Saturday, February 19, 2011

Watson Fights for the Users

Or at least we better hope it does, because I don't think we'd stand much of a chance against it.

To recap, Watson is IBM's new supercomputer designed to take on the best human players in the world at the quiz show Jeopardy! For specifics on how it works, you can refer to IBM's official website. Earlier this year, it was pitted in a two-game match against Ken Jennings and Brad Rutter, the two greatest (human) Jeopardy! champions of all time, in the first ever man-versus-machine competition in the show's history. The games were broadcast over three nights this past week, and Watson proved itself not only able to play, but able to dominate humanity's best. In the end, its total winnings were well more than Jennings and Rutter's combined scores.

It was the most significant development in the ever-evolving “man vs. machine” saga since IBM's Deep Blue trounced grandmaster Garry Kasparov in chess back in 1997. I was in eighth grade when that chess match went down, and I still remember what a big deal it was. Of course everybody recognized that computers could process faster than us. Even a weak computer could crunch numbers faster and longer than any human, but that sort of “intelligence” was one-dimensional. Chess, on the other hand, was a highly nuanced and intellectual game, long considered by some to be the finest and deepest test of complex strategic thinking. Champions such as Kasparov and Bobby Fischer had been respectfully regarded as geniuses, representing the pinnacle of mind. Thus, when IBM produced a computer that could outplay humanity's champion at its noblest game, it was almost as if to say that human intelligence had run its course, and any forward progress lay in the hands of machines. Chess matches played between “mere” human players, of any level, subsequently seemed to lose all significance.

Bringing things forward to today, many of those following the Watson story, fearing the larger implications of an AI once again making a mockery of human will and wisdom, hoped for a human victory in this real-life John Henry scenario. It was not to be. Not even close. But the results are hardly as conclusive even as the Deep Blue match. Jeopardy!, after all, is not just a test of one's knowledge but also one's finger speed. I've seen both Jennings and Rutter play before, and they are exceptional competitors, capable, on any given night, of running the board with displays of extraordinary breadth of knowledge. It is likely that, in this case too, either man could have provided as many correct answers as Watson, if not more. Watson just didn't give them the chances. As a computer, it can manage perfectly and consistently precise timing on buzzing in, which no human could ever hope to match. I think everybody was willing to concede that much coming into this, and if all this contest proves is that a machine can time a buzzer click more perfectly than a human can, well then that's not portentous at all. It's not even really news.

What is news is that IBM has created a computer that can play Jeopardy! at all. Were Watson merely a vast database of facts and a quick trigger finger, it still could not be regarded as intelligent, and it would not be enough to play Jeopardy! The achievement is Watson's unprecedented ability to actually comprehend the game and the questions, and to (usually) provide a correct answer in the appropriate form and in real time without assistance. It had its moments of weirdness, as more complex, multi-part clues sometimes tripped it up and prompted bizarre responses that were nowhere close to correct. But those moments, though not infrequent, did not keep it from competing on a level with the best players in the world. More often than not, it was able to understand the clues in their written, plain English forms, and that's impressive. After all, have you ever typed a query into Google and received in return thousands of results that do not appear in any way relevant? I certainly have, and, based on that, I would have thought that the Watson tech was generations away.

That said, we're still a long way to go from being able to carry on conversations with machines. Watson doesn't really so much comprehend language as analyze it. And it can only look up simple answers out of its database of facts. It can't provide deep reflections or produce new thoughts. It could still make life easier, much in the same way that a search engine does, but if anybody's seriously worrying (or hoping) that human culture and society will make way for robots, you can probably rest assured that we're not quite there yet. (Then again, a year ago I would not have imagined that we'd be here yet.)

No comments: