What's At Stake In IBM's Jeopardy Challenge? - InformationWeek
Software // Information Management
03:35 PM
Doug Henschen
Doug Henschen
Connect Directly

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

Advances in assisted Q&A could find their way into medicine, legal, tech support and compliance applications.

Serious Science

From a computing perspective, this grand challenge is very different than Deep Blue's chess matches. Watson must be able to decipher English-language clues it has never seen before. It must quickly search for possible answers in a fixed knowledge store and then apply myriad analytics to determine its certainty in an answer -- posed, of course, in the form of a question.

Language is often ambiguous, highly contextual and open to infinite meanings, idioms, slang and regional dialects. The domains of knowledge are completely unpredictable, with topics and questions known only to the producers of Jeopardy. The clues might be about history, literature, geography, politics, arts, science, or pop culture.

Watson can't just look things up on the Internet. That would be unfair to the humans. Watson is a stand-alone computer. Its memory is a fixed, 15-terabyte store containing the equivalent of 200 million pages or one million books on diverse topics (okay, maybe that's not so fair).

Rather than using a database, Watson relies on a content store based on the Unstructured Information Management Architecture (UIMA), something IBM developed and has since placed in the open source community. Watson "learned" all the information it stores in advance, meaning the content passed through an analysis/training stage in which it was marked up with metatags denoting entities such as people, places, things, dates and concepts.

Inside Watson, IBM's Jeopardy Computer
(click image for larger view)
Slideshow: Inside Watson, IBM's Jeopardy Computer

Search technology retrieves contextually appropriate information quickly, but the real secret sauce behind Watson is its battery of analytic scoring algorithms. That's what IBM Research spent years refining, and it's the key to a computer's ability to decipher the language used in the clue and score its confidence in having the right answer.

"The hard part for the computer is finding and justifying the correct answer," said Dr. David Ferrucci, the "Principal Investigator" behind the Watson project. "For each of thousand of plausible answers, Watson gathers evidence and uses thousands of algorithms to understand what's most likely to be the right answer."

Developing the knowledge set and analytic functionality was the major hurdle, but when that technology runs on a single processor, it takes as long as two hours to answer a single question with confidence. The next step was putting it all on steroids with a massively parallel processing (MPP) architecture.

Watson is essentially a workload-optimized system purpose built to play Jeopardy. It runs on 10 racks of IBM Power 750 Servers -- standard commercial hardware that's readily available. The total machine operates at 80 teraflops, or about 80 trillion operations per second. With that much processing power under the hood, Watson's performance went from two hours on a single node down to less than three seconds -- the threshold required to match wits with Jeopardy champions.

2 of 3
Comment  | 
Print  | 
More Insights
Newest First  |  Oldest First  |  Threaded View
How Enterprises Are Attacking the IT Security Enterprise
How Enterprises Are Attacking the IT Security Enterprise
To learn more about what organizations are doing to tackle attacks and threats we surveyed a group of 300 IT and infosec professionals to find out what their biggest IT security challenges are and what they're doing to defend against today's threats. Download the report to see what they're saying.
Register for InformationWeek Newsletters
White Papers
Current Issue
2017 State of the Cloud Report
As the use of public cloud becomes a given, IT leaders must navigate the transition and advocate for management tools or architectures that allow them to realize the benefits they seek. Download this report to explore the issues and how to best leverage the cloud moving forward.
Twitter Feed
InformationWeek Radio
Archived InformationWeek Radio
Join us for a roundup of the top stories on InformationWeek.com for the week of November 6, 2016. We'll be talking with the InformationWeek.com editors and correspondents who brought you the top stories of the week to get the "story behind the story."
Sponsored Live Streaming Video
Everything You've Been Told About Mobility Is Wrong
Attend this video symposium with Sean Wisdom, Global Director of Mobility Solutions, and learn about how you can harness powerful new products to mobilize your business potential.
Flash Poll