HAL, THE disgruntled computer anti-hero of 2001 A Space Odyssey– "I know that you and Frank were planning to disconnect me, and I'm afraid that's something I cannot allow to happen" – lives on. Or at least a first cousin of his, "Watson". Running on some 15 terabytes (15x1012 bytes) of data and 2,500 parallel processor cores each able to perform 33 billion operations a second, the latter has this week taken on and beaten the best two human contestants that American quiz show Jeopardycould throw at him. And in the process won $1 million for his creator IBM's charity.
The show, running since 1964, gives contestants the answer to a trivia question on anything from history to the arts, pop, science, or sports ... and expects them to reply with the missing question. To the prompt “a supercomputer that has digested 200 million pages of data, and which is IBM’s latest bid to create artificial intelligence (AI)”, a contestant would correctly respond “The question is: Who is Watson?”
The frivolity of a quiz show notwithstanding, the achievement is quite remarkable, a spectacular follow-up to the 1997 defeat by IBM’s computer Deep Blue of world chess champion Garry Kasparov. Not only is the computing power staggering, but Watson’s programmers have gone far beyond the challenges of chess to develop algorithms, that can tackle the fuzzy logic of human language with its nuance, slang, metaphor and cultural allusions, so-called open-domain question-answering, one of the holy grails of AI.
What seem to us like the simplest of questions hide hugely complex assumptions. And yet the programmers were able to get Watson to find answers from ambiguous clues, such as this: “It’s a poor workman who blames these”. “What are tools?” answered Watson.
But does this mean that Watson “understands”, “knows”, or finds “meaning” as human “intelligence” allows us to do? In truth the computer is simply noticing statistical correlations in vast amounts of data.
As lead IBM engineer Dave Ferrucci has noted in an interview: “For a computer, there is no connection from words to human experience and human cognition. The words are just symbols to the computer. How does it know what they really mean?”
It doesn’t. It guesses, brilliantly.