user-pic

Lights, Camera, Watson!

I am standing in the middle of a TV studio at IBM's research center in Yorktown Heights, New York. This is not the kind of bare-bones TV studio where they have a velvet backdrop in a small multi-purpose room so the executives can come down from their offices and talk about the latest quarterly forecasts to the business channel. No, this is an elaborate game show set, with banks of lights, dozens of cameras and a small army of hair and makeup folks. Craziest of all: They are preparing for a science experiment, one that has the potential to revolutionize how we think about computers.

Here, an IBM computer named Watson is going to play on the quiz show Jeopardy!. For the team of IBM scientists watching, it's the culmination of four years of work, and a very public test of their efforts. The matches they tape today will air February 14, 15, and 16, with Alex Trebek as host. Watson's competition will be Ken Jennings and Brad Rutter, the world's two best players.

watson.jpg
Watson's competition-ready poker face. Image courtesy of IBM.

The biggest challenge for Watson will be to understand human language, with all its subtleties and ambiguities. It is something that we take for granted but for computers it is devilishly tricky. For all the advances in computer science, there is no computer today that can astutely answer a broad range of questions posed in human language.

Why? Well consider the following example: "I shot an elephant in my pajamas." Now you probably imagine that the hunter walked out of his tent, dressed in night clothes, and pulled the trigger. But how do you know?

Maybe it was the elephant that was wearing pajamas. Maybe you were shooting with a camera. How do you teach a computer to understand?

And Jeopardy! questions are even trickier. Try answering this one. The category is Rhyme Time, and the clue is "a politician's rant and a froth desert." Stumped?

The answer is "meringue harangue." Amazingly, Watson was able to correctly answer this one.

Through a combination of incredible computing speed and very sophisticated programming, the IBM team has gotten Watson to the stage where it can now understand Jeopardy! questions and answer them in less than three seconds. At the moment it is playing at the level of a very good player. The question now is: Will it be able to beat the best?

Publicist's note: Michael Bicks is the producer of NOVA's Smartest Machine on Earth, premiering Wednesday, February 9 at 10pm on most PBS stations. Please check your local listings to confirm when it will air near you. Oh, and don't worry: We'll give you the inside science scoop, but we won't give away the ending. You'll have to tune in to Jeopardy! to see who--or what--wins the game.

Michael Bicks

Michael Bicks, Director of Little Bay Pictures, is the producer of NOVA's Smartest Machine on Earth.

NOVApbs Twitter Feed

    Other posts by this Contributor