REFLECTIONS ON COMPUTER CHESS
Dept. of Computer Science
University of São Paulo
email@example.com - http://www.ime.usp.br/~vwsetzer
(This essay contains the main topics of an e-mail sent in June 13, 1996 to my colleagues, concerning a call for participation in an internal chess tournament. The original is in Portuguese, also available through my home page.)
Published in a revised form in the electronic magazine The Southern Cross Review, No. 2, Nov.-Dec. 1999.
1. Time magazine, in its issue of April 4, 1996, published an article by world chess champion Garry Kasparov, in which he comments on his tournament against IBM's "Deep Blue," a computer specialized in playing chess. Kasparov relates that during the first game on Feb. 2 DB sacrificed a pawn. This really shook him up, because he felt that the machine "understood the game" (sic) and was not just calculating the optimal moves. The shock left him so nervous, because he "could smell a new kind of intelligence across the table," that he lost the game.
2. After this game Kasparov talked to the machine's designers, and verified that it had simply calculated every possible move and deduced that 6 moves ahead it would take a pawn, recovering the loss. He regained his confidence, because "The computer hadn't viewed the pawn sacrifice as a sacrifice at all." He started trying to deduce what the program's strategies were and played accordingly. For instance, the machine had many stored openings, so Kasparov started making silly moves at openings, and it could not employ the stored patterns anymore. "My overall thrust in the last five games was to avoid giving the computer any concrete goal to calculate toward. ... In the end, that may have been my biggest advantage: I could figure out its priorities and adjust my play. It couldn't do the same to me. So although I think I did see some signs of intelligence, it's a weird kind, an inefficient, inflexible kind that makes me think I have a few years left."
3. In Scientific American, May 1996, pp. 10, there is an interview with the designers of DB, a parallel system with 16 nodes. "In three minutes, the time allocated for each move in a formal match, the machine can evaluate a total of about 20 billion moves; that is enough to consider every single possible move and countermove 12 sequences ahead and select lines of attack as much as 30 moves beyond that. 'The fact that this ability is still not enough to beat a mere human is amazing', Campbell [one of the 6 IBM prophets behind DB] says. The lesson, Hoane [another one] adds, is that masters such as Kasparov 'are doing some mysterious computation that we can't figure out.'"
4. If the reader has found it strange that I have used the non-academic expression "prophet," here is the justification for it: in the Scientific American issue of October 1990, Campbell made a forecast that "Deep Thought" (the precursor of "Deep Blue" - note the lack of modesty revealed through the former name), would beat any living human player "perhaps as early as 1992." In the later interview, he and his colleagues were no longer so arrogant and did not make any more prophecies.
5. This fantastic learning and wonderful show of humility notwithstanding, they said that Kasparov was doing some "mysterious calculation." To me it is obvious that Kasparov won precisely because he had an extremely limited capacity for calculating. (How many combinations of moves was he able to think of in advance, maybe a couple of dozens?) What he had was intuition, this anti-scientific ability of "figuring out from nothing." As happens with other master chess players, he is not able to describe this intuition process, which suggests a correct move without him having to stupidly test billions of combinations, simply because intuition is not logical, and - my fundamental hypothesis on thinking - it is not physical. So I will conjecture that it will never be introduced into a machine.
6. On this subject, it would be necessary to enter into considerations on the Theory of Knowledge (popularly called nowadays "Theory of Cognition"). Unfortunately, this would deviate from our main subject. But it would be interesting to mention the fact that some researchers look for the lost needle under the lamppost - that is, where they can see something with their present methods and hypotheses - instead of where it was really lost. An example of this is the article by D.J.Chalmers in Scientific American of December 1995, pp.62-68, on consciousness, which produced a big impact. Chalmers says: "Despite the power of physical theory, the existence of consciousness does not seem to be derivable from physical laws. ...If the existence of consciousness cannot be derived from physical laws, a theory of physics is not a true theory of everything. So a final theory must contain an additional fundamental component. Toward this end, I propose that conscious experience be considered a fundamental feature, irreducible to anything more basic." Wonderful, consciousness would be an "Urphänomen," a primordial phenomenon according to Goethe, who said e.g. that light should be considered as such. See, for example, his "Farbenlehre," his non-reductionist Theory of Colors, which in my opinion has resisted consistent criticisms. The following facts, which do not make sense from a purely physical viewpoint, should be investigated in the same way: the origin and the limits of the universe, matter itself, and so on. However, at the end of his paper Chalmers says: "[such ideas] might evolve into a more powerful proposal that predicts the precise structure of our conscious experience from physical processes in our brains." There you have it, current narrow-mindedness has closed the doors again: he was thinking of purely physical processes. By the way, he never mentions self-consciousness, because this obviously would lead to other non-reducible aspects, such as individuality and free will.
7. In its issue of April 1996, pp. 54-62, Scientific American published an article by A.Pentland on MIT Media Lab projects, some of which I had the opportunity of witnessing in 1995. They deal with the recognition of human faces, bodily positions and expressions, movements of a driver, etc. At the end Pentland says: "Other research groups at the Media Lab are working to grant our smart rooms the ability to sense attention and emotion and thereby gain a deeper understanding of human actions and motivations." That is, the old ideal of A.I. ("Automated Imbecility"...) still holds: machines will help us understand what we are. It is a pity that he does not call attention to the fact that, as with chess, all those programs do not simulate our way of thinking or recognizing. They only simulate the results of these activities, thus they tell us absolutely nothing about our "functioning." The enormous computational capacity involved in those programs and their complexity, the zillions of calculations which they have to perform to poorly imitate some of our capacities, should lead to a profound admiration, even veneration, of our own constitution, which reveals a wisdom and a structure of another nature than that which we can insert into any machine.
8. There is a joke saying that chess is an idiotic game, because even computers play it well... In this respect, it would be interesting to mention here what Mr. Wang Sen Feng, who had been a South-American Go champion, told me: there is no program that plays Go very well. Can it be that Go, albeit also a mathematical game, is more "intelligent" than chess?
PS [Feb. 1998]: In 1997 there was another tournament between Kasparov and Deep Blue, now incremented to 128 parallel nodes, if I recall correctly, which tests 200 million moves a second, that is, 36 billion movements in the allotted time of 3 minutes for each move. Kasparov lost 2.5 to 3.5. Many materialists were enthusiastic: a machine had overcome the imperfect human machine. But they did not reflect on the fact that he won 1game and there were 3 ties, in a handicapped setting, because DB had many games played by him in its storage, as well as strategies based upon his games, and he had no prior knowledge of the machine's strategies. How is it possible for a human being to beat a mathematical machine in a handicapped mathematical game, in which the machine can test an astronomical number of moves ahead? One of the possible logical conclusions is that the human being is not a machine, and has capacities that the latter will never posses - and which cannot be mathematically described.