No computer will ever beat a chess grandmaster

What actually happened

This was a genuinely respectable position, not just classroom hand-waving. The Dutch grandmaster and psychologist Adriaan de Groot's work on chess perception suggested masters recognized patterns rather than calculated, and philosophers used chess as the standard example of what computers could not do. Kasparov himself beat Deep Blue's earlier version in 1996.

The 1997 rematch ended 3.5 to 2.5. Deep Blue was custom hardware evaluating around 200 million positions per second, which is not thought in any human sense, and Kasparov's defeat was partly psychological: he became convinced a human had intervened in game two, and collapsed in the final game in 19 moves. IBM dismantled the machine and declined a rematch, which he has been sour about ever since.

What makes the old line look worse is what came after. Stockfish, an open-source engine anyone can download, plays hundreds of Elo points above the human record. AlphaZero in 2017 learned chess from the rules alone in hours, by playing itself, and produced a style grandmasters describe as alien and beautiful. The interesting fact is no longer that computers can win. It is that humans now study engine games to learn the game.