Man Vs. Machine

THE HOPE OF HUMANITY IS IN a good mood. On a bitterly cold Sunday in late April in Moscow, less than a week before embarking for New York City to represent our species in a battle that may one day become the prime landmark in technology's ineluctable march to surpass its makers, Garry Kasparov is padding around his mother's roomy apartment in slippers, a knit vest and khakis. He is happy to chat with visitors about how he is training for his encounter with a sophisticated silicon beast, a supercomputer especially outfitted to whack the human race down a notch--specifically by humiliating Kasparov himself.

This nine-day showdown, a rematch of a battle won by Kasparov last year, is scheduled to begin Saturday at the Equitable Center in midtown Manhattan. The $700,000 winner's share of the $1.1 million purse is the least of the stakes. This man-versus-machine battle may one day live in legend as a real-life variation of the John Henry fable. Every human being who has worried about losing a job to a computer, or has simply wanted to kick in a machine's screen at an error message, is rooting for this compact, darkly handsome 34-year-old Russian to prevail.

That match, of course, will take place over a chessboard, a circumstance that would seem to favor Kasparov. After all, he is not only the world champion, but is almost universally recognized as the best player who's ever lived. At one time that would have been enough. But for the first time in history, the question has arisen: can the master of a pursuit once thought of as the pinnacle of human intelligence beat something that has never lived?

In recent weeks Kasparov has been clear on this matter. "I don't think it's an appropriate thing to discuss the situation if I lose," he answers those foolish enough to ask him directly. "I never lose. I have never lost in my life."

But on this day he is cheerily talking about his training dacha in the nearby town of Podolsk when his mother, Klara Kasparova, interrupts. Throughout the conversation, the fashionably trim, unnervingly fretful onetime scientist has been hovering, trying to thrust him back toward his training regimen, making frantic timeout signals with her wrists.

"Garry, please," she urges, "you are way behind schedule."

"Yes, Mother," says the best chess-playing entity in the world. And, possibly, the last chess champion who will ever utter those words.

The reason is IBM's Deep Blue--a customized RS/6000 SP supercomputer, much improved since last year's dustup. Weighing in at 1.4 tons, it stands in twin black cabinets suggesting two Darth Vader Portosans. Inside are 32 nodes (computing engines in themselves) that work together to make it a computational terror. It does not think; it crunches numbers and evaluates chess positions. Billions of chess positions. The human brain, on the other hand, looks ahead only a few moves, but has something that Deep Blue will never attain: consciousness. If Deep Blue triumphs, however, it will be an emphatic indicator that artificial intelligence need not attempt to emulate the brain in order to surpass it. Computers doing what they do best--computing--can take them farther than we ever expected.

What would it really mean if Deep Blue won? Some people have argued whether chess would be diminished by this upheaval. The real question is whether we would be diminished.

After all, chess mastery was once considered the ultimate in human intelligence. Though computer pioneer Claude Shannon first outlined a strategy for a chess-playing computer in 1950, no one really contemplated mechanical grandmasters then. In the mid-'60s, scientist Herbert Dreyfus bluntly asserted that no computer would ever be able to beat a 10-year-old. He ate his words when an MIT undergraduate wrote a program that beat Dreyfus himself. But even as computers improved, experts felt secure that no machine could play chess at the highest level. Champions seemed to compete on an almost ethereal plane, where pure intellect merged with an artistic cunning that bespoke humanity's finest expression.

In his 1979 book "Godel, Escher, Bach," Douglas Hofstadter, professor of cognitive science at Indiana University, predicted that a chess program could never beat a human champion. He has reassessed, of course--and thinks that when the day happens, we will lose something. "There is a constant sense of encroachment by computers," he says. "What bothers me is the degree to which something incredibly simpler than our brain is starting to be able to do things that we do in surprisingly strong ways. It's taking away from the complexity of what we really are."

Perhaps for that reason, some chess purists hate the idea of a champion's risking his reputation--and ours--against a computer. "It distorts the very essence of the sport," says Boris Khropov, vice president of the Russian Chess Federation. "Does Michael Johnson try to run faster than a kangaroo? Let machines play against machines and people against people!"

