Dozens of IBM researchers spent three years building one computer for what appeared to be one purpose. Watson, whose 10 racks of off-stage servers pack 13 trillion bytes of data, appeared on television for three straight nights this week to compete on the popular long-running NBC game show Jeopardy.
The gender-neutral Watson faced stiff competition in human champions Ken Jennings — whose 74-game win streak in 2004 has yet to be surpassed — and the previously undefeated Brad Rutter, who holds the record for monetary winnings. Yet when all was said and done, in the “Final Jeopardy” round of the third and final night, Jennings quoted “The Simpsons” to declare in parentheses following his written answer: “I, for one, welcome our new computer overlords.”
If you can’t beat ’em, he figured, might as well praise ’em. And neither Jennings nor Rutter could beat Watson. Rutter finished the first night tied with the supercomputer, and Jennings made a good run early on in night three. But Watson ran away with the game, the tournament and the million-dollar grand prize (which IBM split evenly between two charities).
Writing for Slate on Wednesday night, Jennings said it didn’t take long for him to realize that “this was to be an away game for humanity.”
Not everyone was impressed. New York Times television critic Alessandra Stanley wrote after Watson’s first appearance that “It’s not the match of the century, it’s more like the letdown of a lifetime.” Stanley approached the situation with such skepticism and casual indifference that we can only assume she owns a Jetsons-esque robot to dress and groom her every morning before she gets into her flying car that drops her off at the Times office and parks itself while she rattles off orders for her interns using only a chip implanted in her brain.
Stanley criticized the representation of Watson by an on-stage avatar as unfair, wondering why the human contestants couldn’t “consult a backstage ensemble of 2,800 experts,” referring to the number of computers to which Watson’s power can be compared. She concluded that if Watson loses “IBM should be ashamed: the company should have gone all out and sprung for a full 3,000 (computers).”
Perhaps she’s right — there are plenty of reasons that the challenge wasn’t necessarily fair for the proven human champions. But that’s not what makes this event special. It was an exhibition tournament, after all.
Putting all preoccupations with fairness and competitive purity aside, it’s remarkable what this machine can do. As IBM researchers explain in advertisements promoting the special challenge, Watson illustrates a breakthrough in the ability of computers to understand and interpret human language, idioms and even tone. At lightning-fast processing speeds, it applies these capabilities to a vast pool of knowledge, and if its best answer reaches the “confidence threshold,” Watson buzzes in before the humans can wrap their head around the question (or on Jeopardy, the answer).
On Monday night, the $1,000 prompt in the category “Final Frontiers” was as follows: “Tickets aren’t needed for this ‘event’, a black hole’s boundary from which matter can’t escape.” Watson, with 97-percent confidence, buzzed in. “Event horizon,” said the robotic voice.
In Wednesday’s “Dialing for Dialects” category, the $200 prompt would seem simple for a human but quite tricky for a computer. “Sprechen sie plattdeutch? If you do, you speak the low variety of this language.” Watson buzzed in right away: “What is German?”
The “question-answer machine” even makes informed risk decisions on the fly. Consider Tuesday night’s Final Jeopardy answer: “Its largest airport is named for a World War II hero; its second largest, for a World War II battle.” Watson wrongly offered Toronto, but saved itself by taking confidence level into account and only wagering $947.
If Watson’s capabilities were truly confined to winning quiz shows, the project would be a serious waste of time, effort and money. The technological advancements made by the IBM team represent a leap forward in the long-dreamed-of utility of artificial intelligence. According to the IBM employees featured in the specials, the same natural-language capabilities could do worlds of good in improving business logistics and even in helping doctors make faster and better diagnoses.
“It changes the paradigm in which we work with computers,” said one researcher.
But even without such an optimistic outlook, anything that puts science and engineering in the forefront of conversation — especially in a country that lags significantly in education for these areas and will likely feel the effects of that in the future — is a good thing. Anything that does it in a fun and stimulating way is even more valuable.
Matt Aaronson was the Daily’s managing editor in 2010. He can be reached at maarons@umich.edu.