building a better search engine, trivia edition

IBM doesn’t shy away from making big claims, especially in the field of artificial intelligence. Last year, the cat brain simulation by their Blue Matter team made a minor splash on tech blogs, until a few people took a look at the details of the experiment and almost coughed up a hairball when they saw that the team built a digital equivalent of cat brain soup instead of running a genuine biological simulation. This time, IBM’s next big PR effort seems to be far less pretentious and a lot more impressive. It’s name is Watson, and it’s a very effective search engine that can interpret natural human speech and come up with a single correct result in the context of a trivia game, specifically, Jeopardy. And in preparation for Wilson’s possible appearance on the show, the researchers put out a video documenting the machine’s ups as well as its occasional disappointing downs…

I wonder what Wolfram thinks of this, especially as he’s trying to compute much of the known universe with an experimental Mathematica/Alpha hybrid. Watson seems to be able to track down information based on a natural language query and come up with one good answer, something today’s search engines have trouble doing. Technically speaking, if you have a machine that can accurately interpret a human language to answer a question quickly and succinctly, you could push it to keep going and track down additional references for an extensive search which should take a lot less time than using Google. But then again, it’s very possible that a lot of people probably wouldn’t trust a machine, no matter how advanced, to have the right answer all the time and want to see its internal search results to confirm that they’re getting the right facts and figures.