Author: Hristo
Date: 13:59:52 10/27/98
Go up one level in this thread
On October 27, 1998 at 06:04:02, Mike Stoker wrote: >Hello everyone, > >I am a newcomer to the CCC, although I have been passively interested in >Computer Chess for a long time. Please excuse any statments in the following >which appear ignorant!! > >I consider that it is high time to move away from the brute force method of >chess playing - ie calculating 100000 positions per second. Whilst there is no >doubt that this have achieved significant results over the past few years, the >fact that top class humans can accomplish as good or better results looking at >just 2 or 3 positions per second would suggest that we might be missing >something. > >I believe this missing ingredient is planning. Now computers obviously lack >the features of a human which make our planning so efficient - i.e. logic, >pattern recognition, reasoning etc. All of these things are in essence a way of >generalising a wide variety of positions to extract just the relevant details. > >I believe that plans should be built up in much the same way as designs are - >1. Consider the ultimate goal of what we are trying to achieve >2. Break this goal down into a few subgoals >3. Take the FIRST subgoal and break this down into further subgoals >4. Repeat steps 2 and 3 until we can calculate the moves to accomplish the >first subgoal. > >Take for example the case of K and P vs K > The ultimate goal is mate. > Subgoals are "Queen the Pawn" and "Force Mate". > >My assertion therefore is that we never need to CALCULATE past the first >subgoal, when we can recognise that conditions are ripe for accomplishment of >the remaining larger goals. > >Vast savings in calculations can be obtained by working out general rules to >accomplish the lowest level goals. For example in "Queening a Pawm", a program >needs to work out the rules that if a Pawn is closer to the queening square than >a king, then it just needs to push the Pawn, otherwise, it needs to support the >pawn with the king, by taking the "Opposition" etc. > >As a challenge to all chess programmers, it would be interesting to see if >anyone can derive the rules required to promote a pawn without it being >captured, based solely on a knowledge of the legal game moves. This one of the >most simplistic chess positions, but obviously not a trivial task. However, if >we can accomplish this, I believe it will be a giant leap forward in the quest >for more intelligent computer programs. > >Mike. Mike, I totlay agree with the general conclusion about brute-force or "inteligent"(different) approach for solving the "chess". I've tried to start conversations about this in this forum, but most people don't want to think about this. They need results and they need them now! Be prepared to get *edjucated* ... Meanwhile I'm working on a chess engine, which uses a completely "alien" approach. It's tough when there is not enough time to be able to concentrate on the problems I'm facing. Designing a chess engine using the "old" style of building up alfa-beta trees, implementing some subjective(what is the material value of the pieces? How important is the material?) chess knowledge is relatively easier than starting with a fresh approach and getting it to work. So it will take time even for those who are interested to try something different. Will it work?! I don't know. I have some very interesting results ... but the program is still not playing chess. :((( What you suggest as possible approach might actually result in some very interesting chess algorithms ... the only problem that I see is the definition of "subgoals"! This is where things can go wrong or good! :) cheers Good luck. hristo
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.