The Bug That Beat Kasparov — How a Glitch in Deep Blue Rattled the Greatest Chess Player Alive
Stop the clock on the forty-fourth move of the first game of the 1997 rematch. IBM’s Deep Blue has just slid a rook to a square that helps nothing — a quiet, pointless-looking shuffle in a position it is already losing. Across the board, in a Manhattan television studio, sits Garry Kasparov, the world chess champion and by most measures the strongest player who ever lived. He is winning this game and will win it cleanly a few moves later. But the rook move bothers him. He cannot see the idea behind it, and he is a man who can always see the idea behind it.
There is no idea behind it. The move is a bug.
Everything that follows in the match — the resignation in a position he could have saved, the accusations of cheating, the collapse over the last five games — radiates outward from that one meaningless square. To understand how a glitch beat the best player alive, you have to look in turn at the three things the move sat between: the machine that made it, the truth of what it was, and the champion who read it as genius.
The machine that made the move
The thing across from Kasparov had a long lineage. It began in the mid-1980s as a graduate-school project at Carnegie Mellon University, where a doctoral student named Feng-hsiung Hsu was designing custom silicon that did nothing but evaluate chess positions, fast. The first version was called ChipTest. The next was Deep Thought. When Hsu finished his doctorate in 1989, IBM hired him and his collaborator Murray Campbell to keep building, and the project took the name Deep Blue.
The approach was brute force, refined. Deep Blue did not think the way Kasparov thought. It did not have plans or intuition or fear. It had 30 processors driving 480 custom chess chips, and together they could look at roughly 200 million positions every second — searching deeper into the branching tree of possible games than any human ever could, then scoring what it found with rules tuned by grandmasters. The programmer Joe Hoane wrote much of the software; the team’s resident grandmaster, Joel Benjamin, helped sharpen the machine’s chess judgment. The whole machine was a wager that if you searched far enough and scored well enough, you would not need understanding at all.
In February 1996, in Philadelphia, that wager nearly paid off and then didn’t. Deep Blue won the first game — the first time a computer had ever beaten a reigning world champion in a classical game. Kasparov, shaken, regrouped. He studied how the machine broke down, steered the games into the slow, strategic positions where its horizon ran out, and won the match 4–2. Man beat machine. IBM asked for a rematch and spent the next year roughly doubling Deep Blue’s speed.
The rematch opened on May 3, 1997, in the Equitable Center in midtown Manhattan, with a prize fund that paid the winner $700,000. Kasparov won the first game with something close to ease. Then came the rook move on move 44 — the meaningless one — and then he won anyway, and the match looked to be following the pattern of Philadelphia.
What the move actually was
The move had done its work before anyone understood what it was. Years later, the writer Nate Silver pressed Campbell on what had happened, and Campbell confirmed the essentials: a bug had occurred, he said, one that may have made Kasparov misjudge what Deep Blue could do. As Silver reconstructs it, the machine had hit a flaw in its code, found itself unable to choose a move, and fallen back on a fail-safe that simply picked a legal move at random. The rook shuffle was that random move. It meant nothing.
That is the fact at the dead center of the match. No depth, no plan, no twenty-move calculation. A piece of broken code reaching for the nearest legal square. The strongest calculator on earth had, for one move, calculated nothing at all.
Kasparov did not read it as nothing. He read it as depth — as the machine seeing something so many moves ahead that he, the greatest player alive, could not follow it. He went into the second game looking for a ghost.
The champion who chased it
Game two was the one that broke him. Deep Blue played with a patience that did not feel mechanical. On the thirty-sixth move it captured a pawn, axb5, and a few moves later it passed up free material that every computer of the era was supposed to grab on reflex — choosing instead a slower, more positional path, the kind of move a human master would choose. Kasparov, convinced he was losing, resigned.
He did not have to lose. Analysts found afterward that the position was defensible — that he had walked away from a saving resource, a perpetual check that would at least have held the draw, missed by the best calculator on earth because he no longer trusted that the thing across the board could be beaten.
Now he was certain something was wrong. No machine, he believed, played like that. He suspected a human grandmaster had been feeding moves to Deep Blue, and he said so. He demanded IBM hand over the machine’s internal logs. He grew angry, and rattled, and the next three games — three through five — he could not break through. All were drawn.
The match came down to game six, on May 11. Kasparov, playing Black, opened with the Caro-Kann, a defense he knew well. On his seventh move he transposed two moves and played h6 a beat too early — a known mistake, the sort of slip a club player learns to avoid. Deep Blue answered instantly with a knight sacrifice, Nxe6, tearing open the front of Kasparov’s position. There was no recovering. He resigned after nineteen moves, the shortest game of either match, in barely more than an hour.
Deep Blue had won the match, three and a half to two and a half. For the first time, a machine had beaten the world chess champion in a full match under tournament conditions.
What was left
Kasparov did not take it gracefully. At the closing press conference he accused IBM of foul play and demanded a rematch, certain that a fair fight would go differently. IBM had no interest in a fair fight or any other kind. The company had what it wanted — the headline, the stock bump, the proof of concept. It declined the rematch, broke up the team, and retired the machine from chess for good. Deep Blue never played another competitive game. Its two equipment racks were given away; one sits today in the Smithsonian’s National Museum of American History, the other at the Computer History Museum in California.
By 2017, after years of analysis, Kasparov let go of the cheating theory. There had been no hidden grandmaster. The machine had simply outplayed him — helped along, by the most persuasive account of what happened, by the one thing no one designed: a glitch that, on move 44 of game one, made a meaningless move look like genius, and let a champion defeat himself. Not everyone buys that reading. Kasparov’s own camp has waved it off, insisting the odd move changed nothing in his play, and the machine’s builders have always called move 44 just a bug — a code fault, not a match-deciding stroke. It is a theory, in the end — the tidiest explanation anyone has offered for how the strongest player alive came apart, and one that can never quite be proven, because the only witness to what the move did to him is Kasparov.