PORTLAND, Ore.—Later this month, Jeopardy, billed as American's favorite game show, will air a special edition where the all-time greatest human champions, Ken Jennings and Brad Rutter, will compete against IBM's Watson cluster computer.
The grand prize of $1 million goes to the winner, with second place earning of $300,000 and third place of $200,000. Instead of shipping Watson to Hollywood, this special edition of Jeopardy was taped last month at IBM's
Hawthorne Lab (New York), where a mock-up of the original Jeopardy studio was
During trial sessions, Watson beat the humans—just barely. But the real contest will be aired on Feb. 14, 15 and 16.
IBM has set several "grand challenges" for itself in the past, but the Jeopardy challenge is not just a PR stunt—a criticism leveled at its earlier chess challenge that resulted in the Deep Blue supercomputer, which beat grand champion Gary Kasparov, but was built from specialized hardware that could not be readily adapted to other applications.
Watson, on the other hand, was built from the ground up to be adaptable to the kind of business applications for which IBM is famous. Unlike chess-playing Deep Blue, Jeopardy-playing Watson runs on a cluster of 90 commercially available Power 750 servers with
2,880 Power 7 cores and 500Gbytes per second throughput.
To be fair, Deep Blue, which beat Kasparov in 1997, did lead to IBM's 1999 grand challenge to build the world's first petaflop computer, IBM’s Blue Gene supercomputer, which is used today to solve problems requiring massive computation over large data—from weather forecasting to protein folding. However, the DeepQA architecture that has made Watson rival humans in precision, confidence, and speed, was specifically crafted to adapt to business solutions.
"Watson is the result of a long commitment to human language technology, large data analytics, and supercomputing, all coming together," said
John Kelly, director of IBM Research. "We knew that if we could get[Watson] to work on a commercial system, then the roll-out into other fields of Deep Q&A would be greatly accelerated."
Jeopardy host, Alex Trebek (standing), leads Watson discussion (from left): Harry Friedman, executive producer, Jeopardy; Brad Rutter, Jeopardy champion (earning $3,255,102); Ken Jennings, Jeopardy champion (earning $2.5 million); David Ferrucci, principal investigator of Watson DeepQA technology, IBM Research; and John E. Kelly III, IBM senior vice president and director of IBM Research.
IBM’s DeepQA architecture for Watson rests on a foundation provided by its 24 man-year effort to create a Practical Intelligent Question Answering Technology (PIQUANT) which was retread for the Jeopardy Challenge, and which led to the Open Advancement of Question Answering (OAQA)
systems initiative to make question-answering algorithms reusable across applications.
“IBM created the OAQA as a way to promote an open approach to building question answering technology that enables different teams to share components, perform end-to-end evaluations, and advance the overall state-of-the-art,” said Eric Brown, a research staff member at the IBM T.J. Watson Research Center.