New York Times on the Web Forums
CyberTimes
Kasparov vs. Deep Blue
Discuss world chess champion Garry Kasparov's unfolding match against I.B.M.'s Deep Blue computer.
(32 previous messages)
hipbone1 - 03:20pm May 12, 1997 EST (#34 of 37)
It is well known that chess masters recognize clusters of piece-positions and can easily memorize whole boards which have resulted from informed human chess-play, ie which represent "late" stages in real games between people -- but have difficulty by comparison memorizing a random placement of pieces on the board.
I have the impression that great masters also generate new clusters of this kind, which are then added to the chess "archive", as new correlations between disciplines are added to the archive of the Glass Bead Game in Hermann Hesse's novel, *Magister Ludi*
It is my impression that Deep Blue has been programmed to "recognize" (and that's the last time I'll put quotes around an anthropomorphizing term in this post, but I intend it wherever any anthropomophic metaphor is being used) certain human "ways of playing chess" -- I'm not too familiar with the specifics, but I mean by this "gambits" and styles of "end game" etc -- so that in effect it has knowledge that it faces a skilled human opponent and is chess savvy in something of the same way that its skilled human opponent is.
I would be interested in a prolonged series of matches between two Deep Blues, because I would like to see what clusters of piece-positions might emerge when the computers were not relying on knowledge of the "state of the art" human chess archive...
On reflection, a contest between two identical computers which had no such archival knowledge programmed into them would be more interesting as an experiment. The idea would be to watch the computers generating (by means of something close to sheer number crunching ability) new piece-position clusters, "gambits", "styles of end game" etc., of the sort which the great chess masters sometimes generate, to see to what extent the computers recapitulated the "archive" or "chess history" on their own account, and to what extent they developed what would look to current chess masters like "alien" clusters and strategies... which nevertheless turned out viable in terms of play...
I'd like to suggest that once the Deep Blue team -- and / or other interested AI folk -- are satisfied as to the mastery of chess at world championship levels by machines, they should turn their attention not only to Go but also to Hesse's Glass Bead Game... which has itself been viewed as closely and interestingly analogous to the computer and the Web, and crops up on a regular basis in discussions of AI -- see for instance Michael Heim's glossary entry "glass-bead game" in his book *The Metaphysics of Virtual Reality*, New York: Oxford University Press, 1993:
"A fictional game described in Hermann Hesse's novel *Das Glasperlenspiel* (1943), translated in English as *Magister Ludi* (the game master). Discussions of VR often evoke references to the glass-bead game because the game's players combine all the symbols of world cultures so as to devise surprising configurations that convey novel insights. Each player organizes the cultural symbols somewhat like a musician improvising on an organ that can mimic any instrument. The glass-bead game's synthetic, non-linear information play is a forerunner of hypertext and of virtual worlds. Hesse's fiction also touches on some of the human problems underlying the advent of cyberspace and virtual reality, such as the role of the body and of disciplines for deepening the human spirit."
I think we could learn much from the attempt to teach a computer to play my own very simple "HipBone Games" variants on Hesse's Game ( found at http://home.earthlink.net/~hipbone/ ) vs. a human opponent --and also vs. others of its own kind.
Charles Cameron, HipBone Games, <hipbone@earthlink.net>
ghabrecht - 04:34pm May 12, 1997 EST (#35 of 37)
It would be interesting to see Deep Blue face off against Yasser Serwain, or even Karpov, without any change in programming to modify for their play. Some might say that it is only fair that a player, whether it be man or machine, should be able to study their opponent before a match. A computer can stock-pile all information on any specific opponent in data files. What did Kasparov have to study about the computer’s previous games? Could Kasparov ever possibly have remembered or have had access to every move the computer had made in a similar situation? I think that the chess community would have laughed out loud had he walked in with a filing cabinet and started leafing through books during the match. Even then he would need to have over 1000 times the amount of time the machine had to make it somewhat fair as far as search speed was concerned. Kasparov was wrong when he said that this match proved nothing. It proved that the search capabilities that a computer has are much faster than that of any human being. This is something we have known since the advent of the machines.
millerman - 06:36pm May 12, 1997 EST (#36 of 37)
I'm still not impressed. That Kasparov could hold up 3 1/2 - 2 1/2 against Deep Blue's 200 million moves a second is miraculous; he did it by something which the IBM team could only dream of programming - human intuition. The day a computer can beat a human while limited to the human's three moves a second, will be the day I concede defeat. Of course, then they can start trying to make Deep Blue be moved by classical music, followed by parenting skills, then poetry writing...
Long way to go, guys. This isn't exactly HAL.
bradmcc - 07:53pm May 12, 1997 EST (#37 of 37)
Some thoughts:
(1) A recent article in Newsweek (sorry, NYT...) cited Kasparov as saying that in future humans in chess tournaments should have computers to assist them like school kids can have calculators on math tests: so that the human can concentrate on innovation and the computer do the routine stuff. If anyone says that Deep Blue or its successor(s) can innovate, my response is: "Great!" -- that frees the humans to concentrate on higher levels of innovation. To be maximally human is to go beyond all that is or ever has been, in socially constructive ways (to be, in the etymological sense, a: meta-physician).
(2) Deep Blue still has not put an end to the problem, which is not to *beat the best human*, but to find the algorithmic *solution* to chess. As former IBM associate John von Neumann said: "Chess is not a game."
(3) The game of Go will still be around seeking its *solution* even when chess is solved (my understanding is that Go is more algorithmically complex than chess.
(4) Let's assume somebody does *solve* chess. Will we be able to *understand* the solution, or will it be something that makes the solution to the four-color math problem look like "pons assinorum"(sp? -- reference is to a simple Euclidean geometric theorem)? Alan Turing speculated: If a computer ever thinks, we shall not understand how it does it. The *social* challenge of human(e)ly managing computing, including the dependence of society on "incomprehensible computer programs" is surely at least as pressing today as when Joseph Weizenbaum wrote his book: "Computer Power and Human Reason: From Judgment to Calculation".
To post a message, compose your text in the box below, then click on Post My Message (below) to send the message.
Copyright 1997 The New York Times Company
