifts in
the human population over time.  The study might take the following form:
   1.  Within one particular calender year, enter a variety of today's chess
programs in formally rated tournaments to obtain statistically significant elo
ratings for these programs.
   2.  Retire these programs for fixed extended periods of time (say 10 years),
then reenter the exact same programs using the same computer platforms and
reestablish new elo ratings, *without* considering the outcome of the previous
games.

     My assumption is that as humans learn progressively more about chess
theory, the skill level of the average player will tend to increase over time.
This increase could be accurately measured by the gradual *decrease* in elo
ratings of these chess programs over time as their competition gets tougher.
     Now I realize that this kind of study will probably never happen because of
the constraints placed on computers for entering sanctioned tournaments.  But I
wonder if folks here think that this kind of study could theoretically work?



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.