Watson and Niche Intelligence

Congratulations are due to David Ferrucci and his team at IBM for pulling off an impressive (and probably ultimately lucrative) computational feat creating a machine that outstrips humans at Jeopardy. As I watched the final game on Wednesday, while impressed with some of the nuances the computer navigated, I rooted vainly for the human contestants. It has been 14 years since the last publicly touted milestone in AI development when Deep Blue outgunned Gary Kasparov in chess. As in the previous episode, the implications are not yet clear.

Contestants on Jeopardy, in order to answer a question, must press their buzzer before anyone else, but after a signal light turns on. Those who jump the gun are penalized by disabling their buzzer for a quarter of a second. Watson has no eyes or ears, so it cannot see the signal. Instead, it receives an electronic signal and it physically presses the buzzer using a rudimentary finger made from a magnetic coil. The system is able to actuate reliably with a lighting reflex time (I have seen both millisecond or microsecond reported) that consistently beats its human opponents.

Watson also gets special treatment for the delivery of the questions. They are delivered electronically like an email or a tweet and thus Watson, which like all computers can read exceptionally fast, is busily searching its databases and scoring its answers using 2880 parallel processes while the question is read out loud to the other contestants.

Watson does have a mouth and it says quirky things like 'Lets finish off Chicks Dig Me'. But behind this facade, how does it decide when to speak? I doubt it is using speech recognition. After all, IBM recently announced a deal to combine Watson technology with speech recognition technology from Nuance. So there must be a human (traitor!) signaling Watson when it is time to select the next question.

These handicaps illustrate that the scope of artificial intelligence is still quite narrow. Indeed I believe the next decade will continue to see impressive and useful developments in "niche intelligence." Most of us enjoy the benefits of Google and GPS navigation on a daily basis. In fact, I believe the most impressive artificial intelligence in the world is not Watson but the Google search engine. Just because it processors and other silicon bits are distributed globally rather than crammed into a room-sized rack of computers, does that make it any less of a 'computer'? In one sense Google is more human in that its cells are constantly being created and eliminated.

Niche intelligence is not unlike a so-called 'idiot savant'. Dustin Hoffman's character in Rain Man has a remarkable facility with numbers and a photographic memory, but he is unable to function socially or in other dimensions. AI takes niche intelligence to a new extreme, performing marvelously in narrow, well-defined contexts. But at the end of the day, Watson's general intelligence is probably somewhere between an ant and a small rodent.

It will be exciting to watch as the next niche intelligences unfold and deepen the ability of computers to enhance our lives in unanticipated ways. It will be a long hard slog. Will these niches eventually expand and converge into truly intelligent entities? It is hard to say, but I thank Watson's mentors for stimulating and entertaining our minds.