I missed the original showing of the Jeopardy shows featuring the Watson supercomputer. Now it is rerun season, so last evening I got to see the first round in which Watson fought to a draw with one of the trivia champs. I'll have to miss tonight's finale, but I know the outcome.I hope the two gentlemen who lost to Watson don't feel too bad. They were beaten not just by a fast computer, but by a huge team of programmers and experts who had pre-loaded Watson's data banks with things like whole encyclopedias, all of Bartlett's Quotations, and the like. Were Watson's lookup and correlation routines even faster, the men would have been outgunned in every instance except the two for which Watson answered incorrectly.
I'm glad they started the show with a small tour of the hardware behind the avatar. Watson is physically just as large as the mainframe computers I was using in the 1960s, and is accompanied by enough air conditioning equipment to keep a whole block of houses cool. Now let's speculate a little. In the 1960s, the current "supercomputer" was a CDC 6600; ten years later (1976), the first Cray-1 ran the first 100-MFlop calculation. A desktop computer my son and I built last year coasts along in the 100-300 MFlop range. So, thirty-four or -five years in the future, assuming the databases and algorithms are available economically, can we expect a desktop Watson? Possibly.
I have two questions: (1) When will a machine be able to appreciate a subtle joke? and (2) When will a machine no larger than a grapefruit be able to navigate the world with the ease of a typical house cat? A partial answer to both requires us to imagine that a true artificial intelligence needs a body with a density of sensors equivalent to the endowment of a natural animal. A machine that can not just see and hear but smell, taste, feel the breeze in its hair, and that has senses of balance, proprioception and at least the 20+ other senses Aristotle didn't include among the Big Five.
No comments:
Post a Comment