by Paul Hoffman
In my game with Kasparov, his manners were irreproachable. Of course, there was an Internet connection and four thousand miles between us, so that I did not have to endure his physical presence. Mig Greengard, a chess journalist and friend of Kasparov’s, was at the Long Island playing site. When the games began, Mig did his best to stick out his chest and imitate the Russian’s imposing demeanor. My first surprise was that I, and not Kasparov, was playing White: some software glitch prevented the organizers from giving me Black as they had promised. It was normally an advantage to have White, but now all my preparation as Black was useless, and my opening moves were unadventurous. The software had other problems too: the kid who was playing next to me watched one of Kasparov’s rooks mysteriously turn into a knight. The boy was screaming about this while I was trying to concentrate on my own game. The time control was a swift twenty-five minutes a side plus an increment of ten seconds a move, the same as in Kasparov’s rapid match with Karpov. Kasparov’s other five opponents were considerably weaker than I. After fifteen minutes they had all lost, and I was left to play one-on-one with the great man.
Kasparov had adopted a relatively obscure variation of the Sicilian and fluidly developed his pieces. My development, on the other hand, was aimless and phlegmatic. I fell behind by shuffling my light-squared bishop a ridiculous five times to a place it could have reached in a single move. I found myself trying to anticipate threats of his that weren’t even there, adopting a batten-down-the-hatches, curl-up-like-a-porcupine strategy. It was as if I was imitating my young son’s play—he is new to the game and hates to move any of his pieces beyond the fourth rank for fear that they will become targets. I muttered something aloud about my inert position, and Mig consoled me by saying that even top grandmasters find themselves playing passively against Kasparov.
The thirteenth world champion then orchestrated the exchange of queens and tortured me in the endgame by increasing his stranglehold on my wobbly position, confidently marching his king into the game along the exposed dark squares on my queenside. My weaknesses metastasized as the game continued. With only four seconds left on my clock, I lost a bishop on the forty-sixth move and resigned. I was upset by my blunder at the end: I wanted Kasparov to know that I was a better player than this game showed and worthy of writing about chess.
I talked to him a few days later on the phone in Moscow. “You didn’t stand a chance because of your teammates,” he said. “They weren’t too good.” He was being both truthful and charitable. “Your play of course was much too passive,” he continued matter-of-factly. “You must try to do more with White. But somehow you reached a position that wasn’t all that bad. You could have made me work for the win.”
I was still worried that Kasparov thought I was a patzer, and I complained to one of my friends. “Look on the bright side,” he said. “Whenever you play in a tournament for the rest of your life, you’re never going to face anyone stronger.” It was a good point.
FOR PLAYERS OF KASPAROV’S STRENGTH, COMPUTERS ARE THE NEW FRONTIER, and the day is near when machines will totally dominate human players. “They’ll win every match,” Kasparov told me. “And then the challenge will be: can the top player, on his very best day, when he’s rested and has no distractions from his personal life, still win a single game from the machine?” In other words, it will be a success for humans if they get on the scoreboard at all, losing a six-game match, say, 5–1 instead of being wiped out 6–0. This will require a remarkable shift in thinking for the world’s top grandmasters, cocky competitors who thrive on destroying their opponents. The silicon adversary may take the odd blow, but it will never go down.
How soon this day will come, whether it’s next year or five years from now, is hard to predict, but it’s certainly not decades away. The best machines already play world-class chess and have fared well against the likes of Kasparov and Kramnik. Aside from the machines’ success in high-profile matches, their very existence is revolutionizing the way all serious players approach the game.
In the mid-1960s, before the invention of the personal computer, a master prepared for an opponent by going through back issues of chess magazines and books trying to scrounge up any games that the adversary had played. If the master was lucky, he’d locate a few, which would provide limited insight into the opponent’s strengths and weaknesses and the openings that he preferred. Except in the case of a very famous adversary, the number of games available would generally be insufficient to discern the full scope of his opening repertoire.
