Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Assessing GM Strength

Author: Kristo Miettinen

Date: 10:36:44 07/13/99

Go up one level in this thread


On July 13, 1999 at 12:58:31, Robert Hyatt wrote:

>On July 13, 1999 at 12:06:39, dfan wrote:
>
>>
>>On July 13, 1999 at 11:57:17, Kristo Miettinen wrote:
>>
>>>Does anyone have any ideas on how to use computers to objectively assess the
>>>strength of human GM's based on the games of past tournaments?
>>
>>There's an interesting chapter in "John Nunn's Chess Puzzle Book" in which
>>he used Fritz to blundercheck the games of a tournament from early this
>>century (maybe 1920's, I don't remember exactly).  He came to the conclusion
>>that play was overall much weaker in those days.
>>
>>Dan
>
>
>The problem is, he didn't check current tournaments.  I see lots of outright
>mistakes today as well, in GM games.  GMs aren't infallible.  The only question
>is, does the mistake change the outcome, or just the number of moves in the
>game?

In an incomplete information environment, would it be possible to analyze games
with an engine working at tournament speeds, and then use the engine at deep
thinks (or backward analysis from later perceived concequences) to compare move
choices when the GM and the time-constrained engine disagreed?

Style differences could be dealt with by averaging over a sufficiently broad
variety of GMs, as well as by using multiple engines.

The process need not evaluate any game performance perfectly. As long as it is
unbiased, the volume of games can be pumped up to gain statistical accuracy.

I had originally intended to tackle this problem (decades ago, before
GM-strength programs) by examining the advantage enjoyed by the player with the
white pieces as a function of player rating. The advantage must clearly be a
function of playing strength, since asymptotically for omniscient machines it
either reduces to zero or grows to certain victory, while for master-level
players it is in between. The problem is that the technique is VERY insensitive,
so the data volume required for reasonable accuracy would be phenomenal.

Sine cera,

-Kristo.



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.