We recently had the chance to see Watson, IBM’s Jeopardy-playing super-computer, mercilessly pummel Ken Jennings and Brad Rutter, Jeopardy champions. While we believe that Ken and Brad knew just as many or more questions than Watson and Jeopardy is as much about the buzzer as the questions, Watson providing so many correct questions is nevertheless a feat. We also got to see the Nova episode that went into more detail about the development of Watson. IBM tried to use a rules based answer engine first, and then went to a pattern matching engine. We immediately recalled the fellow who trained his pattern matching spam filter to play chess. Anyway, based on what we have seen of rules based spam filters and pattern matching (Bayesian) spam filters, we are not surprised that Watson’s pattern matching question engine performed much better than the rules based engine. You may say we are surprised that IBM did not try that first.
To conclude:
- Let us remember that although the question algorithms give the computer the appearance of intelligence, this computer is no more intelligent than any past computer. That is, it is just pretty good and really fast at matching patterns of a certain type.
- Maybe we need more information, but based on what we know, we would not say Watson is a major breakthrough. IBM just was able to put together a sufficiently powerful computing system to store and retrieve all of the necessary information. We give credit to the smart guys at IBM.
- In the end, it was a buzzer pressing contest and no one can beat a computer at that.