Players in the early 1960s studied chess openings, of course, but because they had access to all of their own previous games and few of their opponents’, they tended to work more on expanding their knowledge of their own opening lines than on looking for flaws and improvements in their rivals’ games. If a player was fortunate to come up with a killer innovation on the thirteenth move of a well-trodden path in the Dragon, he could spring it on an unsuspecting opponent in Buenos Aires and then surprise another adversary three weeks later in Budapest who reached the same position—news of the innovation would not have traveled that quickly. Now the word spreads immediately, thanks to the Internet, and any serious player who checks major tournament Web sites will know about the improvement; he will therefore either avoid the opening line in his own future games, or welcome it because he has worked long into the night and developed a refutation that the innovator overlooked.
The modern information revolution in chess actually began long before the Internet. In 1966, a thick Yugoslavian periodical called Chess Informant began to anthologize games from top tournaments and annotate them with symbols so that readers around the world could follow the analysis without knowing Yugoslavian or English. (For instance, ?! indicated a move that was considered dubious but interesting and +- showed that White had a winning advantage.) The Informant soon became indispensable to top players. Tigran Petrosian was the world champion then, and he contemptuously called the new generation of players “the children of the Informant.”
Kasparov himself helped to pioneer the next explosion in chess information, two decades later. During his matches with Karpov, in the 1980s, he encouraged a group of programmers in Germany, who were the progenitors of ChessBase, to develop a computerized database that stored the games of major players and allowed him to categorize them, subgroup them, and slice and dice them in all sorts of ways. Kasparov could then examine, for instance, all the rook endgames that Karpov had reached, or all the middle games Fischer had played in which the American had two bishops and his opponent had two knights. The database was coupled to strong chess-playing “engines,” or programs, which would review the games, particularly the openings played, in search of better moves and chess truth.
Today, every tournament regular has access to the huge game collections and engines, with names like Fritz, Shredder, and Junior, distributed by ChessBase. Their flagship, MegaBase CD, includes 2,500 games of Kasparov’s, 580 of Pascal’s, and 22 of mine among its millions of games. (My games are pathetic blitz efforts, unsuited for preservation, from an online tournament in which IMs and GMs gave me a harsh schooling in time management.) A search in MegaBase on the Sicilian Dragon, the opening in which Pascal craftily defeated the Bulgarian, turns up thirty thousand games.
A top-tier chess game in the twenty-first century therefore begins long before the antagonists actually take their places at the board. Each cerebral gladiator, assisted by a team of human seconds, the sous-chefs, dissects the other guy’s previous efforts and puts his own favorite openings through the engines in search of innovations that he can unleash during the upcoming contest. As a result, the playing field is completely different than it was in Fischer’s day, but it is a level field nonetheless, because virtually all dedicated players have access to ChessBase software.
Not every competitor, though, has the temperament to stare at a computer screen for days, studying hundreds of games, assimilating what’s been played before, and guiding the machine to search f
or opening innovations. Pascal, for one, hates this sort of preparation and would prefer that the databases didn’t exist. Fischer himself believed that the vast game collections were robbing chess of spontaneity. To de-emphasize the role of memorization, he proposed a whole new variation of chess, called Fischer random chess, in which the initial set up of the pieces varies from game to game and is established just before the game commences. I, however, might have enjoyed those additional hours in front of the computer; after all, I always found studying chess more rewarding—and less stressful—than actually playing the game.
KASPAROV MAY HAVE HELPED USHER IN THE COMPUTER AGE IN CHESS, BUT he is also its most conspicuous victim. In 1997, he played a highly publicized match in New York City, touted as “the last stand of the brain,” against the IBM supercomputer Deep Blue. Not since Fischer-Spassky in 1972 had a chess match garnered so much media attention worldwide. Deep Blue was a 1.4-ton, refrigerator-size calculating monster, with 256 processors, or mini-brains, working together to outsmart its human opponent. Even the normally ruthless Russian looked vulnerable when up against this unflappable silicon Goliath, and despite the fact that his faults were on full display during the match—his hotheadedness, his moodiness, his paranoia, his inability to be gracious in defeat—no one wanted to see him lose to a machine. Kasparov had triumphed over computers before, including a weaker progenitor of Deep Blue the previous year, but he let this one get under his skin from the start.
