University Researchers Prepare Computer to Compete on ‘Jeopardy!’

Photo courtesy of Jeopardy Productions Inc.

Raymond J. Mooney will be rooting for IBM’s Watson computer as it takes on two former Jeopardy! champions next week.

“As a tech person, you’re always rooting for the machine,” says Mr. Mooney, a computer-science professor at the University of Texas at Austin. But this time his allegiance is more pointed: His research is behind Watson’s ability to understand Alex Trebek’s questions (or answers, in the classic Jeopardy! format). In fact, Austin and seven other universities contributed many of the concepts that IBM developers drew on to make Watson work.

Mr. Mooney will tune in Monday at a viewing party the computer-science department is hosting and then watch the final night of the competition Wednesday at IBM’s offices in Austin, where he and his colleagues will participate in a question-and-answer session.

His specialty, natural-language learning, will be important for Watson’s performance on the game show.

“Natural language is ambiguous in so many ways,” he says. Understanding the often pun-laden clues in Jeopardy! will be essential to the machine’s success.

Tomek Strzalkowski, an associate professor at the State University of New York at Albany, another of the affiliated universities, will also be rooting for Watson at a viewing party on his campus.

He hopes Watson’s performance on the program will attract more attention to developments in artificial intelligence.

His research focuses on sustained investigation, which means training a computer to ask deeper questions and provide answers that go beyond just the facts.

“We are looking at hard, open-ended questions,” he says. “Most of the time there is no single piece of information that will answer your question.”

He sees his research having strong implications in medicine and intelligence analysis and thinks it could be used in commercial applications as soon as the next couple of years.

Mr. Mooney and other professors at Texas have worked on similar applications as well, but while the artificial-intelligence community has made great strides in recent years, he cautions against unrealistic expectations following Watson’s performance on Jeopardy!

“We’re still a long way from having a computer that’s as smart as a human,” he says.

The other institutions contributing to Watson’s inner workings are:

• Carnegie Mellon University, whose researchers helped oversee the project and developed algorithms that help Watson find the best information on a topic and recognize the most likely answer;
• Massachusetts Institute of Technology, whose researchers’ work helps Watson process and use natural language;
• The University of Southern California, research from which helps parse large amounts of information and find inconsistencies or gaps within it;
• Rensselaer Polytechnic Institute, whose researchers are developing visualizations to show how Watson works;
• The University of Trento, in Italy, whose researchers’ work focuses on helping computers learn from questions and;
• The University of Massachusetts at Amherst, whose researchers work on search functions.