Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 3000 Rated program & GM's

Author: KarinsDad

Date: 12:44:18 01/06/99

Go up one level in this thread


On January 06, 1999 at 13:40:19, Mika wrote:

[snip]

>
>I agree that the superGMs would learn from a program rated 3000, if such a thing
>were possible. However, isn't it also true that a chess programmer can only
>encode the knowledge which is available at the current time? Chess programmers
>can't program knowledge that doesn't exist. So, the program that is rated 3000
>achieves its rating because it is simply consistently good all the time, which
>is not something that human beings are good at. Accordingly, a program might
>actually achieve a tremendous rating (perhaps not 3000, however), without
>advancing chess knowledge much at all. The improvement for the superGM would be
>in identifying their own deficiencies, not in developing novel heuristics for
>better game play. In other words, the 3000 computer might just amount to the
>same old tired chess principles projected out over 25 or so ply. That will get
>you a great rating, but it makes the machine only worthly of imitation, not a
>real teacher.
>
>IMHO,
>
>Mika

One thing you are forgetting. Since a computer can log the reasons why a move is
inferior (regardless of the number of ply down), the superGM (or any player with
such a chess program) can analyze the logs and make determinations for
him/herself. The program can even use the logs in conjuncture with it's analysis
mode i.e. let's examine the top 20 lines that the computer was considering as
good (based on time spent in that line) until some other factor indicated that
the line was inferior. From this, a human player may be able to analyze the
position and better understand chess concepts for that position and improve
his/her game.

If Kasparov and Shirov were using such a program for analysis, you can bet money
that chess knowledge would be advanced.

As to programming knowledge that doesn't exist, almost all current chess
programs have limited chess knowledge as compared to humans (I'm not talking
tactical knowledge acquired via the search engine and the evaluator, but rather
concrete positional and strategic chess knowledge; programmer's don't beat me up
on this statement; I know that there is some of this knowledge programmed in).
This can be shown by:

a) No chess program can consistently beat the best players in the world in
standard time controls if the program is limited to 10 ply.

b) Kasparov, one of the best players in the world, usually looks only 3 to 5
moves (6 to 10 ply) deep on the majority of his positions.

If you limit the chess program to the effective event horizon of the best human
players (and yes, I realize that a lot of players look deeper in key
variations), then the best chess programs are not as effective as the best
humans. In other words, they do not possess nearly as much chess knowledge.

The real reason that chess programs can compete with the best players is that
they are:

1) relatively consistent (they don't get sick, tired, or distracted) and
2) they can search much deeper than humans in most positions, hence, they can
find good moves based on an ability to detect when a human makes an inferior
move due to reasons beyond the human's event horizon, but within the chess
program's event horizon.

KarinsDad



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.