Author: Uri Blass
Date: 07:50:14 02/01/02
Go up one level in this thread
On February 01, 2002 at 10:36:54, Mike Hood wrote: >On February 01, 2002 at 05:49:48, Kurt Utzinger wrote: > >[White "Hiarcs 7.32"] >[Black "Chess Tiger 14.0"] > >[D] 8/8/K1kn2R1/8/P6r/8/8/8 b - - 0 1 > >121. Ka6?? >121... Rxa4# > >>Such error happens from time to time in computer games and are the exception of >>the rule. And it is clear: nobody will ever be able to reproduce the situation. >>Kurt > >I agree. > >It's obviously a problem of selectivity, somewhere along the line. The very >first chess programs relied on brute force to analyse all possible >continuations, but today's chess programs have only become stronger by deciding >what NOT to analyse. All the same, in the above position I have to naively ask, >"Shouldn't a program at least do a 2-ply exhaustive search from the root?" (Or >is it 3-ply, to recognize that there are no legal replies to Rxa4?) Missing a >forced mate in 4 is something that might happen to me on a bad day, but this >example is a blunder that only a beginner would make. > >It seems to me that this is a special position. Hiarcs is obviously fighting to >avoid a draw by the 50-move rule. (See my original post: a4 in move 111 is >already the 42nd move according to the rule). > >But just think of the implications if this hadn't happened in an obscure >engine-engine game. Just imagine if Black's name were not "Chess Tiger" but >"Kramnik". Instead of this small discussion (which will probably die out after a >few replies) it would be a matter of heated argument for months. It wouldn't >just reflect on Mark. Every computer chess programmer would be up against the >wall. The boulevard press would be less kind than we are in here. Just imagine >the headline: "After 40 years of research computer chess still sucks" > >I'm not a journalist, so I won't speak those words myself :) All I can say is >that this is a strange curiosity, and I hope I won't see it again. I suspect that only hiarcs can blunder like this. This program simply have bugs. I already found strange tactcial errors with hiarcs when it did not happen with other programs. There are cases when I could reproduce the errors if I gave hiarcs to play the same game again but could not reproduce the errors when I gave Hiarcs only to play the same position again. I believe that it has some bug in retaining the hash tables Uri
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.