In 1997, it famously had its Deep Blue supercomputer take on former world chess champion Garry Kasparov but this time it’s switched to a possibly less Olympian form of human recreation: the long-running TV quiz game Jeopardy.
If you don’t know it, the game is actually much trickier than you might think. You don’t answer a question with a straight answer, instead have to work backwards from answers to a question form, thus "5,820" should prompt you to answer "The number of feet in a [US] mile". Plus, the range of questions is pretty broad, running from history to pop music to sports.
Now in its 47th year, the show is one of the most popular on American television and so it’s a much more populist bit of IBM outreach than grandmaster chess. At ‘stake,’ apart from all the PR, is a $1m prize, though what the computer, ‘Watson,’ will do with the money if it wins is anyone’s guess.
It could be argued that this isn’t that interesting a test – if the array is connected to the Web, all it needs to do to beat its two human competitors on the broadcast is a quick database search, right?
Well, IBM says that the AI (artificial intelligence) tested is quite a different form from that or even the kind of deep algorithmic knowledge Deep Blue had.
Instead, it’s about adaptation and flexibility and a test of the way the software can handle the speed of response required (sub three seconds) and the subtle nature of the clues and the lateral thinking you may need to solve them.
Thus Watson is coming to the event, which is on all this week on US telly, stuffed full of analytics software and a 15 Terabyte database of trivia.
That sounds a lot, and indeed is the equivalent of 200 million pages of text/a million printed books – but then bear in mind that Jeopardy attracts some high-power, college professor level contestants, so it’s a reasonable amount of knowledge to try and give it.
(The guys it’s competing with are champions of this peculiar past time, by the way, and are no patsies, with one having racked up the highest number of consecutive victories and the other being the highest money winner, at $3m.)
I went to see the Kasparov-Deep Blue event in NYC and to be honest was underwhelmed. At the end of the day, it was a dude in a room across a chessboard from a guy with a monitor. This time, IBM’s upped the ante, filming the thing in a lab with a crowd of excitable Big Blue-ers in tow, and enough of the US media are excited for the thing to get some attention on prime-time.
By the way, we were joking earlier about what happens to the winnings – IBM will donate every cent of any prize money to charity, which could be as much as the full million, though there are second and third prizes of $300,000 and $200,000 respectively, and the human contestants say they’ll do the same with 50% of any winnings they get.
So – interesting. Let’s see if IBM analytics can do something if not massively useful then at least suggestive, here.
