Computers process enormous volumes of digital content. Much of this content is unstructured text that is created by we, the people. Computers have not been able to process our created content in the same way that humans do (or try to) - by parsing it for it's meaning. The pursuit of systems that can do this - interpret, process, and respond to unstructured, direct language requests, over a broad range of topics, has been a holy grail for artificial intelligence researchers, with considerable recent progress. In this pursuit, the classic test is called the Turing test, for Alan Turing's recognition that once it is not possible to distinguish whether a computer or a human is responding to input typed into a terminal, our place in the world will have changed.

IBM has not yet attempted to achieve this ultimate test in a general, referred manner, but has always pushed our state of the art in machine cognition. This is especially true in solving well formed problems like chess, in which Deep Blue, an IBM supercomputer, was pitted against Garry Kasparov and won. However, taking on something far less structured - like playing a game of Jeopardy, has been a much more daunting task.

Designing a system that could square off against human competitors in a way that would allow the audience to transparently determine who the winner was with randomly selected questions is far more than a great public relations ploy. The computer and human contestants receive clues at the same time (though the computer receives the question as text, to avoid the difficult parsing problems of human speech). Like people, the computer must make judgments about how likely it is to have the correct answer, and trigger the buzzer when it thinks it has the answer. The demonstration serves as a useful benchmark of how well computers can understand complex information requirements, expressed in the form that humans do, in natural language questions and interactive dialog. Computers have to interpret and synthesize natural language text, then integrate and rapidly reason over the machine's available knowledge to deliver meaningful answers.

A recent article in the New York Times describes the amazing effectiveness of this capabilty on subtle questions like:

• A ‘Green Acres’ star goes existential (& French) as the author of ‘The Fall' (Who is Eddie Albert Camus)
• This action flick starring Roy Scheider in a high-tech police helicopter was also briefly a TV series (Blue Thunder)
• The greyhound originated more than 5,000 years ago in this African country, where it was used to hunt gazelles (Egypt)
• Classic candy bar that’s a female Supreme Court justice (Baby Ruth GInsberg)
• He was presidentially pardoned on Sept. 8, 1974 (Nixon)
• In 1594 he took a job as a tax collector in Andalusia (Cervantes)
• In Poland, pick up somekalafjor if you crave this broccoli relative (cauliflower)
• This ‘insect’ of a gangster was a real-life hit man for Murder Incorporated in the 1930s & ’40s (Bugsy Siegel)

Watson may not yet quite be at Ken Jennings level, but with this progress, it is only a matter of time! The practical applications of this - fact-checking, free-form natural language inquiries, etc - are exciting, and a little unnerving. Are you ready for the living room television to ask you, 'What's wrong, Dave?'.