May 12, 1997
Computer in the News:
Kasparov's Inscrutable Conqueror
By ROBERT D. McFADDEN
EW YORK -- When it was all over Sunday, when the greatest chess player in history had been crushed, the machine that had done it -- IBM's RS/6000 SP, alias Deep Blue -- did the magnanimous thing: It kept its monastic silence. After days of Man-versus-Machine hyperbole, those who had looked to Garry Kasparov as the last, best hope could now only bemoan the coming days of ascendant computers.
But after a hard day of oscillating among billions of terrible options, RS/6000 SP -- a pair of featureless black monoliths, each an intimidating 6 feet 5 inches tall and resembling nothing so much as twin amplifiers at a rock concert -- remained unmoved and all but unattended in its air-conditioned closet high up in an office tower in midtown Manhattan.
Its inscrutable face gave nothing away. There were no rows of little lights to blink exuberance, no rich beery voice to gloat. Under its smooth metal skin, the chips, wires and electronic circuits were tightly packed and almost alive with invisible blips, but there was not even a radiator's clank or gurgle to whisper sympathy, nothing to show a caring or a cruel heart.
It would be wrong, of course, to imagine that RS/6000 SP has no personality. Throughout its six-game match with Kasparov over the last 10 days -- indeed, for most of its four-year existence -- it has exhibited qualities of scrupulous care, unshakable calm and remarkable powers of concentration and endurance.
Unlike its opponent, an emotional Russian whose frustrations over the board were often on display, RS/6000 SP never agonized, was never tired, never showed joy or disappointment, though its handlers were seized with elation or concern as it evaluated 200 million chess positions a second and flashed its moves and evaluations over a small screen in a room off the playing venue.
"Why is there such global interest in this match?" C.J Tan, the Deep Blue project manager, said at a news conference after the game. "Because it shows what technology can do for man and how far we can take it. The computer played grandmaster chess using both knowledge and speed."
Deep Blue's ancestry can be traced back centuries to dreamers and charlatans who envisioned machines that could beat humans at the ancient game of chess. One gimmick, invented by Baron Wolfgang von Kempelen in the 1760s, toured Europe as the Maezal Chess Automaton, nicknamed the Turk for the turbaned marionette in a cabinet that made the moves. Inside, it later turned out, was a tiny chess master.
A century later, Alan M. Turing, the British mathematician and computer scientist, developed a program that could generate simple moves and evaluate positions. And in 1945, Konrad Zuse, a German scientist, described a program for chess moves and even developed a crude computer.
The modern ancestors of Deep Blue, however, date at least to 1950, when Claude Shannon, a Bell Laboratories mathematician and one of the inventors of computer science, laid down an early blueprint with a proposal for a chess-playing machine. But significant progress in computer chess did not occur until 1973, when well-engineered programs that used brute-force calculation were developed.
Deep Thought, developed by Carnegie-Mellon University researchers in the 1980s, combined enormous speed and computational power with sophisticated analysis of positions, and became the first machine to defeat a grandmaster in tournament play: Bent Larsen of Denmark in 1988.
The first Deep Blue was produced by IBM scientists working under Tan in the late 1980s, and it quickly became the world's best chess-playing machine.
The second, improved version, the one that beat Kasparov Sunday, was born in 1993 at IBM's Thomas J. Watson Research Center in Yorktown Heights, N.Y. It was big and loaded with potential: 2,800-pound twin towers that were featureless on the outside, except for an on-off switch near the base and an electrical cord to plug it into a wall outlet.
But on the inside were 516 chess processors capable of examining 50 billion positions every three minutes, and it used its enormous calculating ability to find its way through the maze of chess games. It was not able to "think" in the way humans do, with flashes of insight, and had a limited capacity to focus on only the few promising lines of play, as the best human players do.
Moreover, though Deep Blue's specialty was chess, it was, IBM stressed, a general purpose computer that used parallel processing technology to handle enormous data computations in such diverse fields as molecular and fluid dynamics, air traffic control and financial analysis. A cousin would cost about $2 million.
What is it like to play against Deep Blue? Robert Byrne, a grandmaster who is the chess columnist for The New York Times and a former U.S. champion, recalled feelings of growing concern after playing many games against the ever-improving computer.
"No longer do you wonder if the machine will surprise you, but whether it will sweep you off the board," he said Sunday. "I would not quit challenging it, but I would not expect to win or even make a draw any more. Not in 50 games. It's a desolate thought."
In February 1996, Deep Blue was put to its toughest test to date -- a match with Kasparov. Despite its capacity to look at 100 million positions a second, it lost, 4-2, to the man who had been world champion since 1985. But Kasparov paid it a compliment, saying he had felt the stirrings of genuine thought in his opponent, at least in the way the results mimicked thought.
And, as usual, despair was not part of Deep Blue's makeup. Back in Yorktown Heights, preparations for a rematch began. There were almost daily brainstorming sessions with a research team: Murray Campbell, Joseph Hoane Jr., Jerry Brody and Feng-Hsiung Hsu, among others, and with Joel Benjamin, a grandmaster and former U.S. champion.
While the scientists boosted Deep Blue's over-the-board capacity to 200 million positions a second, Benjamin helped the programmers to refine its knowledge of chess: to recognize positional weaknesses and to understand long-range strategy as well as short-range tactics.
Sunday, the results were in: Deep Blue had demolished Kasparov in a scintillating final game of a deadlocked match. There was muted joy among Deep Blue's handlers, while Kasparov reacted with only-too-human words about a big corporation with unlimited resources and a machine that had not proved anything, yet.
But Deep Blue was beyond praise or criticisms. Indeed, in its windowless bare closet high over the city, there was no way to appreciate Sunday's lilac-and-burgundy sunset, no way to glimpse the city lights coming on, or the great Hudson flickering like mercury in the dusk.