ACt

So, in case you haven't heard, there's a little contest going on on Jeopardy over the next three days. IBM have designed a little Q&A system which after years of work has been tailored to answering trivia questions. And what better place to try it out than on Jeopardy.

So, from makers of Deep Blue, I give you Watson. The beginning of the end of human independence.

I kid, I kid.

Anyway, essentially it is a computer with programming that takes in information and then, while focusing on key words, searches the net to come up with the correct answer. Surprisingly, the fact that the answer is a question doesn't bother it. Huh. Makers were trying to get under a 3 second response time, and feel they have done so. And that's my base understanding. I think Mal or Axiom might have more to say on this because it is encroaching AI territory.

The idea is to test the systems against Jeopardy grand champions and then probably use that as enough publicity to launch a new wave of search engine and patterning software on the world. Or, they'll win, dissemble it and not let anyone see the man behind the curtain. Which is what they did with Deep Blue, leading Kasparov to expand on conspiracy theories that the machine was helped by humans. (See: Game Over. Movie about chess and the Kasparov Deep Blue competition.)

Anyway, it should be rather interesting. I most intrigued to see if the Jeopardy question writers have created any particularly tricky questions to try and defeat the program (which is what they should do). I think the Before and After ones should be really tough on it. Also, will the computer answer wrong? Or only ring in when it is 99% certain it is correct. I'm going to lean to the later because I doubt they wrote in a time limit, rather to keep searching as time allows until they are most certain. Finally, I wonder the advantage of the computer having near instantaneous reflexes to ring in while the others, though good, will have human reactions. That should give some of the easier questions to the computer.