Computer Chess Club Archives




Subject: Re: chess and neural networks

Author: Omid David Tabibi

Date: 17:57:40 07/01/03

Go up one level in this thread

On July 01, 2003 at 18:34:05, Christophe Theron wrote:

>On July 01, 2003 at 15:31:55, Tom Kerrigan wrote:
>>On July 01, 2003 at 14:29:25, Ricardo Gibert wrote:
>>>On July 01, 2003 at 14:21:12, Tom Kerrigan wrote:
>>>>On July 01, 2003 at 13:32:19, Ralph Stoesser wrote:
>>>>>Hello *,
>>>>>Why no top engine uses neural networks for positional evaluation in non-tactical
>>>>>situations? Are there interesting publications about neural networks and chess
>>>>Neural networks are for analyzing things that are
>>>>"fuzzy"--voice/image/handwriting recognition, etc. Chess is a very exacting
>>>>game. (It makes a big difference if your rook is on d1 vs. e1.) I doubt neural
>>>>networks will ever be useful for chess.
>>>Hmmm..but Kasparov uses the neural network within his brain to play chess
>>>doesn't he? Are you contending he does not use his brain to play chess? Divine
>>>Perhaps you meant to say something like, "I doubt [the current interpretations
>>>of] neural networks will ever be useful for chess."
>>Kasparov has billions of neurons with 10-50 times as many interconnections. A PC
>>has 50 million transistors and plays chess approximately as well. It's obvious
>>to me that the human brain is not as well suited to playing chess as a computer
>>(esp. considering that most human brains are much worse at chess than a 386).
>As far as I know our neural networks are quite good at mathematics, and that's
>not a fuzzy discipline at all. That's to answer your previous message.
>Considering the current state of computer chess it is very possible that the
>only way to finally demonstrate a real superiority over the world champion (or
>the five best human chess players on the planet) will come from something that
>will have something to do with neural networks.

Neural networks might work fine in endgame evaluations, which is the weakest
spot of current programs.

>I see the current approach at best to be able to demonstrate that it is on par
>with the best human player. I don't know what you think, but I consider this to
>be extremely humiliating for computer science, given the efforts and resources
>that have been spent on it.

I pretty much agree with this. In a recent seminar lecture, I called computer
chess "the most thoroughly researched area within compuer science", which is
true as CC was born even before computers were created (Turing's
simulations...). And still computers are not better than (the best) humans.

>A well trained brain can be world chess champion, a well trained 386 cannot.
>Unless you are extremely optimistic about computer science. :)
>Biological neural networks have proven to be so different and nevertheless so
>close in computing ability to computers that it is obvious that current
>computers have much to learn from them.

Consider the human neural system as an infinite number of parallel (very slow)
processors. It is not good in pure computation, due to the very slow processor,
but extremely well in cognition, due to the massive parallelism!

>So eventually what counts is not the superiority of one approach (Von Neuman
>architecture) over the other (neural networks), what counts is that both have
>unique skills and it there is probably much to gain from the one that is
>currently almost not used at all.
>That being said I know how to do well at chess with a Von Neuman architecture,
>but I have not idea how to do it with neural networks... :)
>    Christophe

This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.