Dr. Eric Brown, IBM research scientist, talks to the West Point community March 3 about WATSON, the supercomputer that beat Ken Jennings and Brad Rutter in a three-game match on "Jeopardy." Brown described some of the challenges IBM faced in building...
Dr. Eric Brown, an IBM research scientist from the T.J. Watson Research Center in Hawthorne, N.Y., briefed cadets and community members March 3 about the WATSON supercomputer that recently beat all-time "Jeopardy" champion Ken Jennings and Brad Rutter in a three-game match.
"IBM has a history of innovation," Brown said. "We started with tabulating machines and then progressed to punch cards and in the 1970s built the first commercially viable business computer with the System 360. In the 1990s, we built Deep Blue, the supercomputer that beat chess grandmaster Garry Kasparov in 1997."
Brown said the idea for a computer to win "Jeopardy" began in 2004 when Charles Little, then the vice president of software research, went to dinner with clients.
"While having dinner, Little noticed that everyone moved to the bar to watch television at about 7 p.m.," he said. "Everyone was watching 'Jeopardy' with Ken Jennings on his way to his historic winning streak. It was at that moment when Little wondered if IBM could build a computer to beat 'Jeopardy.' It would be an interesting and great way to demonstrate technology that deals with human language and can push us to build computers to interact with humans in a more natural way."
Humans have a hard enough time communicating with each other, so to build a system that understands human language sounds like a gargantuan task.
"Human language is ambiguous, grounded in human cognition and there are infinite numbers of ways to express the same thing," Brown said. "There are five dimensions in the challenges of building a computer that understands human language."
The best way is to build the computer to drive and measure the automatic question and answering technology along the five key challenging dimensions of a Broad open domain, high precision in analyzing and determining the best answer, need for confidence that the answer is correct, analyzing complex language and need for high speed (to ring the buzzer within three seconds.)
The technology invented to deal with this is Deep QA. This allows WATSON to generate and score many hypotheses that use thousands of national language processing, information retrieval, machine learning and reasoning algorithms to gather, evaluate, weigh and balance different types of evidence (or data) to deliver the answer with the best support it can find.
WATSON has stored and analyzed the equivalent of one million books including encyclopedias, dictionaries, news articles and plays. By training, WATSON evaluates the best documents to answer a "Jeopardy" question.
"The technology can be applied to medicine for diagnostic assistance and evidence-based collaborative medicine," Brown said. "It can be used for help centers and tech call centers and government to improve information sharing and education."
Related Links:
Army Chief Information Officer/G-6
Social Sharing