In the future that's what may indeed happen--because the abilities of machine and human may be so unbalanced. But for now, we have the unique situation where biology's best is roughly comparable to technology's best. We stand at the brief corona of an eclipse--the eclipse of certain human mastery by machines that humans have created. How well Kasparov does in outwitting IBM's monster might be an early indication of how well our species might maintain its identity, let alone its superiority, in the years and centuries to come.

"This is much bigger than a world-championship match," says Kasparov. His mother echoes the sentiment with a mix of admiration and concern. "In all other areas they've proved that machines are stronger--they move faster, they can get you to the moon. But in the 21st century, the challenge for the human being is whether you can create a machine that's equivalent to the human brain. Humans cannot relax until they are sure they have intellectual priority over the machines. It is a huge responsibility, a huge duty, but Garry has always been willing to try."

And if the computer wins? "It would be a moment comparable to Lindbergh flying to Paris," says David Gelernter, a Yale professor of computer science. "People knew it could be done in principle, but when it happened it was a significant cultural event."

For a moment, it appeared that that milestone was going to be reached last February, in Philadelphia, when Kasparov played what turned out to be a warm-up match against an earlier version of Deep Blue. The contest, like next week's face-off a series of six games, began with little fanfare. Most observers expected, in fact, that the human champion would convincingly vanquish the machine and walk off with the pot.

Then Garry Kasparov lost the first game.

It was perhaps the most shocking single-game defeat in a thousand years of pushing wood. "We were completely devastated," says his technical adviser Frederic Friedel. "There was a theoretical possibility that this machine plays better than God!"

But as Kasparov and his coaches analyzed all of Deep Blue's game, they discovered it had played some moves that God nor grandmaster would never have chosen. Kasparov set out to isolate and exploit Deep Blue's Achilles' heel. The next game, he says, "revealed fundamental weaknesses that could not be repaired during the match." And the match was even. But the champion had to fight to draw the next two games.

The fifth game proved decisive. After 23 moves, Kasparov offered a draw. The Deep Blue team huddled to determine whether to accept. Even though the machine's own ratings had it slightly behind, they decided to play on. Bad move. Almost immediately, Deep Blue made some tactical errors, and Kasparov convincingly turned the game around for a victory that clinched at least a tie.

By the final game, "I knew enough to put the computer in misery," boasts Kasparov, who did just that, vanquishing the computer and pocketing his $400,000 winner's share. But cognizant of his initial setback, he acknowledged that the IBM team had advanced computer intelligence to an alarming state. "I don't know if you can find a second player in the world who could recover on the second day after a painful defeat in front of the entire planet," he says. "But I'm a survivor. I had to go and win the next game. And that's what I did."

Can he do it again? Not if the Deep Blue team can help it. For the last 15 months, five computer scientists--two of whom have been on the project since its origin at Carnegie-Mellon in the mid-1980s--have been working overtime at the Thomas J. Watson Research center in Yorktown Heights, N.Y., remaking Deep Blue into a chess entity unlike any the world has ever seen. They smell success.

"I truly think we have a system which is far superior to what we had last year," says C. J. Tan, who leads the Deep Blue team. "I anticipate that we will win this match, overwhelmingly. If Kasparov plays at the same level as he was last year, we will crush him."

Why so confident they can win in 1997 when they could not in 1996? Because they have virtually remade their machine into a much better chess player: "It's inside-out, upside-down new," says research scientist Joe Hoane.

Deep Blue is smarter. The version of Deep Blue beaten by Kasparov was "a baby, and it didn't know too much about chess," says Feng-HsiungHsu, Deep Blue's original creator; Kasparov too easily exposed those flaws. So for more than a year now, IBM has run a "chess school" for its silicon prodigy. Grandmaster Joel Benjamin, working full time as a team member, would play the machine, find a weak evaluation in a certain kind of position and work with the programmers to deepen its understanding of that situation. As a result, the new version of Deep Blue software can consistently whip the 1996 model.

