An IBM computer named Watson today took on Jeopardy champions Ken Jennings and Brad Rutter in a three-part match. We’re not supposed to know the results until the program air in February, but if a practice run on Jan. 13 is any indication, Watson should beat its human competitors handily.
The Watson challenge is a bit of a stunt, reminiscent of a 1997 challenge in which IBM’s Deep Blue computer beat world champion Garry Kasparov in a chess match. But in many ways, Watson is far more important–to IBM’s business and to the future of computing–than was Deep Blue.
Deep Blue was a purpose-built computer that was extremely good at one thing: playing chess, or more precisely, analyzing chess positions. It was really, really fast, but it wasn’t particularly subtle. It used its hardware, which included specialized “chess chips,” to work out the consequences of a move to greater depth than any human player could imagine. In the end, the project helped rekindle IBM’s long-dormant interest in supercomputers, but Deep Blue itself led nowhere because all it could do was play chess.
Watson, on the other hand. is more a natural outgrowth of the challenges of business analytics. Deep QA, Watson’s parent, is based on an open source project called Unstructured Information Management Architecture that aims to find answers to real-world question from the vast amounts of information that exists outside of structured databases. The Watson hardware is based on IBM’s Blue Gene/p series of commercial supercomputers. In other words, Watson may be a publicity stunt, but unlike Deep Blue, it is promoting real IBM products.
It’s also a riskier venture than Deep Blue. No one really expected Deep Blue to beat Kasparov and all the computer had to do to win credibility was play respectably. Jeopardy seems like a silly trivia game, but in computer science terms, it is much, much harder than chess. It has to find answers to questions that are ambiguously put and that often contain plays on words, things that traditional computers are no good at. furthermore, as principal investigator David Ferrucci points out, Jeopardy includes a significant luck element. Good players who get the three hidden Daily Double questions can dramatically increase their chances of winning a match. But there seems to be a widespread expectation that Watson will dispatch its human opponents.
When Deep Blue beat Kasparov, there was a certain amount of wailing and gnashing of teeth about the defeat of man by a machine. “In brisk and brutal fashion, the I.B.M. computer Deep Blue unseated humanity,” The New York Times wrote on the occasion. I don’t think we’ll see a similar reaction this time if Watson wins. We expect both more and less of computers, and our understanding of their strengths and limitations is better. Watson is not on the verge of becoming self-aware; it’s just a very big, very smart computer with a genius for extracting meaningful answers from the sea of information around us.