Is Chess Finished?
Not since American Patriot missiles knocked Iraqi Scuds out of the skies in the Persian Gulf war has a scientific achievement stirred so much public wonderment. I am referring, of course, to the defeat of world chess champion Garry Kasparov in New York at the hands of IBM’s computer, Deep Blue.
In the weeks since this extraordinary event, two opposing opinions have arisen about its significance. On one side are those who see the match as a major milestone in history, the dawn of the age of artificial intelligence. As Newsweek put it before the encounter even got under way, “we stand at the brief corona of an eclipse—the eclipse of certain human mastery by machines mat humans have created.” And the critic Charles Krauthammer, in a cover story in the Weekly Standard (May 26, 1997) written after Kasparov was crushed, asserted mat in one of its games Deep Blue had passed the famous Turing test for the existence of machine intelligence, its output, at least in the restricted realm of chess, being indistinguishable from mat of the human mind. “We have just seen the ape straighten his back, try out his thumb, utter his first words, and fashion his first arrow.”
On the opposite side are those who downplay the meaning of the human defeat. “Deep Blue is just a machine,” wrote the Yale University computer scientist David Gelernter in Time (May 19, 1997). While the IBM device represents, in Gelernter’s words, a “beautiful and amazing technological achievement,” all its programmers did was to solve a problem which, though admittedly “much harder than adding numbers,” nevertheless amounted to raw calculation. As for intelligence, Deep Blue “doesn’t have a mind any more man a flowerpot has a mind.”
In my opinion, both these views are wide of the mark. Deep Blue’s victory is far from the stupendous feat which some pundits (and IBM publicists) have cracked it up to be, and the fact that it succeeded in beating the world champion by a score of 3½ to 2½ tells us little about the machine’s intelligence or about its impressive ability to calculate. As it happens, the computer played a mediocre game at best. The only questions raised by the match concern the shockingly bad form of Garry Kasparov, the greatest chess master ever to have walked the earth.
The effort to program computers to play chess began in earnest in the 1950′s, and there can be no question that over the past four decades, considerable progress has been made. Fantastic increases in the speed of microprocessors and refinements in programming have combined to bring computers to a level I myself would characterize as decent—if still not on a par with top-level championship chess. I know, from much personal experience, whereof I speak.
Beginning in 1989, an annual tournament, the Harvard Cup (so named by the Harvard students who organized it), has pitted a team of top grandmasters against the best computers in the world, including, on one occasion, Deep Thought and Chiptest, two of Deep Blue’s precursors. I have participated in all but one of these events. From early on, both the strengths and the glaring weaknesses of our silicon opponents have been plain to see.
Just as Gelernter and others have pointed out, the computers’ major advantage in encounters with human beings is calculation. It has been established that the total number of possible chess games which can develop from the starting position on the board not only exceeds the number of atoms in the known universe, but does so by a wide margin. As computers play, they sort through as many positions as they can generate in the time allotted to them, a task which (like humans) they necessarily perform incompletely, but with incredible speed and thoroughness. Deep Blue is said to examine 200 million positions a second. Against this advantage, however, one must place certain deficiencies: computers lack essential traits like intuition, imagination, and, most fundamentally, a capacity to plan.
The ability to calculate is at a premium in chaotic, open positions, where one’s pieces are not hemmed in by one’s own pawns but instead are thrust into the fray. There the computer excels. For the human player, the trick is thus to steer the game into quiet channels in which everything hinges on the formulation of a long-term strategy which can be carried out (in a general and not wholly precise way) over the next dozen or more moves, far deeper into the future than the computer can see.
At the Harvard Cup tournaments and in the many previous encounters between computers and human beings, the machines proved very dangerous in chaotic positions and managed to score a number of sensational wins. But in positions where tactics recede in importance and strategic planning comes to the fore, they were shorn of their strength. Not that the machines ever made crude errors, but they played purposeless computer moves against purposeful human players, and the asymmetrical nature of the competition told on their scores.
In the first year of the Harvard Cup, the computers garnered a total of one victory and one draw in sixteen games. As processing speed bounded upward, their chess play improved, but also leveled off. In 1994, their best year, the computers won thirteen and drew eleven of 48 games; in 1995, even as similar programs were deployed on more powerful hardware, their share of victories and draws fell. Given their tactical prowess, one certainly had to treat them with respect, but chess with any sort of interesting intellectual content, i.e., containing any form of strategic planning, they simply did not play.
Which brings us back to Deep Blue. Has the IBM team broken some kind of barrier? Do Deep Blue’s hundreds of specialized accelerator chips deliver something new, something of which grandmasters, as Kasparov himself has said, must be “afraid”? Judging by what I saw in New York, my own answer is an unequivocal no, and in this I am hardly alone. Many grandmasters, in fact, were struck by the appallingly low level of the play on both sides. The Indian wizard, Viswanathan Anand, spoke of the match’s “disappointing” chess, and to the American grandmaster Patrick Wolff, the computer in some situations played like a “numskull.”
