Blog Archive

Wednesday, May 24, 2017

Backchannel - We Are All Kasparov: When Deep Blue beat the world chess champion 20 years ago, we learned a huge lesson. Just not the one we thought. By Editor and Writer Steven Levy

We Are All Kasparov

When Deep Blue beat the world chess champion 20 years ago, we learned a huge lesson. Just not the one we thought.

(Stan Honda / Getty Images)
The room where it happened was decked out like a faux study—a place where a couple of friends might engage in a friendly game of chess. But the people at the chessboard were professionals, and only one was paid to play chess. One was IBM computer scientist Murray Campbell, whose job it was to move pieces at the instructions of a computer he helped program. He sat with an air of detachment mixed with anticipation, like a passenger on public transit not sure where the bus will stop. The other was world champion chess player Garry Kasparov, whose concentration was intense enough to start a fire in a rainforest. His head hovered over the chessboard as if trying to identify which piece was threatening to betray him. His ankles shook. He was clearly under epic stress. Meanwhile, his putative opponent —a supercomputer housed elsewhere on the 35th floor of this midtown skyscraper — not only did not suffer stress, but did not even know what stress was.
Follow Backchannel: FacebookTwitter
I was in that room, for a few minutes at least, taking a turn at occupying one of its eight seats. It was February 1997, and I was covering the Kasparov-Deep Blue match—the historic contest where IBM’s computer would beat the world champion—for Newsweek. In my own tribe’s form of jousting, I had campaigned for the cover, despite the editor’s declaration that “we will never run a cover about chess.” I successfully argued that this was not about a game of chess, but rather about a much more epic contest between human and artificial intelligence. What clinched it was the cover line I suggested: “The Brain’s Last Stand.” It also helped that no celebrity died that week. So it was that Kasparov’s X-ray eyes and ultra-confident visage graced the newsstands of America, at a time when people actually paused at the newsstands to see what the weeklies put on their covers. And that “Brain’s Last Stand” line would come to be be invoked to this day. Even Kasparov, in a TED Talk last month, cited it twice.
I stand behind that provocation. Even though chess isn’t the toughest thing that computers will tackle for centuries, it stood as a handy symbol for human intelligence. No matter what human-like feat computers perform in the future, the Deep Blue match demands an indelible dot on all timelines of AI progress.
But that’s not the only reason why that six-game match in the Equitable Center is still so important. Two decades later, it’s clear that the significance of that outcome rests as much on how Kasparov was defeated. Though brute force computation and clever algorithms had created the winning positions against him, the champion was shattered by a well-planned psychological attack against him, executed by an IBM effort that leveraged its silicon advantages with human cunning. By the final session of the six-game match — one which began with the two opponents tied in points — Kasparov was a haunted ghost of himself. “I knew I didn’t have the energy for a complex flight,” he writes in his recent book, Deep Thinking, explaining why, early in the game, he made a risky move that effectively ended his chances for winning. The machine had gotten inside the human’s head.
And therein lies a parable.
Nine weeks or so before the match, I had lunch with Kasparov and C.J. Tan, the IBM scientist who managed the Deep Blue team. Both of those men maintained a veneer of cordiality that occasionally slipped to reveal the high stakes for each. Looking over the transcript 20 years later, a few things jump out at me. One was the confidence of each man. Tan had earlier remarked to a reporter that IBM was “not conducting a scientific experiment anymore,” and now he amended that to say, “It’s part of the experiment to how far the computer will go, and we’re doing everything we can to win.” Kasparov was annoyed that the prospect of an IBM victory was even mentioned. “I don’t think it’s an appropriate thing to discuss the situation if I lose,” said Kasparov. “I never lost in my life.”
The other interesting point was our discussion about the psychological aspects of the game. “I hope it will be as small as possible,” said Kasparov.
Today, those aspects seem to loom larger than the technological achievement of Deep Blue. It turns out that Tan’s remark about IBM doing everything it could to win included waging psychological warfare against its human opponent.
One tool was the element of surprise. Going into the match, Kasparov was frustrated that IBM had not shared printouts of Deep Blue’s practice games. He felt at a disadvantage because in a contest with any human, he would have a long history of match performance and would be able to tailor a strategy against that person’s tendencies and weaknesses. The best he could do against Deep Blue was to study the chess minds who helped IBM program its system—but the only grandmaster on staff was the American player Joel Benjamin, who was not top-ranked, and to Kasparov, not even worth researching. “I have better things to do in my life” than study Benjamin’s games, Kasparov told me. But he did suspect that IBM was secretly working with more experienced grandmasters. I asked Tan directly at our lunch if this was so, and the IBM-er replied, “No. Only Benjamin.”
But at the match, IBM revealed that formidable grandmaster Miguel Illescas was on its team, as well as two other grandmasters who were working in consulting roles. (In his book, Kasparov says he had known only that Illescas had played training matches against Deep Blue.) Kasparov had no way to prepare, and he was thrown off balance.
That was far from the only trick that IBM would use. Here’s a small example Kasparov cites in his book. During a match, human players sometimes will play games with the timing of a move. For instance, they might have a firm plan in mind, and if it’s going their way, instead of making the next move in the cascade right away they might let some time tick off the clock, to feign uncertainty. IBM actually programmed in the equivalent. In a 2009 interview with a chess publication, Illescas revealed that sometimes when Deep Blue instantly knew its next move, it would wait minutes before acting. When a chess computer stalls like this, it typically signals that the machine is having difficulty, or even has crashed. When Kasparov made his best move, the machine would play immediately, trying to give Kasparov the impression he had fallen into a trap. “This has a psychological impact as the machine becomes unpredictable, which was our main goal,” said Illescas.
