In 1769, Baron Wolfgang von Kempelen introduced his "automaton chess player" to the courts of Europe. The automaton had the form of a seated Turk, and it was presented as a mechanical marvel, a sophisticated device of levers, gears, drums and cylinders that apparently played a pretty good game of chess.
A sophisticated device it must indeed have been; but one of its more important components was not presented that was the hidden assistant who actually played the games.
Emperor Joseph II of Austria was fooled by the automaton. So were Empress Catherine of Russia, and, supposedly, Napoleon. I don't know anything about the mental powers of the first two, but Napoleon's record as a military strategist suggests that he had the brains to play a decent game of chess himself. So it's natural to wonder, why were these people so readily taken in by Baron von Kempelen?
Partly, I think, because they were not chess specialists. At first sight, the game of chess is exactly the sort of thing that a clever mechanical device (or a computer) ought to be able to do superbly well. The board is small, the rules are few. There are no undefined elements, and no place where chance enters. It should be no big deal, you would think, to design a machine, or program a computer, to play chess.
And indeed, it isn't, provided that the computer is allowed to play as badly, as, say, I play chess (P - K4; resigns). But for the computer to play chess well, at Grandmaster level -- that's another matter. The problem, from the machine's point of view, is not the rules. It's the explosion of possibilities provided as move follows move. Unless decision logic is incorporated into the program, to limit choices, today's computers are simply not fast enough. The combinatorial problems of chess are presumably NP problems, in which the number of steps needed to define a winning strategy cannot be defined by any polynomial in the number of moves.
Despite all that, I think we are close to the last generation. Starting maybe next week, or maybe next year or five years from now, the point of no return will come. After that, no human will ever again be able to beat the best computer programs at chess.
But then, for me, we approach the most interesting questions Why did machines take so long to match humans? If Deep Blue's basic circuitry is a million times as fast as Kasparov, what is going on inside the human's head as he plays the games?
Is there some enormous parallel processing algorithm at work? In the related field of machine vision, Rosenfeld long ago pointed out that a human can recognize someone else's face, and put a name to it, in less than fifty cycles of our built-in wetware. Computer vision algorithms need millions or billions of cycles and do the job less well.
Is it possible that the trick is not parallel processing, but an ability to grasp a situation in chess not as a placement of particular pieces and a sequence of particular possible moves, but as a single entity, the way the rest of us look at and grasp the sense of a painting? That also seems possible, even likely.
In fact, do the best human chess players all employ what is essentially the same program, making allowances for variations reflecting their differences of temperament? Or, if we could compare and contrast the mental processes of Morphy, Alekhine, Steinitz, Capablanca, Fischer, and Kasparov, would they seem more different than they are similar?
Or -- my own favorite question -- is there something quite different at work inside all their heads, some non-algorithmic process that we have not yet begun to guess at? It is difficult to suggest this, without being reminded of Roger Penrose's controversial theory, described in his books THE EMPEROR'S NEW MIND and SHADOWS OF THE MIND. Penrose suggests that human thought is not at all like the algorithms embodied in our computer programs. He does not, unfortunately, say quite what it is, though he suggests that thought involves the collapse of a large quantum wave function.
I want to suggest another line of thought here, consistent with Penrose's ideas but not dependent on them. Whatever Kasparov is doing, it works. He is able to play a winning game of chess -- i.e. solve a NP problem -- in what is, by computer standards, a ridiculously small number of his hardware cycles. If humans can do this in one field, chess, they should be able to do it in others, provided only that they are willing to do the necessary preparation work.
Here is what we do. We take a group of talented young people, and we expose them to the history and theory of some famous NP problem. The traveling salesman problem will do nicely, it has been studied for a long time. Now we ask our talent pool - - without requiring that they tell us how they get their answers -- to provide their solutions to some knotty example of the traveling salesman problem. If they succeed, we gain new insight into the way that humans think. If not, we still reward them amply for their efforts.
That last comment is important. Maybe the reason we don't have good solutions to the traveling salesman problem is that there is insufficient motivation for anyone to provide them. Beyond all the rules and all the protocols, chess players need strong motivation as much as anyone pride, curiosity, maybe even greed.
Which leads me to my final point. The games between Deep Blue and Kasparov in Philadelphia have not yet begun as I write this. That leaves me free to explain exactly how I would like things to turn out. Thus-
Man and machine draw the first five games, each taking longer and longer to make their moves. In the fifth game, Kasparov appears to be a little weary. In the sixth and final game, he makes a horrendous and ugly error, which Deep Blue at once pounces on and forces checkmate.
At this point, Kasparov leaves the stage. He returns carrying a fire axe. He chops the computer to pieces, saying, perhaps through an interpreter, "Now who's the clever one?"
This sequence of events tells a good deal about human motives, computers, chess, and possibly me. CS, 4 Feb 1996.
[ The Game | The Program | Commentary | Grandstand | Deep Blue ]
[ IBM home page | (C) | (TM) | IBM Small Planet Pavilion | Internet 1996 World Expo ]