Deep Blue, Garry Kasparov and the chess upset that showed AI’s potential : NPR

Garry Kasparov, left, takes a pawn in the first minutes of a chess game against IBM’s Deep Blue computer in Philadelphia on February 10, 1996. Feng-hsiung Hsu, right, Deep Blue’s lead designer, types a move into the computer.

Tom Mihalek/AFP

hide caption

toggle caption

Tom Mihalek/AFP

Could a machine surpass the best human mind in the world? Thirty years ago, it was still an open question, but a historic confrontation between a chess grandmaster and an IBM supercomputer answered it.

On a cold February day in 1996, hundreds of chess fans descended on the Pennsylvania Convention Center in Philadelphia. They held scoreboards and watched a giant video screen.

In the next room sat Garry Kasparov, world chess champion for 11 consecutive years. In front of him: a human substitute playing on behalf of a supercomputer that IBM called Deep Blue. It was designed to test the limits of artificial intelligence and could calculate around 100 million chess moves per second.

Computer programs had been able to beat a human at standard chess since the 1960s, but in 1996, most experts – including chess grandmaster Michael Rohde – still believed that the mind of a champion had the edge over Deep Blue.

“When he’s trying to make a decision, he sees all the possibilities, but it’s very difficult for him to evaluate whether one position is slightly better than another,” Rohde told NPR. “And that’s where humans still have a big advantage.”

As the match began, Kasparov played aggressively – probing, attacking.

Deep Blue responded. He didn’t feel tired or distracted, and he practiced in every match Kasparov played.

As the hours passed, the balance changed. Kasparov miscalculated; Deep Blue didn’t. And then — checkmate: for the first time ever, a computer had beaten a world champion in a regulation game of chess.

“I likened it to Mission Control in Houston, you know, when we landed on the moon,” said chess grandmaster Joel Benjamin, who helped IBM develop Deep Blue. “It was that kind of electricity, you know? Everyone was jumping up and down and getting really excited. It was really exciting.”

Kasparov would regroup, winning the next three matches and drawing two to win the 1996 match. He spoke two days later about his shocking initial loss to Deep Blue.

“He attacks, you know? He finds the shortest route to a weakness in your position,” he said. “He doesn’t hesitate, he has no doubts, he’s not afraid of your illusory threats. And that’s why, you know, it was [the] it was the worst, and you know, it was a well-deserved massacre.”

Technology continued to advance, and a year later an improved Deep Blue would outright defeat Kasparov in a six-game match, winning two and drawing three.

Today, computational intelligence surpassing that of Deep Blue is commonplace, even on the smartphones in our pockets. Yet AI still surprises us with its ability to imitate us. And like Deep Blue, it has its skeptics.

But that game in Philadelphia on February 10, 1996 marks the moment we first proved that an intellect we had built could beat the best among us at our own game.