Deep Blue is more powerful. In early April, a newer, faster RS/6000 SP was delivered to the Watson Research Center. It essentially doubles Deep Blue's strongest weapon: the ability to search through millions of possibilities for the strongest move. Last year it could scan 100 million positions per second. The new machine, says Tan, can do 200 million a second. "Sometimes, 300 mil- lion," he says.

Deep Blue is more nimble. The programmers have been working on tools to allow them to make more significant adjustments between games. In 1996, they would notice some problems--for instance, a misevaluation of how a bishop worked in a certain position--but were afraid to make changes because they couldn't easily gauge how that change might affect other parts of the system. New software will allow them to make that sort of adjustment with confidence. "We'll never be as flexible as a human being," says Deep Blue scientist Murray Campbell. "But we'll be more flexible than last time."

You can't dismiss these computer scientists as nerds--the competitive juices are running at a near boil. Grandmaster Benjamin, in fact, sees a lot of similarities between chess masters and master programmers. Both are brilliant and obsessive, though "I think chess players like to drink more," he says.

Still, at every turn the IBM team will tell you that the Deep Blue effort is not so much an attempt to wrest a part of humanity away from our species as it is a scientific experiment. (It's also great publicity for IBM's supercomputers, which are sold to companies involved in data-mining, molecular analysis and routing airplanes--in fact, the actual Deep Blue computer that will play next week has already been sold to an unidentified corporate customer.) "We're in this era right now, in 1997, where computers can help us solve problems better than we've ever solved problems before," says Campbell. "If the problem is playing chess better than the world champion, all of a sudden, today or this year or next year, we can do that. The important thing is whether we can then also solve other problems that people can't solve--problems that are really important. "

If so, perhaps the lasting lesson of Deep Blue's extraordinary skill is that problems that seem to belong to the realm of human intelligence are potential sitting ducks for massive computational power. Even Kasparov, who half-jokingly vows to "defend the integrity of chess against this scientific idea to split everything into pure numbers," agrees that this is the way of the future.

"I have feelings that almost everything created by nature could be explained in pure numbers," he says. "The numbers can be huge, but theoretically, I think it's doable." In saying this, he is echoing the predictions of scientists who believe that we may ultimately discover rules by which complex systems operate and computers will be able to more accurately simulate everything from the weather to the bouncings of a roulette wheel. Already we use computers to try to figure out the palpitations of the body and the intricacies of international relations. In comparison, chess is a simple matter. All that stands in the way is Garry Kasparov.

Ultimately, this match will come down to a different kind of chess. Unfortunately for Kasparov, some of his most powerful weapons are not as potent against a mechanical foe. His trademark is tossing lightning bolts at the board--making bold, unconventional moves that throw opponents off their game plan. But Deep Blue doesn't flinch at that stuff: it simply keeps calculating. Taking risks plays right into its hands. "At every move, you're on the edge," says Kasparov's adviser Freidel. "You make one mistake, and it's over."

On the other hand, all computer chess programs to date suffer from a generic flaw: a vacuum when it comes to long-term strategizing. When there's nothing really happening in the game, it flounders.

In fact, depending on the particular situation, Deep Blue is either smarter than any human being could ever be--or maybe as clumsy as a weekend patzer playing for dollars at Washington Square Park. And that's where this battle really will be won or lost: which player will force the game on its own turf? Kasparov must create situations where Deep Blue doesn't "understand" the positions. On the other hand, Deep Blue's program must be clever enough to nudge the game toward situations where it can correctly evaluate the chessboard.

The IBM team believes that by taking the machine to "chess school," it has minimized those events where it suffers those characteristic weaknesses of a chess computer. "When I play the machine now, I can no longer use a anti-computer strategy to pick apart its weaknesses," says Benjamin. "That's why I feel it has a real chance to win."

