'
'
'

Monday, February 14, 2011

IBM’s Watson Computer Will Win. I Know. I Played Against It.

Today’s the big day. IBM’s Jeopardy!-playing supercomputer Watson plays the first of three nights against the game’s two greatest champions, Ken Jennings and Brad Rutter, on ABC at 7 p.m. Eastern. The winner gets $1 million. The victory will decide once and for all whether a sizable computer can deracinate humans from a game only they (well, Merv Griffin) could invent.

How do I know Watson will win? Because I played two rounds against the “world’s smartest machine” last week and lost badly. The other human contestant was A.J., a private-equity megamind with a tweed blazer and multiple degrees from Ivy League institutions. Him I could handle. Watson, though, was OFF-THE-HOOK good. It claimed more than $30,000 in winnings each time. IBM said I was the only writer to have played against the machine. Ego-wise, a bittersweet scoop.

Sony Pictures Entertainment, which owns Jeopardy!, had better attend here. The last stunt like this was when another IBM supercomputer called Deep Blue beat Russian grandmaster Garry Kasparov at chess. Has anyone talked about chess since then? In fact Kasparov has been going around criticizing computers for not being smart enough. Well, tovarisch,tonight a new era of computing begins. The Trebek era is drawing to a close.

If you need a backgrounder on what Watson is and how it works, you can  watch this video or listen to this NPR radio program. The press is all over this Jeopardy! Challenge, and rightfully so. Such holy-crap innovation from an American company is just what we need at a time when everyone is concerned about America falling behind China and its ilk. Jeopardy! may seem like a silly application for such an expensive effort by one of the world’s largest technology companies, but IBM is testing a system that opens frontiers in software intelligence. Google is about search results. It’s up to you to pluck the best result from the list. With Watson, you ask a question (a certain way) and you get a range of answers with the thought process behind each answer, the confidence level behind each answer and all the relevant links to the reference used to derive each answer. This is something new.

There are many reasons Watson is good at Jeopardy!. It has something like a million pages of documents and a geospatial database in its memory. It can run the board on categories like “Northernmost Capital Cities,” in which you’re given a trio of capital cities and have to name the northernmost one. Watson doesn’t really “get” that the category is looking for geographical information off the bat. That isn’t a simple assumption to deduce from the category syntax, but it catches on quickly that all it has to do is compare the latitude of each city in the trio and give the one with the highest number. It buzzes in correctly on Stockholm, Bogota, Pyongyang, Algiers and Kathmandu. A.J. and I knew some of these, but that’s when Watson’s other superior trait kicks in; the machine is lightning-fast at buzzing in answers. The best Jeopardy! champions get to the buzzer first more than 60% of the time. Watson can sometimes push that buzz-in rate above 70% and has improved its accuracy every few months until now it lives in the same scatterplot chart region of speed and accuracy as a Jeopardy! champion. Only Jennings’ scatterplot looks better. But for mere mortals, if you can’t get even a crack at one-third of the clues, what’s the point of being there?

Watson is really good at the bottom of the board, too, those $2,000 plums that often stump humans. One was in the category “Saints Be Praised.” The answer: “This New Testament martyr is the patron saint of bricklayers.” Get it? I didn’t think so.  Watson did, answering “What is St. Stephen?” How the hell does it know that? Oh, I forgot, it has total recall of everything.

I am a fairly good Jeopardy! player. I made the show once. I lost with a very big lead going into Final Jeopardy by getting it wrong. The same tragedy befell me in the second game against Watson. (The first game, drawn from this January episode, we will not talk about. Suffice to say they had to lend me money so I could finish.) In the second game, based on this October 2010 episode, I had $16,200 before Final Jeopardy! Watson had $31,800. AJ had $8,000. Statistically, I could have won if Watson got it wrong and I doubled my bet with a correct answer. The category was Mammals. My mind riffled through the mammal files: Hair, milk, birth to live young, bats, whales, dolphins, chimps. Got it. I doubled my bet, minus a dollar. The clue: “One type of this aquatic animal gives milk that’s 65% fat; pups are weaned in 4 days, the least of any mammal.”

It wasn’t a whale, which is what AJ said. It had to be a highly evolved sea mammal. A porpoise, I wrote. No. It was a seal. A hooded seal to be exact. Watson wrote “What is a seal?”

How the hell did it know that?

After its Jeopardy! fame fades, Watson is going to get down to serious work. The IBM team led by computer scientist  Dave Ferrucci is already deploying Watson in health care. The same way IBM fed Watson Wikipedia, the Bible, a geospatial database–the equivalent of a million pages of documents–it has begun to feed Watson electronic medical records, doctors’ notes, patient histories, symptoms, the USP Pharmacopeia. Here’s the amazing thing: The machine is getting faster at learning. Teaching it to play Jeopardy at a championship level took four years. Teaching it to deliver reasonably accurate answers to diagnostic questions took only four months. I can see IBM selling Watson as a Web-delivered service to doctors and hospitals seeking answers to patient presenting with problems. Watson considers everything and creates evidence profiles (the types of information it relies on, weighted based on their reliability and utility) that feed into diagnoses graded on varying levels of confidence.  These can be offered up as charts on an iPad  showing a doctort Watson’s thought process. It’s like peering into the mind of a House, M.D. The doctors make the final call but they can assess possibilites they may not have seen and can click right to source material used to compile Watson’s answers. This is powerful stuff.