Learning necessary for chess champion?

Mike Gherrity

unread,

May 16, 1997, 3:00:00 AM5/16/97

to

[ Article crossposted from comp.ai ]
[ Author was Mike Gherrity ]
[ Posted on 16 May 1997 08:07:12 GMT ]

Is it possible that at the level of chess play between Kasparov and Deep
Blue, the one who best learns their opponent's weaknesses wins the match?
In the February, 1996 match, Kasparov lost the first game but went on to
win the match by taking advantage of weaknesses in Deep Blue that he
observed in the early games. The IBM commentary on game 6 of this first
match (http://www.chess.ibm.park.org/deep/blue/archive.html) says

"Clearly, Garry had learned a great deal about his
digital opponent, which he brought to bear in this
masterful display."

Rather than add a learning capability to Deep Blue to defend against such
human abilities, the Deep Blue team decided to counter this threat by relying
on human observers to do the learning, and then tweaking the computer between
games. From CNN (http://www.cnn.com/WORLD/9705/03/chess.rematch/index.html)

"We also made our program more flexible, so in
between games we can change our strategy to counter
what Garry will do," says Chung-Jen Tan, the
manager for Team Blue.

So it appears that the Deep Blue team has concluded that championship level
play requires learning the opponents strategy, and then defending against it.

Is it then true that the caliber of chess play is so high that one cannot
become champion without the ability to learn and respond to the opponents
play within the length of a match? If this is true, then Deep Blue could
never be champion by itself, since it doesn't learn by itself. In order to
answer these questions, it would seem necessary to not allow any changes
to Deep Blue during the entire match, not just during each single game.

mike

--
Michael Gherrity | gher...@nosc.mil
NCCOSC RDTE DIV 4221 | http://papaya.nosc.mil/~gherrity/
53140 Gatchell Road | (619) 553-5322 (office)
San Diego, CA 92152-7463 | 553-4732 (fax)

Scott Pallack

unread,

May 16, 1997, 3:00:00 AM5/16/97

to

On 16 May 1997 08:08:23 GMT, gher...@io.nosc.mil (Mike Gherrity)
wrote:

>Is it then true that the caliber of chess play is so high that one cannot
>become champion without the ability to learn and respond to the opponents
>play within the length of a match? If this is true, then Deep Blue could
>never be champion by itself, since it doesn't learn by itself. In order to
>answer these questions, it would seem necessary to not allow any changes
>to Deep Blue during the entire match, not just during each single game.
>

By the same token, the human player should not be allowed to consult
with his seconds, do research, or use any tools including a computer
between games.
Scott Pallack
sky...@wizard.com
banglIj yItlhabchoHmoH; cheghbe'chugh yIwam 'ej yIHoH.

Tom C. Kerrigan

unread,

May 17, 1997, 3:00:00 AM5/17/97

to

Mike Gherrity (gher...@io.nosc.mil) wrote:

> Is it possible that at the level of chess play between Kasparov and Deep
> Blue, the one who best learns their opponent's weaknesses wins the match?

Eventually a point will be reached where a machine will search so deep
that a human, no matter how well prepared, will lose every game due to
(possibly "minor") mistakes. I don't think we're at this point yet, but DB
proved that we're making progress. Garry couldn't win games, even with
supposed positional advantages. It's only a short leap to DB getting the
advantages and winning.

Cheers,
Tom