Author: Bob Durrett
Date: 07:47:29 11/04/03
A Few Thoughts about Pruning and Evaluation: Recently, I asked why Crafty was slower than the top commercial programs and Bob Hyatt responded by saying that the top programs are using some very sophisticated pruning techniques. Bob’s comment was very thought-provoking for me. Bob seemed to be pointing to the direction for future advances in chess engine program design. I have often heard it said that the difference between the way a human thinks and the way a chess engine thinks is that the human only considers a few moves in each position whereas the chess engine “has to” look at everything. Obviously, a chess engine which is doing pruning is not looking at everything. When a human selects only a few candidate moves, the remaining moves have been “pruned,” in effect. However, this does not imply that the human consciously or subconsciously considers each and every move on the board. We do not know what the subconscious mind does, of course, but it appears that the subconscious mind does not consider and reject each and every move except for the chosen candidates. Indeed, it is not clear that the subconscious mind “considers” anything. Perhaps what happens in the human mind is that candidates are “somehow” found not by a rejection process but instead by some other means. In other words, candidates are NOT “all moves remaining after other moves have been rejected.” Instead, they are, I suspect, selected by some means which does not consider the merits or demerits of rejected moves. It seems much more likely that humans select their candidate moves based on a subconscious intuitive evaluation of the position. The person becomes consciously aware of this when candidate moves seem to jump out of the chessboard. In reality, they are surfacing to the conscious from the subconscious. Conscious activity does steer the subconscious mind however. For example, if one consciously and deliberately examines each piece to get its true evaluation in the position (considering all rule exceptions), then the subconscious “works on” the information found in real time. Similarly, general positional considerations, such as (a) weakened king position, (b) tempo advantage, (c) spatial advantage, (d) restricted range of action of a piece, (e) vulnerability of pieces, (f) coordination and dislocation of pieces, and (g) weakness and strength of the pawn formation can be deliberately considered by the conscious mind, essentially steering the subconscious mind to think about them too. [Elaboration on these position evaluation elements may be found in “Chess Strategy” by Gufeld & Kalienchenko, ISBN 0 7134 8775 5.] One might think of the subconscious mind as being a very complex chess engine. It is dumb but very good at what it does. It can be steered, and to some extent controlled, by the conscious mind. One thing is clear. “Intuitive” evaluation of a position is not done by methodically evaluating each move that exists in that position. Rejection of moves, to arrive at a list of candidate moves, is NOT done. Instead, the position evaluation process is entirely different. To what extent can/do chess engines mimic the human mental processes outlined above? Unfortunately, the methods used by the top programs must be kept secret for profit reasons. A programmer(s) of a top engine cannot afford to give his/her competition anything. The idea of trying to mimic the human mental processes is not new. Many will categorically say that trying to do so must be a mistake. Nevertheless, it seems likely that there are lessons to be learned and new ideas to be found from study of the human mental processes. The idea of a conscious steering the subconscious is interesting. A chess engine could be programmed to have the computer equivalents of the conscious and subconscious, with the conscious steering the subconscious. The subconscious might be where most of the actual work is done, whereas the conscious might be devoted to consideration of “overview” strategic considerations. This is just an example. Other ideas abound. To summarize, there are new ideas to be found from study of human mental processes. [The science of the human mind is still in its infancy.] Pruning ideas include methods based on selection rather than rejection. Evaluation ideas may include "steering" based on general strategic considerations. Indeed, this may be a way around the "planning roadblock." A block of software could be set up to consider strategy and long-range considerations, calling upon the subconscious for assistance in evaluation when needed, and thus serve as a silicon equivalent of the conscious part of the human mind. Enough rambling. 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.