What's So Great About IBM's Jeopardy!-Playing Machine?

What is Watson?

It's the name of IBM's latest high-profile supercomputer, this time designed to play Jeopardy! rather than chess. The New York Times has a a forthcoming magazine profile on what Watson is, how it differs from IBM's past supercomputing projects and what challenges it had to overcome. The project sounds impressive, but Watson's success apparently has more to do with sheer computing power than any major language-processing breakthrough.

Watson was designed with practicality in mind, unlike Deep Blue which famously beat Chess Grandmaster Gary Kasparov in 1997. That skill -- the ability to play chess really well -- couldn't be used to make much money. Solving the "question answering" problem, however, has plenty of applications. (Just ask Google.) There is a lot of money to be made from developing a system that can "understand" questions posed in normal, everyday speech, rather than selected keywords, and return meaningful results quickly.

Watson never truly obliterates the competition and sometimes it even loses. Still, creating a question-answering machine is a difficult task because of the vagueness of language and making one that answers questions based on a pun, play on words, or other verbal trickery deployed by Jeopardy's creators is even harder.

So, how did IBM approach the problem? Rather than develop a machine that can decipher semantics -- what do these words mean and how do they relate? -- IBM took a shortcut of sorts and developed a high-speed computer that "thinks" in probabilities:

Watson uses more than a hundred algorithms at the same time to analyze a question in different ways, generating hundreds of possible solutions. Another set of algorithms ranks these answers according to plausibility; for example, if dozens of algorithms working in different directions all arrive at the same answer, it's more likely to be the right one. In essence, Watson thinks in probabilities. It produces not one single "right" answer, but an enormous number of possibilities, then ranks them by assessing how likely each one is to answer the question.

That's quite a feat, but it's not emulating human thought. It has side-stepped the problem altogether, relying on massive computing power and storage, as well as probabilistic number-crunching to approximate how we parse language.

In an IBM-produced video, the host of a fake game of Jeopardy! offers the answer: "In R.E.M.'s 'It's The End Of The World As We Know It' two of the men with the initials L.B." Watson replies with "What is 'I Feel Fine.'" It's not that the supercomputer didn't know the song or the band -- clearly both were in the database because its response was a phrase from the song -- it's that the computer didn't even understand an elementary question.