Author: Mark Young
Date: 14:58:19 02/27/99
Go up one level in this thread
On February 27, 1999 at 16:19:46, Leon Stancliff wrote: I find the data interesting, but the ratings data from ICC or any online chess server will be skewed. I do not know if you can draw any kind of conclusions from them. The are just to many sub pools of rating in ICC. 1. You have GMs who will play anyone. 2. You have GMs who only play humans 3. You have GMs who only play humans or computers they can beat 4. You have GMs who only play other GM’s or IM at the top of the ICC ratings list. Then you have a similar sub pools of ratings for computer accounts. This makes comparing chess strength by chess server ratings almost impossible, between GMs and computers, or even GMs to GMs or computers to computers. For Chess server ratings to have any meaning at all, they could not allow players to choose whom they will and will not play. This would be very unpopular. Mark Young > While scrounging around in the research area of ICC, I found what I think is >very interesting data. I have been interested for many years in the human vs >computer chess ratings. It is but a matter of time before any person who desires >will be able to own a computer chess program that plays at lower grandmaster >level, even at time limits of 40/2. > > Why does this seem to be such a hot topic? Automobiles make faster time than >the top human runners. Machines are also superior to humans in other areas. The >reason this is so intriguing is that we are here considering human thought >processes, and also that the competition is very close. > > Here is the data I found. I first looked at the Fide ratings of the >grandmasters on ICC who were in the 2500 to 2600 range. There were about fifty >of these persons. I compared their ICC Blitz ratings with their Fide ratings. >The average Fide rating was 2532. The average ICC Blitz rating of these persons >was 2686, or 154 points higher than the Fide rating. > > The average Blitz rating of the seven highest computer programs was 2929. Thus >the computer programs were clearly superior to the grandmasters in the 2500-2600 >range at Blitz times. > > Next comes the shocker! Although very few grandmasters are willing to play at >ICC standard times, there were ten who had played enough games to be >significant. These ten grandmasters had ICC standard ratings of 2456. The same >ten had Fide ratings of 2541. Note that the standard ICC rating is 85 points >lower than the Fide rating. Obviously these standard games were played at much >shorter time limits than 40/2. Still, these figures would cause us to be >suspicious that top computers which are rated above 2500 are more than capable >of competing on even terms with grandmasters rated under 2600 Fide. > > Come on grandmasters. Let us see some more standard games at 30 0 against the >top computer programs! Losses do not take away from your prestige. Are you game?
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.