Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: I think TIGER is mega strong and a great improvement Don't PLAY MORE!

Author: James Swafford

Date: 18:42:42 10/03/02

Go up one level in this thread


On October 03, 2002 at 21:18:51, stuart taylor wrote:

>Anyway, how do the results of 1000 games PROVE anything either?
>Even if program A wins 950 and loses 50, the next 4000 games might have program
>A losing 3950 and winning only 50, final score A=100 B=4900.
>So you could also say that 1000 games also means nothing at all.
>


You're getting into number theory here.  I agree that what
you say is possible, but it's very unlikely.  As you are probably
aware, statistical inferences are given with a confidence interval.

In the case of the 4000 games, in which one program wins 3950, the
confidence interval would be very very close to 100% that the program
winning 3950 is stronger (assuming the match was fair).  Close enough
to call it 100% in my book.

--
James



>(yes, I realize that the margin of likely error is lower, but it's still far
>from a proof. It still could be very worthless. We only know from experience of
>number patterns, which is a little bit of "intellligence". But there could be
>much more "intelligence" applied too, if anyone is interested, and has any).
>
>And how much more so in humans which have good and bad times.
>Was Fischer really better than Spaasky when he won 12.5 to 8.5? Fischer may have
>been mentally stronger in that situation, and stastically it means nothing
>anyway. Maybe Fischer was dirt compared to Spaasky at chess. Out of 100 games it
>might have ended up with 80 to Spaasky and 20 to Fischer.
>S.Taylor



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.