The first time "Jeopardy!" champion Ken Jennings went on the popular game show, the ultimate prize was millions in cash. But when he faces off against IBM's super computer Watson this week, he's not just playing for money -- he's playing for all of mankind.
For the past four years, top artificial intelligence researchers at IBM have been preparing their mega-machine, Watson, to compete on "Jeopardy!" against all-time champions Jennings and Brad Rutter.
To win, the machine needs not just to know the facts, but to replicate a human's understanding of all the puns and wordplay woven into the game -- and then be the first to the buzzer.
This week, the much-hyped man vs. machine match-up airs on national television. And, Jennings said, the pressure is on.
"One of the first things I thought was, 'This time, I'm not just playing to pay my mortgage or something, or to feed my kid's college fund, I'm actually sort of representing 7 billion human beings against our new machine tyrants,'" he quipped. "That was a lot of responsibility. I didn't want to let people down."
When "Jeopardy!" first called him a couple of years ago to let him know that IBM was working on a supreme game-show machine, Jennings said he was "skeptical."
As a former computer programmer himself, he said, he knew the computer's limitations and doubted if IBM actually could pull it off.
But when he watched taped matches of Watson playing against top human contestants, he realized that beating the computer was hardly a foregone conclusion.
"Clearly, it was playing at a very high level. It sort of effortlessly handled the kinds of things I thought computers couldn't do," he said. "It could understand wordplay, it could understand things that were more conceptual than a single fact."
The three-day match, which airs tonight, Tuesday and Wednesday, was taped in January. Jennings can't reveal the winner, but said this: In speed, at least, Watson takes the top prize.
"Obviously, humans can't duplicate the precision of Watson's reflexes," he said. "If all three players are buzzing, it's going to be Watson because we just can't match its reflexes with our goopy neurons."
But when the machine gets something wrong, Jennings said, "it gets it spectacularly wrong."
For example, Jennings said an IBM developer told him that when asked for the Russian word for "goodbye," Watson gave the answer "cholesterol."
"To me, that's just crazy," he said. "There's no way a human player could duplicate that kind of mistake, but Watson has no idea. It just doesn't have all the checks and balances we do."
It wasn't until he saw the machine in action that Jennings also realized that Watson isn't as good at short clues, as it gets less time to think.
And then there are the so-called "train wreck" categories, he said.
"One of the developers was telling me that it'll have streaks of just terrible luck. ... [In] a category that has some angle that it's not getting, it still does fail pretty spectacularly," Jennings said, adding that in a practice session, children's book titles proved especially difficult for Watson.