Although it doesn’t have much to do with social networking (or anything to do with it, for that matter) I wanted to write this week about a rather remarkable achievement in computing.
Unless you’ve been living under a rock, you know that this was the week that the syndicated game show Jeopardy! aired a three-part tournament pitting IBM’s newest computer wunderkind “Watson” against two of the best human players ever, Ken Jennings and Brad Rutter. Watson (spoiler alert) won easily.
Don’t get me wrong: I don’t in any way, shape or form belittle IBM’s achievement with Watson (well, actually I do a little—read on). It is truly remarkable that Watson, unconnected to the internet, drew from a vast well of facts that ranged from Beatles songs to Art History (although it seemed, ironically, to not be able to parse questions about keyboards or U.S. Cities, both categories I would have guessed it would do well in). Without a doubt, the ability to parse Jeopardy! clues and come up with the correct questions as often as it did was amazing.
I was hoping that Ken and Brad would do a bit better. Before the show aired, I thought that perhaps the humans might stand a reasonable chance against Watson. After all, people have the advantage of first-hand contextual knowledge about a lot of subjects, they are much better at associative memory and they had the advantage of listening to Alex Trebek read the clues, so they could anticipate when they could buzz in rather than react to a signal that the clue was complete (as Watson did). I realize after I watched the show that Brad and Ken never stood a change, and in a sense the deck was stacked against them.
For one thing, all the questions were presented in textual format, easy for Watson to parse, analyze, and look up the answers. There were no “video” questions that would have required Watson to “see” a picture or understand what the clue presenter was referring to in a video clip. So, there was no complex visual contextual analysis for Watson to try to decipher. Advantage Watson.
Although Watson was indeed relying on a signal that it could ring in—rather than have to anticipate when Alex was done speaking—everyone has to realize that computers are lightning fast: microseconds (at most) after the signal was given, Watson was able to send the signal to actuate the buzzer to ring in. No human can match that speed, and a lucky button press might *occasionally* beat the computer to the punch but not very often. Also, if the human rings in early, their buzzer is locked for a fraction of a second: this is something that Watson was *incapable* of doing. And, if you think about it, they could have built in a “human-sized” delay into the button actuator Watson used. What do you think adding a few dozen milliseconds to the button press would have done to Watson’s success rate? Again, advantage Watson.
Another point: Watson doesn’t get nervous, tired, or frustrated. And while Ken and Brad were likely well rested and have nerves of steel, they very much appeared frustrated at times, which certainly might have had some effect on their performance. Once again, advantage Watson.
The final advantage is simply one of timing. IBM rehearsed Watson for months before the contest was taped, pitting it against IBM employees at first and then actual Jeopardy! Champions. Does anyone think that IBM would have suggested the contest if they weren’t *convinced* that Watson could beat the best players on the planet?
So congratulations to IBM and Watson (and don’t feel sorry for Brad and Ken, they were well compensated for a day’s work). The lessons learned from building Watson will impact all our lives in ways we probably can’t even begin to imagine.
People are still better than computers at lots of things. First and foremost is that we can build computers like Watson: computers can’t (yet). Arthur C. Clark once observed that “any sufficiently advanced technology is indistinguishable from magic,” and Watson is well on the way to proving that any sufficiently advanced Artificial Intelligence is indistinguishable from real intelligence.
No comments:
Post a Comment