Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Computerchess Fallacies Part 3: Chess Programming

Author: Dann Corbit

Date: 16:59:04 03/13/03

Go up one level in this thread


On March 13, 2003 at 19:35:30, Rolf Tueschen wrote:

>With this message I want more answers and explanations than simple opinions.
>For a longer time now we see how programmers try to make their products smarter.
>So it is a good idea to implement say a rejection of the Trojan Horse Sacrifice.
>But real chess players do know that this is just one single method to bust chess
>computers. Eduard Nemeth has actually found a new hobby with his discoveries of
>how stupid the engines still are. For instance he proves that Black does also
>reject the Trojan if Black could win immeadiately when White has played without
>the Ra1 so that Black could take Nb1 if the white Queen moves away to h5...
>
>But all such examples are given without a good description. Let me try a first
>explanation - although I am NOT an expert of computerchess programming.
>
>To me it seems that we have here a typical fallacy of computerchess. Chess is a
>game that is dominated by the reign of the concrete details of a situation. Now
>the point is that even in chess we follow certain rules (and we know) IF the
>concrete changes in the position are 'irrelevant'. But how do you want to
>program relevancies? All what you could do is implementing a defense - roughly,
>but you can't do it with flexibility! The program can't differentiate if
>something is suddenly important. Or let me assume, that this could well be done
>technically but then that would be so time consuming that the overall strength
>of the program would decrease.
>
>So, now a little reminiscence of something I often repeated already. Given the
>human super GM would see a real incentive in winning agains chess computer
>programs, they would study more about programming, then they would understand
>what we have now discussed and then they would make pudding out of the DEEP
>THINGS. Because flexibility (not a chance rotating) in a sense of smartness is
>impossible to implement. All the programmers can do is implementing a lot of
>suspense. But by definition that could be discovered after a few training
>sessions. And then  when all tricks have been detected, computerchess is solved!
>
>
>The actual show-events however do prove NOTHING because there was no incentive
>to find out for the players. They got their recompensation right at the
>beginning of the show.

Interesting gedankenexperiment.

However, there is no meat in the sandwitch.



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.