Author: José de Jesús García Ruvalcaba
Date: 10:36:31 04/09/99
Go up one level in this thread
On April 09, 1999 at 02:41:01, blass uri wrote: > >On April 08, 1999 at 14:30:38, Chris Carson wrote: > >>On April 08, 1999 at 12:44:50, Steven Schwartz wrote: >> >>>I know many of us believe that certain chess programs >>>perform better against humans than they do against >>>computers (and vice versa), but does anyone have any >>>actual objective evidence to this effect? >>>- Steve (ICD/Your Move) >> >>Just my opinion (and $100 will get a cup of coffee). >> >>Chess is chess. No difference. > >I agree that all programs are strong against humans but I do not agree that >there is no difference. > >A program can be better against humans if it is using different evaluation >function when the opponent is in time trouble and does not allow him(her) to do >a draw by repetition even if the position is slightly better for the human. >It should not allow the opponent to simplify the position in this case. > >This idea is not productive against computers. >another productive idea is not to play openint that humans are good against >computers. > > >> Today's top programs >>on fast machines are top caliber players (several masters >>have posted as much here and GM's have written as much in >>Chess Life). Most (Soltis is an exception, see latest Chess Life) >>strong players (master and above) use computers as study partners. >> >>In my opinion, Humans learn and adapt faster (unless stuborn) >>than a machine. I have not seen any data that shows that a high rating >>in comp vs comp would not correspond to a high rating against humans >>(perhaps not exact, but close). AGEON and other events show this as >>well, computers have done very well in these events. Most GM's are >>scared to play a top program on a fast computer in public. >> >>Chess is 99% tactics. Computers excell at this. Both Human and Top >>programs have very extensive (good) opening book knowledge. Both have >>good endgame knowledge (not true of programs in the 80's or before, but >>very true today). > >I do not agree about good endgame knowledge. >I think that this is one of the main weaknesses of computers > >I saw cases when top programs did the mistake of going to a lost pawn endgames. > >They can do other mistakes in the endgame > >examples: Anand won Rebel10 in the last game in the match because of a >positional mistake of Rebel in the endgame. > >Fritz5.32 won Hiarcs7 in one of the games of the ssdf because of a bad game of >Hiarcs7 in the endgame. > >Uri I agree with the endgame part. I have seen many computers misplaying the endgame. I think that computers can be very strong in endgames with very reduced material, if they handle tablebases smartly. They can be strong in some complex queen endings (I remember a game Deep Thoght-Alex Fishbein where the computer won from a slightly inferior queen endgame). But in general, the endgame is the weakest part of current chess programs. Still, they are abismally better than the old programs in the endgame.
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.