DB Tweaking Between Games
Mike Gherrity
I notice that both IBM and the press are claiming that in the DB/GK "Man
versus Machine" match, the machine has won. Here is what IBM says at
http://www.chess.ibm.com/home/may11/story_1.html
> It was the first time a current world champion has lost a match
> to a computer opponent under tournament conditions.
And from the May 13th London Times:
> Defeated Kasparov accuses machine of cheating
>
> FROM JAMES BONE IN NEW YORK
>
> GARRY KASPAROV was routed last night in the final decisive game
> of his Man-v-Machine chess challenge against the IBM
> super-computer Deep Blue.
However Kasparov did not lose to Deep Blue, he lost to the Deep Blue team --
including six people. In a *REAL* "Man versus Machine" match, there would
be no human interaction with the machine from the time the match started
to the time the match ended. However since the human champion can consult
with other humans between the games of the match, the machine should be
allowed to consult with other machines between the games of the match.
The human should not be allowed to consult with machines between games,
and the machine should not be allowed to consult with people between games.
In the first six games played over a year ago, the Deep Blue team said they
were not able to change any program parameters between the games of the match.
Kasparov lost the first game, but by the sixth was easily able to defeat
the machine -- Kasparov could learn, the machine could not.
In the rematch, Kasparov said he observed weaknesses in the machine, but
was apparently foiled when he tried to take advantage of those weaknesses
in the next game. Since Deep Blue still can't learn, could it be that the
human team members also observed these weaknesses and tweaked the program
between games so they weren't there for the next game? The machine wasn't
doing the learning, the people were. And could it be that these tweaks
introduced new weaknesses in the machines play?
The rules of the match do not forbid such parameter tweaking between games
(see below). If this really was a scientific experiment, then IBM should
announce whether any such tweaking occurred between games. It is of interest
to the AI community whether Deep Blue was good enough to play the entire
match without any human input between games, or whether Deep Blue relied on
humans to do its learning for it. If the latter, Kasparov was playing
against the human members of the Deep Blue team during the match also. It
was not a "Man versus Machine" match.
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)
These are the only match rules I could find that would have anything to do
with changing parameters, but they only apply during the game, not between
games. This is from http://www.chess.ibm.com/watch/html/c.8.html
>13. At any time during play, IBM may replace
> any or all of the computer hardware and/or
> software being used to play the games provided
> that:
>
> a. Any work carried out in the playing
> hall is absolutely necessary, and
>
> b. All work is carried out only when it
> is not Kasparov's turn to move.
>
> Any work that can be carried out via the mirror
> terminal in the Deep Blue control room shall be
> carried out there. In the case that work is
> necessary to perform maintenance in the
> playing hall, Kasparov will receive a recovery
> bonus of five minutes on his remaining clock
> time to compensate for the disturbance caused
> to his concentration by the on-site repairs.
Douglas Rupp
Mike Gherrity wrote:
>
> However since the human champion can consult
> with other humans between the games of the match, the machine should {only}
> be allowed to consult with other machines between the games of the match.
This is cute, DB uses the WWW between games to draft chess playing
computers
everywhere to help it figure out how to beat its human opponent.
Somehow I
think people might object to this.
Douglas B. (aka D.B.) Rupp