Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Are Conventional Chess Engines Almost "Peaked Out"?

Author: Daniel Clausen

Date: 08:45:29 11/07/02

Go up one level in this thread


Disclaimer: The following is just my opinion - consider yourself warned :)


On November 07, 2002 at 11:12:16, Bob Durrett wrote:

[snip]

>But what about the BIG deficiencies of engines which everybody knows about
>but, so far, noone knows how to eliminate?  For example:  deficiencies in
>long-term strategy, planning, endgame, etc.

I see the situation where we are in CC atm like this:
We are in a 3-dimensional landscape with mountains, valleys, hills, plateaus,
etc. The ultimate goal is to be on the highest point in this scenery. (that is,
height is a measure of playing strength in this analogy) We are at one of these
mountains, but not on the top of it. What we're trying to do at the moment is to
reach the top of that mountain we're on at the moment, or in other words, reach
the local maximum. You can do that pretty well with changing something a bit and
test whether it pays off or not.

The only problem we might face is this: The mountain we're currently on is not
the highest one in the landscape. In this case the strategy "changing something
a bit and test whather it pays off or not" doesn't really work.

We (the computer-chess community) tried other mountains in the past, but maybe

(a) we just didn't try enough and therefore were only at the bottom of the
mountain (afterall the steepness of a mountain at the bottom is not necessarily
a measure for its max height)
(b) we just tried the wrong mountains


I think that the chances that we're not on the right mountain is rather big
(given the number of potential mountains - and no, I don't know how many
mountains there are :) But the real question probably is how much higher the top
of the right mountain would be compared to the top of the current one. If the
difference is not that big, it doesn't matter that much.. but who knows how big
it is...


Trying out different mountains takes a lot of time, and the "little changes and
test" approach is not suited to find it. Therefore I think that it will be a
non-commercial (at least at the beginning) chess-programmer, or a scientist-team
of some sort (like the Hsu-team (don't start to fight, please :)) who would find
a much higher mountain than the current one. Simply because for example the
Fritz team can't say "Hey, we know that Fritz9 is weak, but we're on the way to
this other mountain and Fritz13 will be much stronger than the competition!".

As I said at the beginning, that's just my opinion. :)

Sargon



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.