My favorite story of the last couple of weeks is about Watson, the I.B.M. supercomputer who’s been working out on pro-grade Jeopardy players in real-time test matches. The idea is to fine tune him for a series of televised matches this fall against the best of the Jeopardy best. This being Jeopardy, of course, Watson must understand playful, sophisticated language and then search through something like all of human knowledge to come up with a correct answer in the 3 or 4 seconds it takes Alex Trebek to read the clue. We humans can do that, at least some of the time, and now a computer can, too.
Understandably, the article only hints at the complex architecture involved in pulling this off, but two qualities seem apparent: that Watson automatically assimilates the information given to him and that his success is based on applying meaningfulness to a series of choices. The first quality seems like a kind of superpower — “watch me absorb Shakespeare by noon” sort of thing — and the second feels like criteria we might use to distinguish a kind of intelligent behavior. Watson is only three years old.
Here’s a bit from the article that leaves the game show theme behind:
I.B.M. plans to begin selling versions of Watson to companies in the next year or two. John Kelly, the head of I.B.M.’s research labs, says that Watson could help decision-makers sift through enormous piles of written material in seconds. Kelly says that its speed and quality could make it part of rapid-fire decision-making, with users talking to Watson to guide their thinking process.
While it’s fascinating to imagine the many ways specialists might collaborate with Watson — in medicine, in law — I wonder why they would build another version when Watson can add anything that’s placed in front of him to his already large knowledge base. Sure, the programming might need to be adjusted to account for non-Jeopardy-like human interaction, but the point is that adding subjects like neural anatomy and environmental law to his memory won’t displace expertise about, say, the highest waterfall in the world.
If we follow this alternative far enough, we find ourselves with something more than a search engine or question-answer system, something like an intelligent research library that can negotiate its contents with a variety of readers. The more expert the question, the more precise the answer, and such answers might well draw from reference book entries, research articles, classroom lectures, and blogs. Watson might be the ideal reader the publishing industry has been searching for.
Perhaps this isn’t likely to come about in the short term, but we might imagine a time a generation from now when our primary subscribers are not individuals or institutions, but forms like Watson that spend every moment of every day reading what we make, remembering it, and telling us about it later — a form that anthologizes human understanding as a sort of transcendental media. I like the symmetrical sense of supposing that the information economy, so vastly automated, is leading to another based on extrahuman comprehension.