IBM supercomputer takes on Jeopardy champions

Tonight's the night for those keen on man versus machine confrontations as Ken Jennings and Brad Rutter, former champions of the US quiz show Jeopardy, take on a rack of IBM Power 750 Servers and a custom-built analytics engine.

Jeopardy is one of the longest running game shows in US history, and differs from most quiz shows by providing answers and requiring the contestants to work out the questions.

A sample answer might be 'Mount McKinley', to which the question required is 'What is the highest mountain in the US?'

"Jeopardy really represents natural language," said Harry Freidman, executive producer of the show.

"You have to understand the English language and all the nuances and all the regionalisms and the slang to play the game and get the clues. It's not just about a piece of information."

For the next three nights the former champions will pit their wits against the IBM Watson machine. The shows have been pre-recorded and the rumour is that the IBM system wins, but this will not be confirmed until the final game on Wednesday.

IBM developed the system as part of an exercise in data mining technology, since the show relies on matching results with initial data against a limited time which penalises early responders who get the answer wrong.

"The rate of growth of information is surpassing our ability to understand it and extract knowledge from it," said John Kelly, director of IBM Research.

"We decided that we needed to build a system that could extract knowledge at a much faster rate from vast amounts of data than any other human being or computer system can do."

This not the first time that IBM has taken on human games with machine interfaces. The IBM Blue Gene computer was able to beat a former world champion at speed chess, and recently computers have beaten humans at Go, one of the oldest board games.

"I wonder how many people will begin to believe that IBM Watson actually thinks? I'm sure it will be a dazzling display of finding," said Mike Rollings, research vice president at Gartner, in a blog post.

"Watson has to find all kinds of things to process the question (e.g. noun, verb, the question) and it has to process even more to find, weigh and select possible answers from historical information. But it is not thinking."