Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Deep Blue/IBM vs. Gary

Author: jonathan Baxter

Date: 23:10:28 04/22/98

Go up one level in this thread


On April 22, 1998 at 22:15:27, Doody Ungson wrote:

>5. On Deep Blue's rating- I wish most of you would read the latest books
>on Deep Blue vs. Kasparov  (i.e. one is written by a chess instructor
>Bruce Pandolfini). Gary Kasparov was definitely playing his hardest to
>WIN. As the previous TIME MAGAZINE article wrote about the match. He was
>playing to prove that MAN was still superior. The machine especially on
>all the drawn games found the proper combinations and moves to MATCH
>Gary's best skills. Even the best Grandmasters following and analyzing
>the game were not accurate in their analysis because Deep Blue at times
>was playing equal or beyond Gary's rating - definitely above anyone
>else.  Gary tried his hardest to win- throwing his best shots and Deep
>Blue with its enormous crunching power and of course a good chess
>program matched his chess skills move by move. Deep Blue was definitely
>playing at Gary's peak strength. The reason Gary lost was because of
>exhaustion. Period. That is the only explanation. Since Deep Blue never
>gets tired, it definitely has an edge over the world's TOP grandmasters-
>ANAND, KARPOV. Whoever plays it- 6 games or more.

DB would *not *continue to win if it was in open competitiion. To think
it will shows a lack
of understanding of the way computers play chess. DB simply uses brute
force to analyse every possible variation some 12-14 ply ahead (and much
deeper) in the end-game. But there are strategic considerattions in
chess which often dictate one side or the other is lost, even if it
might take 60 ply to do it. Although a lot of energy has been invested
by the chess programming community in coding evaluation functions that
capture such strategic knowledge, I think it is fair to say that there
is simply too much of it and it is often too "fuzzy" for anything more
than a small fraction to have been captured by today's evaluation
functions (DB's or anyone elses). This means that all programs must have
weaknesses in the way they play. A six game match is definitely not long
enough to find these weaknesses out, but in open competition they
certainly would be discovered. I can just imagine the conversation
between DB's programmers and IBM management after the win against
Kasparov:

Management: "If we let this thing loose will it crush everyone?"

DB team: "At first, yes, but given enough time to analyse DB's games,
the top GM's will start to find weaknesses and beat it:"

Management: "Oh well, scrap it then. We can't afford to jeopardize all
the good publicity."

DB Team: <stunned silence>

On another note, I think many GMs, Kasparov included, get psyched out by
the tactical ability of computers. They seem not to have a good
appreciation of exactly how simple the program is. Instead they judge
its play in their own terms, and get really freaked out by a player that
in some kinds of position has much greater strength that themselves,
especially the top players who are used to winning because they  can see
deeper than the rest. I think if the GMs learnt more about how the
programs worked, they would be more comfortable playing them.

Jonathan Baxter



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.