A computer beating the world chess champion once could be called an upset. Beating him across a match meant something less easy to dismiss. In May 1997, at the Equitable Center in New York, IBM's Deep Blue beat Garry Kasparov 3.5 to 2.5 over six games. IBM's account of the match describes the event in the language it acquired almost immediately: a public test of whether computers were catching up to human intelligence.

That wording makes me uneasy, although I understand why it stuck. Deep Blue wasn't a mind arriving in public; it was a formidable chess machine built for a clean, bounded problem. Britannica records the useful prequel: Kasparov had beaten it 4 to 2 in their 1996 match, before the reworked system won the rematch a year later. In other words, the public didn't watch an intelligence wake up. We watched an engineering team return to a defined task and finally clear it.

Still, the last game had the theatrical force of a verdict. Chess had long served as a convenient stand-in for thought itself: rules visible, skill legible, a champion sitting under lights. A machine could beat humans at calculation without disturbing many assumptions about being human. Once it beat Kasparov, the distinction felt thinner, at least on television. I remember the match chiefly as newspaper imagery, a man at a board facing a box that offered no face back. The picture did much of the philosophical work.

The aftermath made the symbol harder to inspect. According to a twentieth-anniversary account in The Conversation, IBM declined Kasparov's request for another match and dismantled Deep Blue; it also released detailed logs only after the machine had been decommissioned. The same account notes that later analysis found serious mistakes in Deep Blue's play. None of that reverses the score. It does matter to the meaning we assigned to the score, because a rematch and an inspectable machine would have turned a revelation back into an experiment.

IBM had every commercial reason to stop at the perfect frame. The computer had done the job the public understood: it had met the champion and won. Another contest could add technical knowledge while weakening the image, which is a poor exchange when the image already travels further than the technical story. This is not evidence of a conspiracy, and Kasparov's suspicions are not needed to make the point. A company can retire a machine honestly and still freeze an event into mythology.

Deep Blue also exposes a confusion that AI keeps inheriting. Success at a sharply specified task is impressive, but it is not the same thing as a general theory of intelligence. The victory belongs beside the later argument about scaling search rather than encoding intuition: an approach can work brilliantly before anyone agrees on what its success should mean. Today we make the same mistake in a noisier setting, reading a fluent answer or a successful tool call as proof of a much larger capacity.

What lingers from 1997 is not that a computer won at chess. That was real, and worth the attention. It is that IBM let the machine leave public life at the exact moment it became a story about all machines, while everyone else was still deciding what question the match had answered. I would have preferred the untidier version: another match, more logs, less icon. Technical history improves when its famous objects have to keep working after the applause.

Sources: