Author: Russell Reagan
Date: 21:33:33 10/29/04
Go up one level in this thread
On October 29, 2004 at 17:36:40, Anthony Cozzie wrote: >Almost. The computer provides an opponent strong enough for anyone, as well as >good analysis. However, there is one thing that the computer is very bad at: >teaching humans how to play better chess. A computer can't say: I think white >should play on the queenside so as to shatter the black pawnstructure. It can't >say: Black shouldn't take this pawn because he will have to wade through a >tactical melee for the next 30 moves. Do you think we can accomplish this with current software tools and hardware? If not, do you think we will be able to in the future? It seems like we don't have such a thing now because a) no one has put enough effort into it, or b) it can't be done (well) under the current chess engine model (i.e. alpha-beta based tree searcher). Maybe a project like Symbolic will be a step in this direction. I think that if it is done, it will not be under the present chess engine model. Chess engines are removing chess knowledge in favor of extra search depth as hardware and algorithms improve. Ironically, humans are progressing the same way: less dogmatism, more pragmatism. So maybe what you describe could actually be a step in the wrong direction? Interesting ;) So where is the optimal point for best instructive value? Showing a 40 move combination will not be terribly helpful to us humans, but perhaps the examples you give aren't ideal either (slightly too dogmatic). Being human, there is a limit to our pragmatic side in searching game trees, but something more position specific than "white should play on the queenside" would probably be more instructive. For instance, top modern players recognize that certain "positional weaknesses" are not really weak at all. A bishop that is blocked in by its own pawns may be very useful since it can aide in protection. Remember, "positional advantages" are really just mental crutches to aide our poor tree searching ability. In some ways the present chess engine model provides a lot of instructional value. I have recognized that usually when I reach a position where I feel "stuck", when I analyze it after the game the computer's recommended moves are usually pawn moves. The computer could demonstrate the value of a blocked in bishop if you are willing to try a number of variations and see how the computer refutes them. So, in some ways we have what you describe, but in the form of user-tool instead of student-teacher. >Note that in terms of absolute chess strength the machines still have quite a >ways to go before they approach perfect play, but that they are still far >stronger than the average player (elo 400) so it doesn't matter much to most >people ;) Are you saying that the average player is 400 ELO below computers, or that the average player is 400 ELO? Neither statement sounds correct to me :)
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.