Watson Day 1: Can AI Really Beat Human Players at Jeopardy!?

by Gabrielle DeMarco on February 14, 2011

The battlefield is ready. Tonight’s Jeopardy! contestants have studied thousands of books. They have improved their Shakespeare trivia and bulked up their knowledge of obscure word origins. And one very special competitor has been running on all 100 algorithms for months now.

Tonight, we finally get to meet Watson, the latest feat in high-powered, high-thinking technology from IBM and the brainchild of several RPI alums. Tonight, Watson will begin its two-game tournament with former (and we think human) Jeopardy! champions Ken Jennings and Brad Rutter.

To honor this very cool occasion and the strong RPI links to the technology, RPI is throwing a three night viewing party for students and other guests at EMPAC starting tonight. This evening’s events will begin with remarks from President Shirley Ann Jackson, and one of the creators of Watson, IBM Research Scientist Chris Welty. Welty is also an ’85, ’89, and ’95 grad of the Institute. President Jackson and Welty will join fellow RPI computational gurus James Hendler and James Myers for a conversation on the fundamentals of the wonderful Watson. RPI will then light up the HUGE, high-def screen in the EMPAC Concert Hall for everyone to watch the game live.

Welty’s main research area is knowledge representation. At a very basic level that means Welty was very involved in “teaching” Watson language, including how to process and respond to it. Much of this teaching came in the form of ontologies, which are used by computer scientists to encode meanings in a language that computers can understand. Think something like the computer equivalent of grammar, which interestingly enough Watson also needed to be taught.

This work will pair nicely with the expertise in ontologies that Hendler will bring to the discussion. As one of the inventors of the Semantic Web and a guru on nearly anything that involves the World Wide Web, Hendler should provide interesting insight into Watson’s programming and potential future. For example, what if Watson could be let loose on the WWW? In order to play by Jeopardy! rules, the system is banned from Googling as us armchair Jeopardy! contestants are known to do during the game.

As the new director of CCNI, Myers will insert some good perspective on the powerful computational infrastructure that Watson represents as well as how high-powered systems like Watson and CCNI can continue to advance. Watson runs on massively parallel processors. For us non-cyberinfrastructure experts that means that the system breaks down problems into many smaller ones for faster, easier processing.

Given all this, the conversation before the big show should be interesting. How did IBM overcome the major challenges of getting a computer to not only answer basic trivia questions, but to play a game as complex as Jeopardy! with its humor and scholarly double entendre? No AI has come close to this to date. It seems near impossible for any machine to beat a normal person at Jeopardy!, let alone the likes of Jennings and Rutter who literally press the buzzer for every question before even considering an answer! My poor laptop (standard for the day, but arguably no Blue Gene) practically wheezes to its death every time I look up a single file among only a few hundred others. What does it take to go from something like my humble PC to Watson? I look forward to the experts take on the technology and of course the big showdown.