Author: Christophe Theron
Date: 14:47:12 12/26/01
Go up one level in this thread
On December 26, 2001 at 14:51:10, Uri Blass wrote:
>On December 26, 2001 at 12:17:02, Christophe Theron wrote:
>
>>On December 26, 2001 at 08:10:41, Otello Gnaramori wrote:
>>
>>>On December 26, 2001 at 05:41:07, Andre Godat wrote:
>>>
>>>> The other day, on a different thread, somebody posted a message in which he
>>>>stated that, for all he knows, there may be a "retard" somewhere who can play
>>>>chess better than Kasparov. It is obvious to me that there's a gigantic
>>>>dividing line between a memory savant like "Rain Man", who can memorize every
>>>>game database in existence and still contribute nothing to the game and a genius
>>>>who can can add his own original touch and take the game to its highest level
>>>>ever.
>>>> If I pretend that Fritz is a real person--and I do it every day--there is no
>>>>question that he's an idiot-savant. In analysis mode, he just calculates until
>>>>I tell him to stop. He doesn't care about money or women or even chess itself;
>>>>the game is, for him, just a math problem he's trying to solve. But humans
>>>>don't look at a billion positions and say, "Well, this line is +1.41 and that
>>>>line is +1.51, so I'll go with the latter." It simply isn't possible, but that
>>>>may be what "Rain Man" would try to do.
>>>> Even in chess, Fritz can be amazingly idiotic. He has no intuition or
>>>>strategic sense. He has thousands of opening moves in his book, but he doesn't
>>>>really KNOW any openings in the sense of having an understanding of the typical
>>>>middlegames resulting from those openings. Yet his very ignorance is a source
>>>>of some remarkable novelties. He doesn't "know" that Black is supposed to
>>>>attack on the kingside in the Dutch Defense, so if he sees something on the
>>>>queenside that he likes, he'll play there.
>>>
>>>One of the things to come out of the effort to build chess programs is that
>>>games like chess require very little "intelligence".
>>
>>
>>
>>Yes and no.
>>
>>1) once some piece of "intelligence" has been identified and included into a
>>chess program in the form of an evaluation factor or an algorithm, we just stop
>>considering this as "intelligence". So "intelligence" is like horizon: you
>>believe you have made a step toward it, but realize it is still as far as
>>before. Our definition of intelligence is not only fuzzy, it changes constantly
>>as we are making progress in IT. Computer Chess is probably one of the main
>>contribution in our change of mind about what "intelligence" is.
>>
>>2) the kind of intelligence that is missing in today's programs (and I regret
>>very much that people starting to work in chess to not try to work on this) is
>>the ability to extract knowledge by itself and re-use it later. If chess
>>programs were doing this, you would not say that chess requires very little
>>intelligence. But you know that they don't, so you feel safe saying that chess
>>requires no intelligence. I still believe that chess requires intelligence for a
>>human player. In other words, without intelligence human players could not
>>become as strong as they are at chess.
>>
>>
>>1 and 2 is the reason why I believe that today's chess programs are a dead end.
>>They represent a high level of technology, but this technology leads to nothing
>>else than strong chess.
>>
>>
>>
>> Christophe
>
>
>I think that the end of chess programs is going to be only when it will be
>impossible to do significant software improvement.
>
>Significant software improvements are improvements that make the program at
>least 50 elo better on the hardware that is practically used by humans.
>
>Do you think that it is going to happen in the near future?
I don't think so, but I think at some point the only way to improve will be to
incorporate a way for the program to learn without the programmer, to remember
its experience and improve on it, and to adapt its play to its opponent.
The computational skills of computers are today probably close or even above the
best grandmasters in the world.
The difference between Kasparov and the best software is that Kasparov will not
do the same blunder twice (to put it in simple terms), and that Kasparov is able
to understand his opponent, his weaknesses, and play accordingly.
Programs do not adapt, and that's why they can be easy targets even for
relatively weak chess players.
But experience has shown that without this ability to adapt (because for example
of lack of knowledge of the chess programs), a grandmaster do not have a clear
advantage against today's top programs on today's hardware.
Today's top chess programs are, in this regard, only the half of a real chess
player.
>Do you think to try to do programs for another game that is more hard for
>computers like go when it happens?
I don't think I'll seriously try another game.
First because I do not want to repeat the steps I have taken to build Chess
Tiger and end up with just a wonderful and powerful calculator.
Second because if I have enough time (and money) I'll try to do some research in
real IA. And in this case Chess is a game that is more than sufficient to
experiment with (and with a totally different approach this time).
Christophe
This page took 0.02 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.