IBM is putting the finishing touches on a computer program that will compete against human contestants on “Jeopardy!”
Comparing such an achievement to Deep Blue, the venerable tech company’s chess-playing program that beat world champion Garry Kasparov in 1997 is like comparing checkers to chess.
Chess is a game of simple statistical probabilities, a lot of them it’s true, and pieces with clearly defined powers.
“Jeopardy!” presents more daunting challenges for computers, which must weigh nuances of language including double entendres, puns, and analogies faster than Ken Jennings on crack.
The machine has been dubbed Watson in honor of IBM founder Thomas Watson. It is the culmination of a 3-year project involving a team of 20 with expertise in language processing, information retrieval and machine learning.
“The big goal is to get computers to…converse in human terms,” said David Ferucci, an AI scientist and the team’s leader.
In the contest, Watson will receive questions as electronic text, whereas the human contestants will, as usual, see the question and hear it spoken by host Alex Trebek.
Watson will use a synthesized voice to respond and select follow-up categories.
It will not be connected to the Internet during the contest, instead rendering answers from text that had been processed and indexed in advance.
Harry Friedman, the show’s executive producer, indicated he might invite Jennings to carry the flag for humans.
In 2004 Jennings won 74 consecutive contests and collected $2.5 million along the way.
In prepping for the contest, Watson will have stored a large chunk of the Web as indexed by Google, but it’ll mean nothing if the machine can’t understand the context of each clue.
For example, the sentence “I never said she stole my money” can have seven meanings depending on which word is stressed.
“We love those sentences,” Eric Nyberg said. “Those are the ones we talk about when we’re sitting around having beers after work.” The computer scientist from Carnegie Mellon University is on the development team.