Machines like Deep Blue are typically very materialistic—they’ll happily grab a pawn or a piece, even if they have to wait out a blistering attack, provided they don’t foresee themselves actually getting mated. A human grandmaster, on the other hand, will generally avoid snatching material if the price is being on the defensive for a dozen moves—he knows that a single oversight may lead quickly and irrevocably to checkmate. It’s like inviting someone to fire a gun at you: if you fail to dodge a bullet, it’s all over immediately. But if you can stand the pressure of being under the gun (and who could stand this other than an emotionless automaton?), this may be a great strategy: when your enemy has exhausted his ammunition, you start firing back and he is defenseless. Most human players, however, find it easier to attack than to defend: if you’re the gunman and you fire a bullet wide, you won’t lose on the spot and may even have time to recover before the counterfire.
Kasparov handily won his first game against Deep Blue in 1997, but was suffering in the second. At one juncture in the middle game, he clearly expected Deep Blue to penetrate his home territory with the White queen and pick off a pawn or two. This materialistic maneuver would have allowed Kasparov a modicum of counterplay—not enough, perhaps, to save the game, but more than he had. But the machine did not operate according to expectations: instead of going after the pawns, Deep Blue quietly clamped down on prospective counterplay by a simple shift of a bishop. Kasparov was dumbfounded that the machine was capable of playing such a refined, effective move. He resigned in disgust eight moves later. Subsequent analysis revealed that as bad as Kasparov’s position appeared, he could have achieved a draw with a long sequence involving a rook move that he had not considered. This was the first time, he said, that he had ever resigned a drawn position. Not only was he dismayed by his own colossal misjudgment, he was also confused by the computer. How could the same machine that had played so brilliantly earlier, foregoing a pawn meal in favor of a quiet bishop shift, now mistakenly allow him to wiggle away with a draw? Deep Blue’s erratic play was more characteristic of a human being, he thought, than of a machine.
Kasparov was also clearly frustrated that it was impossible to get a psychological advantage over Deep Blue, because the computer was simply playing pure chess: it didn’t care whether it won or lost. He could have spat at the machine or taken off all of his clothes and stood on his head, and it wouldn’t have made any difference—Deep Blue couldn’t be distracted. The machine kept coming at him, and Kasparov collapsed under the pressure of an opponent that he knew would never tire. It was the most stressful chess he had played since the aborted marathon with Karpov in 1984 and 1985.
He lost the sixth and final game—and with it the match, by the score of 31?2–21?2—in only nineteen moves, his shortest loss ever. The game had lasted only a little over an hour, and Kasparov was pummeled the whole time because he had suicidally chosen a known inferior and passive line of the Caro-Kann Defense—a particularly odd selection given that he himself had defeated the Caro-Kann when Karpov previously played it against him in four key games. By the end of the match with Deep Blue, Kasparov was a broken man, a bloodied, knocked-out boxer who could barely crawl out of the ring. “The pressure got to me early,” he told me years later. “By the last game I was in no condition to play chess—or do anything else.”
His recollection was an understatement: at the time he had been so wounded that he lashed out and suggested that Deep Blue might have cheated—a charge that IBM denied—by getting advice from human chess masters during the games.10 That’s apparently the only way he could justify to himself its humanlike play. Did he really suspect that IBM had secretly enlisted Karpov or another GM to feed moves to the machine? The world’s number one human player demanded a rematch. IBM, which was offended by his accusations, turned him down, ordered its programmers to clam up about how Deep Blue worked, and mothballed the machine. The press joked that Deep Blue had turned Deep Yellow.
