Author: Matthew White
Date: 16:19:18 12/03/02
Go up one level in this thread
On December 02, 2002 at 19:58:49, Bob Durrett wrote: >On December 02, 2002 at 19:09:36, Matthew White wrote: > >>It seems to me that one of the reasons that most chess engines have difficulty >>in blocked center/closed center positions is their lack of planning ability. If >>they had a specific plan in mind and stuck to it, rather than trying to stumble >>onto the winning line, then they would probably play much more effectively in >>these positions. >> >>Matt > >But what about the process which produces good "plans"? How would that process >be implemented in a chess engine? That is something that a better player could probably answer more effectively than I, but I think that one possibility is to make an engine "goal oriented," and then give the engine the necessary "knowledge" in order to carry out those goals. I read in a book once that any plan is better than no plan at all, so rather than lumping all knowledge into a composite number, it might be useful to evaluate aspects of a position and use those aspects to determine sort order. Another possibility might be to encode a "plan" in the opening book. When a player studies an opening, it is not just to learn the moves that make up that opening, but also to learn the plan behind the opening. I have watched several computer chess games where the book told the engine to play a gambit opening, and the program did not understand how to exploit the gambit, so it lost. If the opening was encoded with information which slightly altered the evaluation parameters. For example: in a gambit opening or in an opening with an open center, failure to develop pieces should be penalized greatly and king-safety should be rewarded, as should having pawns in the center. Conversely, in an opening with a blocked center, or a mobile center, rapid piece development is not nearly as important, but pawn trades in the center should be discouraged until development is complete and the king is out of harm's way. I think that meta-information associated with each opening is probably the easiest short-term way to help a computer plan. However, if the computer could be "taught" to recognize important features in a position, then it would be much simpler to formulate a plan. Matt > >First one must "stumble upon" a viable plan. Only then can one "stick to it." > >The hardcopy book, "Judgment and Planning in Chess" by Dr. Max Euwe, first >copyright 1953, ISBN 0679143254, describes a process whereby good plans might be >produced. The suggested process involves "judgment" as a part of that process. >One might argue that chess engines already do some of that "judgment," although >that's not clear to me. > >That book addresses only human chess. I'm not sure Euwe's suggested process >would be optimal for use in chess engines, however. By their very nature, >modern PCs are inherently sequential, unlike the human brain. Maybe there are >other plan generation methods better suited for chess engines. > >Bob D.
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.