Wednesday, February 16, 2011

Artificial/Natural Intelligence

Since my wife and I are nerds, the cool thing this week wasn't Valentine's Day (although I did cook her a nice candlelit dinner), but the heavily hyped Jeopardy match pitting human heavyweights Ken Jennings and Brad Rutter against Watson, the computer system built by IBM.  After two games, Watson pretty much wiped the floor with the human competition, although the second day was much closer.  It was a little shocking how easily it could beat them, although to be fair both humans were also competing against each other so it may not be a true mano-a-mano contest.

Still, this is much more impressive than the Deep Blue victory over Kasparov.  Chess is amenable to brute force, algorithmic strategy.  Even Go, a game renowned for its intuition and positional play, is slowly succumbing to computational play.  In this case, it's impressive that Watson is able to parse and understand human language, particularly English with all of its idiosyncrasies and all of the puns and wordplay in Jeopardy.  You can start imagining a future where you can simply talk to a computer instead of interfacing with it through less than intuitive commands.

One nice thing about the episodes is that they showed the top three answers that Watson is considering, along with the confidence levels it has for each, which itself is pretty innovative.  It's a nice insight into how Watson "thinks" about the answers.  I'm also impressed that IBM still has such a strong research division and enough funding and management support to pursue this types of projects.  It's only a company that believes in research that can really kick start innovation like this.

On the other hand, I think claims about artificial intelligence and worries about Skynet are overblown---Watson is essentially a sophisticated database lookup.  I just don't see real "intelligence" in computers yet.  I've always thought predictions of artificial intelligence are exaggerated.  It's similar to claims about nanotechnology.  I've started doing some more work in the nanotechnology field, so I've started seeing some of the actual science close up.  Although I'm not a computer scientist, I am a chemist and I find that the people making the most predictions about nanotechnology tend to not be chemists.  People like Ray Kurzweil are notorious for this, with claims of the technological singularity approaching or dreams of space elevators.  Having worked with nanotubes and nanoparticles, there is no way we're making a space elevator any time during our lifetime.  I also have similar reservations about true artificial intelligence or predictions of uploading our consciousness to the internet.  Despite all of the huge scientific leaps we've covered, I think people just don't realize just how crude our tools still are.

In any case, Watson isn't completely omniscient.  At the end of the first game, the Final Jeopardy category was US cities.  Watson answered "Toronto".  I can imagine everyone at IBM smacking their heads at that moment.

Anyway, some cool links:

You can play a simulated game of Jeopardy youself at the New York Times.  I beat Watson 40-10!

You can also watch an episode of Nova that gives a behind-the-scenes look at Watson.

No comments:

Post a Comment