EVEN PEOPLE WHO DIDN’T KNOW HOW TO PLAY CHESS TOOK KASPAROV’S DEFEAT personally, as if it had been a terrible defeat for humankind. Editorials lamented the triumph of a machine in an area that had been widely regarded as a hallmark of human intelligence. And Wall Street rewarded IBM: the stock shot up 7 percent during the ten-day match, adding $11.4 billion to the company’s valuation. In fact, though, Deep Blue had merely demonstrated that computers were good at what they have always been good at—calculation. The fact that a car can outrun a human athlete doesn’t detract from the New York marathon; why, then, would we let a machine’s success at chess somehow diminish human chess performance?
When electronic computers were in their infancy in the 1940s, pioneers in the nascent field of artificial intelligence set the goal of building a machine that played high-level chess. Unlike many other intellectual endeavors, the game was appealing to programmers because the machine’s competence at chess could be judged precisely by pitting it against rated human players under controlled conditions; the machine would earn a rating, and subsequent improvement could be quantified by how its rating increased. To measure a machine’s ability as a poet, say, would have involved subjective judgments. AI researchers hoped that, by building a chess computer that played as well as the world’s elite, they would better understand the nature of human intelligence.
The irony in Deep Blue’s success is that it provided no insight whatsoever into human chess playing; man and machine actually approach the game completely differently. Deep Blue chewed through three million moves per second, while Kasparov looked at, perhaps, one to three. Even at a speed of millions of moves per second, Deep Blue could not usually look very far ahead because the number of possible continuations was staggering. There are, for instance, more than 1040 (that’s the number 1 followed by forty 0s) conceivable legal chess positions—and even more possible chess games. “There are more chess games,” Kasparov likes to point out, “than there are atoms in the universe.”
The strength of world-class human players lies in their ability to decide who stands better in a given position, and this ability—general chess knowledge, if you will—is hard to build into a machine. According to popular wisdom, a grandmaster is a kind of human computing machine: when he chooses a move, he explores numerous continuations in his mind’s eye—if I push the king pawn, he’ll fork my rooks, but then I’ll trap his queen—at lightning speed with incredible precision. But this view is wrong: calculation is not the only, or even the main, secret of the master’s success. He depends much more on pattern recognition than on the exploratio
n of a mind-numbing aggregation of moves. “Calculation,” wrote Father William Lombardy, a grandmaster and Roman Catholic priest, “most often comes after the goal is achieved, the moment when a winning position converts into a mathematically forced win.”
The Dutch psychologist Adrian de Groot found that of the thirty-eight legal moves possible in the typical position, the master ponders an average of only 1.76. In other words, he is generally choosing between two candidate moves that he recognizes, based on the hundreds of thousands of positions he has played himself or seen others play, as contributing to the immediate or long-term goals of the position. Humans are better than computers at long-range strategic planning, where delicate, methodically executed maneuvers ultimately carry the day. Deep Blue and its silicon heirs excel in hand-to-hand combat, tactical dogfights in which brute computational strength prevails. Rather than despair that Homo sapiens has been somehow belittled by Deep Blue’s victory, we could rejoice that human beings uniquely possess knowledge and intuition that computers are not yet close to approximating. After Deep Blue, grandmasters who battled machines often defanged them by the deliberate anti-computer strategy of avoiding messy brawls.
Although the match with Kasparov showed off Deep Blue’s computational prowess, it did not actually settle the million-dollar question of whether man or machine plays better chess. The event was stacked against Kasparov, and not for the spurious reason that the machine was a fraud. The contest with Deep Blue was inherently unfair because, unlike high-level encounters between humans, Kasparov had the huge disadvantage of playing without having seen a single game of his opponent’s, while Deep Blue had for examination hundreds of Kasparov’s games. The match had been sponsored by IBM itself, and Kasparov believed that the company had a conflict of interest in ensuring that it was conducted fairly.