http://www.washingtonpost.com/wp-dyn/content/article/2011/02/11/AR2011021107031.html It's coming on right now on CBS. Watson is on today, tomorrow, and Wednesday. Should be interesting to see how this thing does.
Part 1 <iframe title="YouTube video player" width="640" height="390" src="http://www.youtube.com/embed/BfNBWJTGEEA" frameborder="0" allowfullscreen></iframe> Part 2 <iframe title="YouTube video player" width="640" height="390" src="http://www.youtube.com/embed/TFe2pJETNuw" frameborder="0" allowfullscreen></iframe>
It was interesting. Some predictable flaws with handling the oddly worded questions on Jeopardy, but still pretty impressive.
It missed a question by repeating a wrong answer given by another contestant. There's no way for it to "hear" answers given by another contestant.
If there is a way to input the answer given by another contestant, that can fix that issue. I just don't think it would be possible in terms of time... and how fast Jeopardy moves.
Pretty amazing stuff. It's interesting to see the moments of difficulty that Watson has, but overall, I think it's can be classified as a major success so far. The sheer fact that it can process so many millions of pieces of information in enough time to buzz in and answer a question is mind-boggling to me.
I am wondering how much game theory they programmed into Watson in regard to which questions it picks based on other people's scores or how much it bets on daily doubles. Also since it can't hear other player's answers I'm guessing it can't steer the game into a category that it believes the other players will have trouble with. They should've also made Watson's avatar a red camera lens.
YOU IDIOTS!!!! You're all helping the machines!!!! Don't you get it?!?!?!? FATE. THERE IS NO FATE BUT WHAT WE MAKE!!! NooOOoooOOOoooOOoooooooooooooo!!!! Spoiler
I watched a PBS show (Nova?) on the making of Watson last week. It already tried out for Jeopardy once and got rejected (producers were extremely impressed, but it did not have a good track record in tryouts), but after another year of development it made the cut (producers were floored, it killed). It was incredibly entertaining to see the process Watson goes through, especially if you're a programmer. I had a nerd boner when Watson got almost every question wrong in one category. What it did not realize was the category wanted a month as a response. But on the last question it finally 'learned' and got the $1000 question right. IIRC I thought they were supposed to implement this in production. That's how it figured out the 'month' example I mentioned.
I'm blown away by this. I understand the HOW but still scared of computer overlords....it is exciting though for the future to see how this plays out, but every endgame scenario is the same with the computers eventually ruling us.
He (whoops, it) doesn't seem to preform as well on the high dollar questions, giving the humans a chance to catch up.
I'm impressed also, but I guessing it doesn't "think" per se. I think it just averages out all the common words associated with the words in the jeopardy "answer". For example, if you Googled "drunk, hotel room, cocaine". Your top hits would mostly likely include "Charles Sheen". I assume it works in much a similar way. So, don't expect Skynet soon.
