DEEPER IN THOUGHT

By MICHAEL KRANTZ

It's move 16, and Deep Blue is thinking. Or rather, Deep Blue's 512 processors are reviewing 200 million chess positions per second in order to create the illusion that Deep Blue is thinking. And it isn't really Deep Blue either. It's what the guys at IBM's Thomas J. Watson Research Center in Yorktown Heights, New York, call Deeper Blue: the second generation of the original Deep Blue, the infamous chess program that one year ago threw a stunning uppercut to human self-esteem by winning the first game of its six-game match against world champion Garry Kasparov. Kasparov, of course, went on to score three victories and two draws to win the match and save mankind; the 33-year-old Russian isn't considered the best player in history for nothing.

The Deep Blue team, led by senior manager C.J. Tan, has been plotting revenge ever since, and is now prepping for the rematch, which will take place in Manhattan in May. Today, in this cramped lab at T.J. Watson, Deep and Deeper are playing their first father-son game, a sort of silicon Oedipal struggle. The first 15 moves are what chess types yawn at as "standard"--established openings. Very safe. No surprises.

Move 16 is when Deeper Blue pauses to "think." Finally, its human monitor announces, "F4." F4? An excited buzz sweeps the room. F4! Deeper Blue has advanced the knight's pawn two squares, loosening its kingside defense with an assumption of the superiority of its position that would surely be considered arrogant if a carbon-based life-form were making it. "This move was special," murmurs Joel Benjamin, a former U.S. champion and current Deep Blue consultant. The room nods in agreement. Deeper Blue is thinking.

Pretty soon, Deeper Blue is kicking butt. From F4 onward, its inexorable kingside march swallows one pawn after another, and Deep Blue resigns 18 moves later. The room erupts in applause. The same thought is on everyone's mind: the new program is better. The new program is a lot better. We're gonna crush Kasparov like a bug.

The bug in question isn't nervous, though--at least not yet. "I'll have to play well and have a couple of surprises, but I feel that my chances are still superior," Kasparov says over lunch in Manhattan the next day to an audience of six, including Tan. "I know quite a lot, and I'll control my temper and my psychology."

In person, Kasparov is something of a surprise. Handsome and burly, he has a temper and psychology more befitting a garrulous European uncle than a genius geek who spends his life hunched over a chessboard. During appetizers he enthralls the table with discourses on a diverse array of topics, including hot chocolate (the world's best is found at Cafe Angelica on the Rue de Rivoli in Paris), Kremlin politics ("Russia has no choice other than Lebed!") and his infant son Vadim--"I want to stay on top long enough for him to recognize his father as a champion."

Over entrees a smug Tan hits Kasparov with the news of Deeper Blue's smashing victory over the program that made him sweat last February, and suddenly he focuses, laserlike, on his favorite subject. Before last year's match, he admits, the chess world felt "a computer would have very little chance of beating a top grand master." That myth faded quickly. Halfway through Game 1, faced with daunting circumstances--"an open position, my king is exposed, many weaknesses"--Kasparov undertook a blitzkrieg aimed at Deep Blue's king, the sort of hell-bent gambit that has devastated every pretender to his throne. "Any human being," he explains, "feels uncomfortable feeling his king under pressure."

Deep Blue didn't flinch. His gambit, Kasparov admits, was "a complete disaster, because the computer simply doesn't care. If the threats are not real, it sees that. So the machine simply took all the pawns and defended its king." And for an industry that IBM had built in the first place, scored the first win over a world champion. "Then I realized," he says, "that this will be tough."

But so is Kasparov, and in the ensuing games he mercilessly exposed his opponent's weaknesses. Deep Blue is a data-grinding engine of staggering proportions: a 1.5-ton supercomputer able to sort 40 billion combinations in an average three-minute move, shining its searchlight far into a game's future to find a winning strategy. When your opponent is Kasparov, though, it's (thus far) impossible for even a 1.5-ton supercomputer to search far enough to be sure it chooses wisely. "Deep Blue sees everything in the searchlight very well," says research scientist Murray Campbell. "But after that, in the black beyond, it has to guess. And humans guess better."

What the program lacked was intuition--the ability to set traps, hatch plots, smell danger and generally enact the violent and paranoid predator from which the human race evolved and to which all great chess players return. What's left is playing percentages. Deep Blue refused to follow a strategy it recognized as a likely loser, even one that any decent grand master could see offered the best chances for victory due to, say, a blunder by a rattled foe. The machine just didn't go for it.

So if intuition remains solely the province of human intelligence, why not just fake it? Why not teach the machine how the chess building blocks it can understand relate to one another? Benjamin has spent the past year helping Deep Blue's programmers encode thousands of positional evaluation rules, leavening the program's computational prowess with what one might call street smarts. "The hardware can detect certain features of a chessboard," says Campbell. "Rooks on open files, pins, pawn structure. It's a matter of assigning weights to how important these features are in a given position."

In the final game of last year's match, for instance, Deep Blue let its bishop get trapped on the edge of the board, with little power and zero mobility. The awful tragedy of the edge-locked bishop wasn't fully salted into its code base at the time, so the poor computer was oblivious to the depth of its positional peril, and Kasparov won the game handily. But things won't go so easily for mankind this time around. Says a pleased Benjamin: "Deeper Blue understands more about bishops--when they're good, when they're bad, how to use them better. It understands rooks better. It understands knights better."

After lunch last week, Kasparov ran into the rest of the Deep Blue team in the lobby of his midtown hotel. Hands were shaken all around, but the smiles seemed a bit strained. There will be lots of emotion on both sides of the board come May 3. Everyone involved knows the match will make history, whichever way it goes. Last year's virgin Deep Blue campaign brought chess its widest audience since the Fischer-Spassky cold war match in 1972. "Chess is of secondary importance to the wider audience," says Kasparov, who nonetheless hopes to launch a chess-themed Website called Club Kasparov later this year. "It's the social contest. It's about the machine."

He's right. Modern history teems with tales of the potential usurpation of mankind by its own technology: John Henry vs. the steam drill. Dr. Frankenstein vs. the monster. Linda Hamilton vs. the Terminator. The genius of chess lies in the sublime tension between logical analysis (call it Truth) and human intuition (call it Beauty). Our fascination with Deep Blue derives from fearful wonderment at the possibility that computers, which have already surpassed us at the former, may soon produce some chilling emulation of the latter. Kasparov, the latest standard bearer in humanity's war against our own obsolescence, is stoical in the face of the challenge. He muses that God, observing tomorrow's computers, may feel something akin to grandfatherly pride. "Maybe the highest triumph for the Creator," he says, "is to see his creations re-create themselves."