Inside Watson, IBM's Jeopardy Computer

IBM has spent four years and untold hundreds of millions of dollars developing Watson, a computer that can play Jeopardy. In fact, Watson answers questions so quickly and accurately that IBM challenged Jeopardy and two of the show's all-time champions to a match. Jeopardy's producers and the human contestants agreed. Taping took place January 14, and the three computer-vs.-human episodes will air February 14-16. Cynics may call it a publicity stunt, but the project has brought real advances in c

Doug Henschen, Executive Editor, Enterprise Apps

January 24, 2011

17 Slides
InformationWeek logo in a gray background | InformationWeek


Meet Watson, a supercomputer designed to play Jeopardy. Watson is what IBM calls a "deep question and answer" application that runs on 10 racks of commercially available IBM Power 750 Servers. It's essentially a Jeopardy-optimized system on a massively parallel processing (MPP) platform that performs at 80 teraflops, or about 80 trillion operations per second. Watson has a 15-terabyte memory, but it's not a database. Rather, Watson was trained in advance on some 200 million pages of content -- on topics ranging from history and science to politics and pop culture. The resulting content store is based on IBM's (now open source) UIMA (Unstructured Information Management Architecture) with entities such as people, places, things, dates, and concepts meta-tagged for retrieval. When Watson is presented with a Jeopardy clue -- submitted as a text file when the clue is revealed -- search technology retrieves thousand of possible answers. Thousands of algorithms (the real "secret sauce") developed by IBM Research then score the possible answers and measure the confidence in best answer available. If confidence is high, Watson will ring in -- through a mechanical actuator much like the one human contestants use.

SEE ALSO:

What's At Stake In IBM's Jeopardy Challenge?

IBM Empowers Smarter Cities

10 Most Powerful Supercomputers

About the Author

Doug Henschen

Executive Editor, Enterprise Apps

Doug Henschen is Executive Editor of InformationWeek, where he covers the intersection of enterprise applications with information management, business intelligence, big data and analytics. He previously served as editor in chief of Intelligent Enterprise, editor in chief of Transform Magazine, and Executive Editor at DM News. He has covered IT and data-driven marketing for more than 15 years.

Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.

You May Also Like


More Insights