Ray Kurzweil talks about robot Jeopardy! champ Watson

November 10, 2012

Forbes | Ray Kurzweil is a leading thinker, inventor, and futurist known for his track record of accurate predictions. In this video, shot in Kurzweil’s office near Boston, he talks to Techonomy founder David Kirkpatrick about Watson, IBM’s artificial intelligence computer system that famously defeated Jeopardy champions on the TV game show.

Kurzweil explains Watson’s human-like intelligence by referencing the Turing Test, an assessment that measures a machine’s ability to exhibit behavior indistinguishable from that of a human. On November 13 Kurzweil is releasing a new book, How to Create a Mind, which applies neuroscience research to the possibilities of super-intelligence. Kurzweil will speak at next week’s Techonomy conference in Tucson, AZ.

Kurzweil: Computers will never be able to command the subtleties and vagaries of human language. That is where human intelligence soars and really reveals itself. I actually agree with that sentiment. And that was Turing’s view when he based the Turing Test on human language. And that’s one of the reasons why Watson is so impressive, because [Jeopardy] is a human language game, and it deals with the subtleties of language. In the query are metaphors and similes and jokes and puns and riddles. It’s not straightforward language.

But not only does it understand that, it got its knowledge by reading natural language documents. People think that the scientists hand-coded all these facts about the Queen and the 16th century and all this information it has to actually respond to these queries. That’s not what happened. It actually read Wikipedia, and several other encyclopedias; in fact, it read 200 million pages of natural language documents—

Kirkpatrick: Kind of like spidering the web in Google—

Kurzweil: ­—And then read and understand them in their natural language. If you read a Wikipedia article, it’s not straightforward, it’s not written I some kind of computer code. I read them and sometimes I have questions—what does this mean?—and I’ve got to another Wikipedia article in order to resolve an ambiguity. It’s complicated.

It actually read Wikipedia. And then 200 million pages and it can come up with the right set of references to respond to a very obscure Jeopardy query in three seconds, it’s very impressive. And people say well, it didn’t really understand human language at human levels. That’s actually true. If it did understand human language at human levels we’d be at the point where computers were passing the Turing Test and we’re not quite at that point

Kirkpatrick: And would you call that point the singularity?

Kurzweil: No, but I’ve been saying consistently it will be around 2029.

Kirkpatrick: When computers can pass the Turing Test.

Kurzweil: And Watson should give us a lot of encouragement that that is well on track.