Author: James Robertson
Date: 18:14:03 12/27/99
Go up one level in this thread
On December 27, 1999 at 16:21:56, John Warfield wrote: >On December 27, 1999 at 16:07:25, Leon Stancliff wrote: > >> I have a win expectancy chart from USCF which gives the expected percentage of >>wins at various point differentials. The difference in ratings which you have >>supplied is 105 points. The win expectancy for the higher player in this case is >>64.6%. 64.6% of six games is 3.8 games. We would expect Rebel Century to win the >>entire match with either 3 1/2 to 2 1/2 or else 4.0 to 2.0, with more likelihood >>of 3 1/2 to 2 1/2. As well all know, in a small number of games the variation >>from the expected percentage can vary widely. >> The results from the games Rebel Century has already played at 40/2 against >>grandmasters and international masters shows just over 2500 for the grandmasters >>while playing on the AMD 600 and about 2480 when playing on slower hardware >>against the international masters. >> I for one sincerely believe that by the time the Russek match is completed, we >>will have sufficient evidence to establish a rating for Rebel Century which will >>be correct within 25 Elo rating points. >> I am also keenly interested in seeing what the new alliance of Rebel and Chess >>Tiger will produce. I really think Rebel is playing just over 2500 Elo on the >>AMD 600 Mhz. I think those who have been downplaying the ratings of the most >>powerful programs will be swallowing their adam's apple when Rebel-Tiger takes >>the stage! > > > I don't think they will be swallowing their adam's apple, for they will >always have some excuse tucked away. For instance 1. The humans didn't play >anti-computer chess So the ratings are bogus 2. Human had a bad cold the >morning he got up to play the match 3. There was to much noise ect, ect, People >will come up with all kinds of things in order to avoid being wrong. So I >wouldn't look forward to your results. Bottom line there will never be any >satisfactory way of establishing a grandmaster rating for computers if people >don't want them to be grandmasters. Another way is to say that the computer must >Earn a 2600 elo to be a real grandmaster, Knowing full well that this is >impposible since a computer will never be allowed to compete with humans under >normal tournament conditions. There is a difference between earning the GM title and being of GM strength. A computer will have a terrible time earning the title as comps aren't allowed in nearly all tournaments. Therefore a comp will probably never "be a GM". As for being of GM strength, that problem is more easily solved. I personnally think there will be computers of GM strength within a few years. When a computer can maintain a rating of > 2500 for more than a few games, I'll be convinced. James
This page took 0.01 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.