Garry Kasparov has an editorial in the Wall Street Journal (via Matthew Yglesias).

Kasparov says:

For 50 years one of the recurring motifs of the science of artificial intelligence was a chess computer beating the world champion. It was almost an obsession in the community, with scores of teams in dozens of universities dedicating their academic careers to achieve this goal. The discussion never ceased over the decades: When would this momentous day come?

It came on May 11, 1997, at 3:30 p.m., when IBM's multimillion-dollar machine, Deep Blue, narrowly defeated me in a six-game match. One of the greatest milestones of artificial intelligence had at long last been achieved.

He's right. It was a milestone. But not so much for AI. As I argued last month, Deep Blue's defeat of Kasparov was more of a milestone for processing speed than anything else. Over the past fifty years, computers have gotten a hell of a lot more powerful, but not a lot smarter. Though there is a correlation, the two are not the same (this is the problem with the arguments of people like Vernor Vinge and his singularity).

True human-level AI, across such domains as natural language processing and flexible problem-solving, will not occur with the advent of pure processing speed. Breakthroughs in algorithms and architecture will have to occur as well.

Kasparov argues that IBM was negligent to dismantle Deep Blue, that there could have been much to learn about intelligence and cognition from studying it. Although I don't understand the rationale behind scrapping Deep Blue, and though it would have been interesting to keep it around, I'm not sure it would have been that valuable to understanding cognition. It was essentially an ultra-powerful decision tree searcher built for an extremely limited domain, with little to teach about generalized thought, pattern recognition, or other important elements of cognition.