A computer glitch might have led to the supercomputer Deep Blue's famous upset victory over chess grand master Garry Kasparov in 1997, according to an IBM employee quoted in a new book.
Deep Blue was defeated by Kasparov in 1996, although the match did contain the first single-game victory of a computer program over a world champion. The 1997 rematch was another story, however: Deep Blue won the deciding final match after three consecutive draws. But in Nate Silver's new book "The Signal and the Noise," one of Deep Blue's operators suggests that a glitch in the computer's software might have been at the heart of at least one of the games.
At the end of the first game, Kasparov had forced Deep Blue into an unsalvageable position after 43 moves, and Deep Blue's response was to move its rook in a way that didn't make any sense to its human opponent. This may have rattled Kasparov, who could not understand the move and may have decided the computer was playing at a higher level than him.
As it turns out, the move really didn't make any sense. According to an anecdote from the book, initially recounted by the Washington Post, a bug in the program, which the engineers thought had been fixed, made it so that when Deep Blue was cornered, it picked a move completely at random.
Murray Campbell, who worked on Deep Blue and other supercomputers for years, told Silver:
A bug occurred in the game and it may have made Kasparov misunderstand the capabilities of Deep Blue. He didn't come up with the theory that the move it played was a bug.
Silver's book is about how people and machines make predictions, and why some work and others don't. In Kasparov's case, his internal model of Deep Blue could have been thrown off by the bug, making him overestimate the computer's cleverness.
"The Signal and the Noise" goes on sale this week; for a full review of it, click over to the Washington Post.
Devin Coldewey is a contributing writer for NBC News Digital. His personal website is coldewey.cc.