Kasparov scoffs at this. "This is crap," he says. He thinks that his only vulnerability really comes from not being able to examine Deep Blue's previous games, a disadvantage that will lead him, he says, to treat the first two games as "intelligence missions" to probe the computer for vulnerabilities. "If they're so sure that that they've covered up those generic weaknesses, let them give me those games!" he says. "I will be able to build up a strategy which will be most painful and poison to the machine."

Who's going to win? (The match, incidentally, can be followed on the Web at www.chess.ibm.com; IBM expects millions to do so.) Most observers give the edge to Kasparov. At the age of 22, he was the youngest world champion ever, and 12 years later, he has never been stronger. In a recent tournament played in Las Palmas, Spain, he convincingly whipped a field of the six best (human) players on Earth. "I'm a hard worker," he says, "I have the talent and the determination." And despite the complaints of his mother that he's typically spending too much time on non-chess diversions--a new wife, a 6-month-old child, a collection of business ventures, a charitable foundation and a passionate interest in national politics--Kasparov has trained very seriously for this match.

Ironically, he attributes his surge to... computers. "The existence of computers is good for me," he says. Even when he's not training for a competition against a machine they are an utter mainstay in his work. "I can improve my chess with such a perfect mechanism." Kasparov even believes that in future high-level human-vs.-human competitions, grandmasters should be able to consult chess computers during the match. It's basically the same argument given by educators who believe that students should be free to bring calculators into their math tests--the "stupid" work of figuring can be done by machine, freeing the human to do the "creative" work of figuring new combinations. "People want to see great moves, great combinations, and I think that allowing us to use the computers--man plus machine--will boost the quality of chess." Just as it has in other walks of life, in creating everything from blueprints to business plans.

In that same way, there's a deep irony in this epochal clash between cell and circuitry. Deep Blue is a machine, but its training consists of programming and chess lessons from impassioned human beings. Kasparov is the standard-bearer for humankind, but he's sparring against a computer running a sophisticated program. The preparations on both sides mirror the relationship that all of us have developed with the silicon interlopers into domains we once controlled. We're not competing but collaborating. If computers were yanked from our presence, planes would be grounded, cars would stall, phones would go dead, cash registers would fall silent, printing presses would stop and the bull market would be hamburger. Silicon is our ultimate prosthesis; the industrialized world is a cyborg culture, and much of humanity's intelligent work is performed, however uneasily, with our digital companions. Computers and people are in this together. At least for now.

That's why we approach this week's historic matchup with more than a little trepidation. The terms of the partnership are subject to change. We humans may be cagier, but they are getting smarter by quantum leaps and bounds. And there's one word that we dread hearing from our silicon counterparts: checkmate.

In the space of 50 years, computers have gone from simply crunching numbers to challenging humans at their own games. A brief story.

1946 ENIAC: The first fully electronic computer, it weighed 30 tons and contained 18,000 vacuum tubes.

1959 GEOMETRY THEOREM PROVER: An early artificial intelligence program, GTP solved geometry problems more quickly by eliminating answers that were obviously wrong.

1965 DENDRIAL: Programmed with rules of chemistry, it combined them with logic to figure the molecular structure of chemicals much faster than humans.

1966 ELIZA: By simpy restructuring the user's sentences and spitting them back out, this program mimicked a psycho-therapy session.

1967 MAC HACK: The program became the first meaningfrul chess opponent tho human players, competing against them in a tournament.

1975 SAM: The Script Applier Mechanism could "read" and paraphrase newspaper stories and answer simple questions about their contents.

1970s AARON: The result of mor than 20 years of work by its artist creator, the world's first "nonhuman artist" creates its own paintings without human help. Its works have sold for $2,000. 1990 TIERRA: The program generates evolution in a machine by creating digital creatures and placing them in an electronic environment. Their world, complete with birth, death and mutation, is a virtual mirror of our own.

1994 CHINOOK: A checkers program, it battled the then checkers champ, Florida A&M matho professor Marion Tinsley, to a draw in six consecutive matches. He had lost only nine games since 1955.

1997 THE MARS ROVER: On July 4, NASA'S Mars probe will touch down and release it. It will explore the planet on its own and rpoert back.