Author: Russell Reagan
Date: 23:28:54 05/13/02
Go up one level in this thread
Your idea sounds very interesting. It's ideas like these that will help chess programs get stronger in the future, because we all need to explore new territory if we want to make any more drastic improvements. As for whether your idea would work or not, that's harder to say. There are a lot of things that I've thought, and others have thought of, that sound absolutely great, but when you start thinking about how to implement them and make them work in reality, it gets more complicated and a lot of times you don't get the kind of results you expected. A perfect example is using neural nets to "train" a chess program to learn. Last time I heard, there were some neural net programs playing decent chess, but not even close to master level. Somewhere around 1500-1600 I believe. I think your idea would work at first, just like neural nets would help a program to learn at first, but I'm not sure if it would boost the program beyond the best engine's that don't use that technique. It's worth a try though I think. I imagine it going something like this. Your program starts off with no evaluation function. It has no knowledge. It begins playing opponents and plays horribly. At first it just guesses for evaluation scores, more or less randomly. It will lose, but I would suspect that it would improve quickly to a decent level of play. It probably wouldn't take many days for a program on ICC playing 24 hours a day to get some losses to look at, and at the very least it would develop a sort of "material" evalutaion quickly, and it would have an idea that knights and bishops are worth more than pawns, and rooks are worth more than those, and the queen the most valuable, etc. And it would be able to beat beginners by simply outplaying them tactically. So that seems like it would work at least at first. I'm not sure how much tuning could occur once some major things like that are discovered by the program though. But, we'll never know until someone gives it a try. All very interesting. Russell
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.