Author: Robert Hyatt
Date: 13:47:57 04/11/02
Go up one level in this thread
On April 11, 2002 at 15:59:18, Uri Blass wrote: > >The fact that they change their mind to Kh1 after enough time is a proof that >they know enough. >I believe that even crafty knows enough to prefer Kh1 after enough time. As the author, I can absolutely guarantee you it doesn't "know enough". If you want, I will be happy to post a simple position where most programs will play the right move for the wrong reason. And then find another similar position where the wrong move is played for that same wrong reason. This is about search. Not about knowledge... > >knowing enough does not mean knowing that Qe3 is a draw but knowing that the >endgame after Kh1 is better than the position after Qe3. > >knowledge in evaluation does not mean knowing the final result of every position >but knowing better to evaluate which option is better. The only straw in the ointment is that "better" here is only discernable after a horrendously deep search to follow all the checks. If you are sure the checks are tolerable, Kf1 is fine to play. Move one pawn in the indicated position and Kf1 is perfectly ok. So there is no "positional reason" to prefer one over the other. There is definitely a very deep tactical reason, however... > >The knoweldge in the evaluation that programs have is enough to evaluate that >the position many plies after Qe3 is relatively better for black than the >position many plies after Kh1. > >Uri And you trust those evaluations? And you trust making a move because the static evaluation says one thing even though the search doesn't see a thing about what is _really_ going on? It should not be horribly difficult to create a position where a similar thing is going on, but where playing Qe3 (for black) actually loses to a deep tactical threat. But without seeing that threat, the program assumes the wrong thing and makes the wrong move. I don't like serendipitous moves...
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.