Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Natural Language Advice And Test Suites

Author: Bob Durrett

Date: 09:04:57 12/17/02

Go up one level in this thread


On December 13, 2002 at 11:14:44, Dana Turnmire wrote:

>It keeps coming up that positional tests aren't accurate because programs may
>find the correct solution for the wrong reasons.  Chessmaster 9000 has a unique
>feature (as far as I know) that explains *why* it recommends the moves it does.
>Would that solve the problem of knowing if it found the right answer for the
>wrong reasons?  If so would it be possible to add such a feature to the other
>programs?

The fundamental difficulty here is that chess engines do not "think" as humans
do.  Why was the PV selected?  The answer seems to be a composite of
explainations for: (1) why the search code produced the results that it produced
and (2) why the position evaluation code produced the results that it produced.
There is more:  The full explaination would have to explain the way the search
code utilized the results of the position evaluation code and, possibly, visa
versa.

Explaining "what the computer was thinking" cannot be done in human terms.  If
you want an explaination in terms of human thinking, you are going to be either
disappointed or misled.

Bob D.



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.