Indeed, nothing in New York was a departure from the usual machine chess of the kind witnessed at the Harvard Cup. In some positions Deep Blue excelled at calculation, but in others it made moves that were aimless and just plain dull. What is unaccountable, however, is that Kasparov played even worse, and without a trace of the blinding bolts of brilliance we have been accustomed to see issue forth from him.
Kasparov lost two games in the match. Of the first defeat, in the second round, Charles Krauthammer has written that the computer played “Brilliantly. Creatively. Humanly.” I would call this a gigantic stretch. As we know from the instantaneous nature of its responses to Kasparov’s play, the first nineteen moves the computer spit out in this game were drawn from its vast database of chess openings; its microprocessor array was, in effect, still asleep. The task Deep Blue performed here is roughly comparable in complexity to what a computer does when searching through an electronic card catalog at a library, which is to say not very complex at all.
Only on move twenty, after Kasparov played a move that was not in its memory, did Deep Blue’s calculating engine engage: it paused and began to “think” for the first time in the game. The computer liked what it found. Not only had Kasparov allowed it to come through the treacherous shoals of the opening without having to calculate and possibly go astray, but he had acquiesced in a cramped and future-less position which he then proceeded to worsen with a few less than optimal moves.
Yet even after Kasparov found himself in a truly horrendous bind, the computer made an entirely pointless move. Although not of the type that did damage to its own position, this was one of many moments showing that the computer was playing without any sort of discernible plan. If anything, game two was an astonishing display of the difficulty even a brute capable of assessing millions of positions a second will encounter in mastering the mysteries of chess; after acquiring a winning position, Deep Blue committed two errors in its own domain of calculation, the second of which gave Kasparov the chance to force a draw.
Unfortunately, instead of seizing the moment, the champion resigned, perhaps the only case in history where a player of his caliber surrendered unnecessarily in a drawn position. But Kasparov’s premature capitulation should not obscure the fact that in its most praised game (as indeed throughout the match), Deep Blue made senseless if typically computer-like moves and was thus far from passing the Turing test for artificial intelligence.
Kasparov’s second loss, in the final round, is even more difficult to comprehend. Wielding the black pieces, he made a known mistake on the seventh move, which the computer, once again consulting its library of openings and not having to calculate at all, instantly exploited to obtain a winning position. Twelve moves later, Kasparov conceded defeat. His seventh move was either an unprecedented (for him) slip of the hand and mind—a hypothesis supported by the mortified expression on his face—or a conscious and inexplicable decision to surrender the game and the match.
What really happened to Garry Kasparov in this match is something I cannot adequately explain. There is some evidence to suggest that the same champion who shows no mercy whatsoever when dispatching ordinary mortals does have special psychological difficulties meting out similar punishment to machines. In 1994, Kasparov suffered a loss in an elimination tournament in London while playing against the computer Chess Genius, a defeat he accompanied with much talk about the computer’s unprecedented and phenomenal strength. But in fact Chess Genius was not the giant-killer it seemed to Kasparov. Two rounds later, Viswanathan Anand prevailed over Chess Genius with ease, using only four of the 25 minutes allotted him.
If Kasparov has another flaw, it is his curiosity; he is a relentless seeker after “truth” in chess. This, indeed, is one source of his power in competition with his fellow man. Against computers, however, he seems to favor moves designed to elicit information about how the program is constructed rather than going directly for the jugular.
Such complexes aside, there may be still another and simpler explanation for Kasparov’s failure in New York: a chess player can have a bad day or, in this case, a bad week. Who can forget how the Cuban world champion José Raoul Capablanca blundered away a crucial game at the famous 1929 Carlsbad tournament when his wife walked into the playing hall while his mistress was sitting among the spectators in the front row? In a purely sporting sense, computers do enjoy a significant edge: they play on the same level day in day out; they do not suffer from fatigue or emotional distress or distractions of any sort; they do not have particular anxieties about playing against other computers; and they never worry about getting caught cheating on their wives.
But they lack something else: in a word, creativity. And from the absence of that quality in New York, I can affirm that what we saw was neither the dawn of artificial intelligence nor even a demonstration that the problem presented by chess itself has at last been solved. David Gelernter is correct to say that the computer plays chess the way a “calculator adds or a toaster toasts.” But he is wrong to conclude, as he does, that chess has succumbed to the calculator.
Chess may be just a game, but it is also an art, and it is in the human dimensions of that art—the haphazard trajectory of imagination, the expression of passion, and the struggle of personalities—where not only the beauty but the actual strength of grandmaster play resides. Convinced of this as I am, I doubt that the kind of computers we know today will play chess any time soon on the level that Garry Kasparov displays when he is in form.
Of course, with the world champion losing to a machine, the activity to which I and others have devoted our lives has unquestionably suffered a blow to its reputation. But I am reminded again of how we all marveled at those Patriot missiles during the Persian Gulf war—a perfect 33 out of 33 attempted interceptions, General Schwarzkopf bragged at the time. Only on subsequent analysis did we learn that the Patriots failed to bring down more than a tiny fraction of the total number of Scuds launched by Saddam Hussein.