Author: Bruce Moreland
Date: 00:19:17 10/31/98
Go up one level in this thread
On October 30, 1998 at 06:07:45, Graham Laight wrote: > >On October 29, 1998 at 22:31:51, Bruce Moreland wrote: > >>There are some products where new versions are demonstrably better than old >>ones, or at least no worse, in every respect. >> >>A chess program is not an example of such a product. Chess programs are like >>cats, they are not particularly predictable, and they odds that they will react >>like you want them to are by no means high. You can make a very minor change to >>a chess program and suddenly something that worked well no longer works as well. > >I've seen other programmers complain about this problem in the past as well. FYI I am not complaining about it, it's just an aspect of a chess program. If you make a minor change to the eval or to the search, you get different moves, and sometimes a tactical shot is found a ply earlier or later. >This is one of the key reasons why I think it would be useful to think about >systematic ways of managing knowledge. I wish the CCC members were more open to >discussing this. It is an area in AI (artificial intelligence) in which progress >is being made. I'm open to discussing anything. I don't think the ideas that you've expressed have been particularly new or useful. Anybody who thinks about the subject for the first time thinks about trying to mimic human thought processes, and anybody who has read a tactical combinations book that is broken down into chapters on various "themes" has considered trying to identify themes in positions and make moves based upon what is thematic. Also, I think some of your specific direction has been wrong in the past. You've spoken about trying to identify elements of a position with enough confidence that you could generalize and apply a precomputed solution. I think this is way too rigid, the amount of effort it would take to produce something that could identify themes and confidently select a thematic move would be so great that it would probably be easier to write something that would understand the position so well that it could create its own plan on the fly. Sorry to be so brutal, but I don't want to get tarred with the close-minded brush. It is not that I think you are wrong because you're advocating the wrong paradigm, I simply think that what you have suggested in the past is wrong. >If one was programming a business application, then making the "rules" >accessible to other programmers would be of high importance. I bet that most >chess programmers cannot predict accurately the effects of the knowledge in >their evaluation functions in various different positions - and especially the >effects of lots of different types of knowledge working together! > >One day, I hope that someone will build a chess engine in which enthusiasts can >put the knowledge in for themselves - and be able to predict how it will effect >the evaluation in various types of position. > >Of course, programming chess this way might impact performance - the "God" of >most programmers. I think you're trivializing the attitudes of people who build chess programs. It may be true that some of the programmers have weird goals, but as someone who has spent a lot of effort on a program, I can appreciate what must have gone into the other programs. And I can assure you that the successful programmers are most concerned with program strength, and they are willing to do anything, even try new things or admit they are wrong about some cherished idea, in order to get it. bruce >> Of course, if you want to say that the change was good, you have to be able to >>find something that the new version does better. >> >>A different version of a chess program, especially a very different version, >>will find some shots faster than the old one, and some slower. >> >>bruce
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.