Like it or not, but the robots are coming, and they’re not messing around. I know this because I’ve seen an IBM -developed artificial intelligence (AI) named Watson defeat two human opponents (all-time great champions, no less) in a round of Jeopardy. Granted, it was merely an exhibition round—the actual robot versus human tournament airs in mid-February—but it portends something fantastic: a world where our benevolent masters, mechanical men, will lead us to sweet salvation. Or, failing that, a world where AI is useful enough to assist humans in areas like healthcare and government. That’s cool, too.
IBM has been developing Watson, an AI that runs on 10 plain ol’, off-the-shelf Power 750 servers (running Linux!), for past four years. (That’s “him” up there. Thomas J. Watson would be proud.) But saying “the past four years” doesn’t really do justice to the fact that IBM has been conducting high-level research since its founding back in 1911. Consider Watson the culmination of IBM ingenuity going back long before anyone even knew what an iPhone was.
Needless to say, it takes an awful lot of horsepower to make Watson tick. Let’s put it this way: the top-of-the-line MacBook Pro has a dual-core processor. (You’ll recall that Motorola flipped out last week when it introduced the very first dual-core mobile phone, the Atrix.) Two is plenty, right? It sure is if all you’re concerned with is tweeting what you ate for breakfast this morning, but if you’re trying to compete against two Jeopardy champions, trying to beat them at their own game, you’re going to need a lot more.
Try 2,880 cores.
Let that sink in for a moment: two thousand eight hundred and eighty eight cores.
That’s the kind of processing power you need to call upon to even approach the raw capabilities of the human brain. Shame we tend to waste all that brainpower watching Dancing With The Stars.
(A more in-depth look at Watson’s genesis will air on PBS’ NOVA on February 9.)
All of the data Watson will ever know is present on more than 200 million locally stored “pages,”; Watson isn’t merely a fancy way to access Wikipedia. (IBM wouldn’t give the precise storage capacity of Watson, instead using the old “millions of pages worth of data” metaphor.) So when Watson faces off against ex-Jeopardy champions Ken Jennings and Brad Rutter (remember: the actual shows will air in mid-February; we’ll remind you when the time comes), that’s it—he knows what he knows. Watson will sit there, interpret Alex Trebek’s questions as he asks them (good thing Trebek speaks English because that’s all Watson understands), then do his best to answer as quickly as possible. He won’t second-guess himself, either: if Watson knows the answer Jennings and Rutter might as well exit stage left. Watson is unstoppable.
IBM has stressed how difficult it was to create Watson. Another famous IBM AI, Deep Blue, beat famed Russian chess champion Garry Kasparov in 1997, but Deep Blue had one over on Watson: chess is far more math-based than Jeopardy. Computers are pretty good at math. Deep Blue never had to know how many hit singles Michale Jackson had; it never had to know when the last time the New York Mets won the World Series; it never had to know when the first iPod came out; it never had to know when the War of 1812 was. Deep Blue had chess—math— running through its veins, so to speak, while Watson has to have the sum of human knowledge at his fingertips at all times.
The War of 1812 occurred in 1812, by the way.
And it’s not as if Watson will be going up against your local pub trivia champions like it’s an episode of The Office; he’ll be going after the all-time Jeopardy champions, Ken Jennings and Brad Rutter. Jennings holds the all-time appearance record of appearing 74 times in a row on the show back in 2004-05, while Rutter holds the all-time money earned record of $3.25 million.
Should Watson win it will represent a triumph of AI over man, yes, but it will also represent a triumph of mankind for mankind. Think of it this way: a team of researchers, flesh and blood just like you, came together to create a piece of software running on stock hardware that was able to out-think, out-know two Jeopardy grand champions.
It’s pretty wild.