I don't get excited about the singularity the way some on Science 2.0 do (and certainly elsewhere) though I admire the optimism.    So when I got an email from a publicity person at PBS about NewsHour Science Correspondent Miles O’Brien’s report on the upcoming match between Jeopardy masters Brad Runner, Ken Jennings and the super computer Watson I had to wonder if this would lead to more claims about an upcoming singularity and exponential leaps in artificial intelligence.

Don't get me wrong, Watson is a smart computer, but not because of its horsepower, about 6,000 home computers, but because its language software makes it different from predecessors in the same way Apple's iPod interface was a leap over the hundreds of MP3 players that existed previously.   

But saying Watson can 'understand' language leads to the same confusion as the word 'existence' does in singularity debates.   David Ferrucci, Watson’s creator, and his team of two dozen programmers spent nearly 4 years writing the algorithms to teach Watson language skills so they like it is 'artificial intelligence' and, in a narrow way, I agree.

But I am not thinking 'artificial intelligence' like I am going to be uploaded to a machine by 2045, despite what Ray Kurzweil claims in order to sell books, I instead think that Watson's ability to understand language is going to revolutionize medicine.    Right now, if I search for something, Google (sorry, Yahoo and Bing, when you send even 1/500th the traffic Google does you can get first billing on occasion) tries to figure out what I mean.    Not a great idea for a doctor, much less a concerned patient.

Watson, as David Murphy at PC Mag notes, uses its 80-teraflops to sort through 200 million pages worth of information, which sounds a little brute force.   And human contestants know a lot also.  What Watson might do better than humans, in a game like this or in medicine, is never guess.  If it does not know an answer as a certainty, within 3 seconds, it is not going to try.   People, being people, will instead act on a hunch.    And Watson does well not guessing.   Here is a clip courtesy of Engadget a few weeks ago:



Imagine the power in medicine if a machine can figure out what is wrong with you, even given Watson's 85% success rate?    Sure, if you get a cancer or vasectomy diagnosis you want to have a human second opinion but for those obscure diseases "House" has to tackle, it could all be done much more easily, and without humiliating medical residents.   And the cost is non-existent once the information is available.

Watson has value and Watson is fun to watch but it's not completely over for humans.    Watson still has trouble with "fill in the blank" questions" which means he can rule at Jeopardy but likely get creamed at "Wheel of Fortune".

However,  that is no solace for Miles O'Brien, who gets creamed when he attempts to go up against Watson.



The things we do for science outreach.

Peter Pachal (also at PC Mag) sums up the whole thing nicely:
In any case, the era of thinking machines isn't quite here yet. But the era of thinking humans, aided by machines that understand them better, and can react faster than ever before, is about to begin. Watson-like computing won't fix all the world's problems, and it certainly won't destroy the world (sorry, Skynet fans), but when you've got a problem, there's a mountain of data to sift through, and you need an answer now, it'll be the machine equivalent of your personal A-Team.
The three-day man vs. machine match starting tonight pits the quiz show's two most successful players, Ken Jennings and Brad Rutter, against Watson for total "Jeopardy" dominance.    I won't be watching live, I'll be at a Valentine's Day dinner with the wife.  Unlike Watson, I can get a date.

Here is a Nova special on Watson you can watch until "Jeopardy" is on: