IBM computer to compete on Jeopardy!
By John Lister
In 1997 an IBM supercomputer made history by defeating chess grandmaster Gary Kasparov. Twelve years later the firm says it’s ready to take on Jeopardy!’s Alex Trebek.
For those who’ve escaped its clutches, the TV show is a basic quiz format with the twist that contestants are given an answer and must give the correct solution in the form of a question. The show started as a play on a United States television scandal where ratings-friendly contestants on quiz shows had been given the answers in advance.
IBM used the show as the basis of a research project which began in 2007 and involved developing a system named Watson (after IBM creator Thomas J. Watson).
The project’s leader, David Ferruci, says the show’s format poses two particular challenges to a computer. It favors contestants with a broad range of knowledge rather than understanding a particular topic in depth. And it requires contestants to decipher which parts of the supplied answer are relevant to the required question. That involves processing a wide variety of information simultaneously, rather than solving one task quickly – traits which are more suited to humans than computers.
Watson attempts to answer these challenges via a BlueGene supercomputer designed specifically for parallel processing. It follows on from the task IBM faced in developing a computer which could compete at the top level in chess, a game requiring players to consider numerous possible moves and the subsequent potential outcomes.
According to IBM, it’s not just advances in processing technology which make its machines better suited to Jeopardy! today. For the system to work, it needs a large amount of general knowledge to be available in an electronic form. That’s more readily available today – though while competing on the game, the computer will have to store it’s ‘knowledge’ internally and won’t be allowed an Internet connection.
Ferruci’s team say that when the IBM machine competes on the show, it will not only give its answer (or rather solution), but give an estimate of how likely that answer is to be correct.

Related:





Stumble It!