The turning point of the match came in Game Two. Kasparov had won the first game and was feeling pretty good. In the second, the match was close and hard fought. But on the 36th move, the computer did something that shook Kasparov to his bones. In a situation where virtually every top-level chess program would have attacked Kasparov’s exposed queen, Deep Blue made a much subtler and ultimately more effective move that shattered Kasparov’s image of what a computer was capable of doing. It seemed to Kasparov — and frankly, to a lot of observers as well — that Deep Blue had suddenly stopped playing like a computer (by resisting the catnip of the queen attack) and instead adopted a strategy that only the wisest human master might attempt. By underplaying Deep Blue’s capabilities to Kasparov, IBM had tricked the human into underestimating it. A few days later, he described it this way: “Suddenly [Deep Blue] played like a god for one moment.” From that moment Kasparov had no idea what — or who — he was playing against. In what he described as “a fatalistic depression,” he played on, and wound up resigning the game.
After Game Two, Kasparov was not only agitated by his loss but also suspicious at how the computer had made a move that was so…un-computer like. “It made me question everything,” he now writes. Getting the printouts that explained what the computer did — and proving that there was no human intervention — became an obsession for him. Before Game Five, in fact, he implied that he would not show up to play unless IBM submitted printouts, at least to a neutral party who could check that everything was kosher. IBM gave a small piece to a third party, but never shared the complete file.
Kasparov was not the same player after Game Two. He fought to draws in the next three games, but in addition to the added mental pressures of dealing with what he clearly believed was his opponent’s skullduggery, he was physically wearing down. Though both sides were tied going into the final match, Kasparov approached it with dread. Asked in the press conference after Game Five about a comment Illescas made that he was now afraid of Deep Blue, Kasparov said, “I’m not afraid to admit I’m afraid!” Quite a difference from his pre-match confidence.
Indeed, Game Six was a debacle. From where we journalists were sitting, Kasparov seemed disengaged from the start. Afterwards, he claimed that he “wasn’t in the mood of playing at all.” On his seventh move, on what should have been a routine opening-game move, he made a mistake so egregiously awful that there were cries of disbelief in the auditorium where spectators were gathered. It was almost like he was throwing the game. He played in a desultory fashion for a few moves, and then resigned in obvious disgust. In a chaotic post-game press conference, Kasparov alternated between rage and depression.
The master had been mastered.
After the match, I pushed very hard for a one-on-one with Kasparov. We met in a ballroom of the Plaza hotel, where his team had been staying. The space was empty except for a few generic dining chairs, the kind used at banquets. We sat knee-to-knee — like chess players, but of course no board separated us. Kasparov immediately repeated a demand he had made in the press conference: that IBM agree to a rematch, under more favorable conditions.
And of course, he railed about not seeing those full printouts. “There is no information,” he complained. “I’m not interested in segments! I’m interested in the whole printout! It’s their obligation!”
But even at that stage, he was clear why he had lost. “I never got over Game Two,” he said to me. “It was sitting in my mind.” And then he summed it up: “It was a single individual fighting one of the largest corporations in the world.”
Indeed, IBM’s stock jumped up after the match. The company never agreed to Kasparov’s demand for a rematch.
Today, Kasparov is no longer competing for chess titles. He is a political activist squaring off against a more formidable opponent than even IBM : Vladimir Putin. His new book is a departure into a chapter of his life that defines him more than he’d like. He now talks about how the future of chess lies in collaborations between human and machine players. In his recent TED Talk, he didn’t revert to his complaints about IBM in the Deep Blue match.
In his book, however, he can’t help but revisit it — the printouts, the tricks, the misdirection, the grandmasters. He does say that he no longer believes IBM cheated its way to victory. But then he trots out a detailed scenario, rooted in that same Illescas interview, in which IBM might have made changes on the eve of the final game that specifically targeted the move he made that ultimately undid him. He implies, vaguely, that IBM planted Russian-speaking security guards in his private space, which might explain that last-minute shift. Not that they cheated. But still.
I dwell on these suspicions, even ones that may border on paranoia, for a reason. Amazingly, when the Deep Blue match occurred, AI was in its “winter” peri0d. Now it is flowering. We hear of amazing machine learning accomplishments on a daily basis. But in 2017, we view them differently. We view them as inevitabilities.
The prime example is last year’s contest, during which DeepMind’s AlphaGo program thumped an 18-time world champion in a series of five games. Go is a much more challenging feat for a computer than chess. Yet AlphaGo did not need to resort to any of the tactics that IBM used to distract, deceive, and ultimately destroy Kasparov. The human champion, Lee Sedol, ended with respect for his opponent and awe for how far computer science had come. But though the match deservedly received attention, it was nowhere near as mythic as the Deep Blue match was. The ground has shifted. Given enough time, money, and machine learning, there’s no cognitive obstacle that machines will not surmount.
When I covered Kasparov-Deep Blue match, I thought the drama came from a battle between computer and human. But it was really a story of people, with brutal capitalist impulse, teaming up with AI to destroy the confidence and dignity of the greatest champion the world had seen. That leads me to believe it’s not Skynet that should worry us about AI, but rather the homo sapiens who build, implement, and employ those systems.
Don’t get me wrong. I’m still on board with the scientists who believe that advances in AI will make life better for all of us. Ultimately, using the power of computation for cognition is a great and historic human enterprise. But may I add a codicil to that declaration?
Always check the printouts.