Watson the IBM computer on Jeopardy?

Last night through wednesday Watson is going up against Ken Jennings, who won 75 jeopardy games a few years ago, and other top players on jeopardy. Watson is wired to 15 some trillion bytes of memory and like 10 750 powered servers; however, Watson actually got some answers wrong, or wasn't quick enough to buzz in before the other players. I don't know exactly how algorithims work, I think it's like a code or somehting, but can a computer not answer certain phrases or idioms? Is Watson rigged to allow the other contestants to answer questions because I noticed last night that Watson tied with another contestant both with $5000 and Ken Jennings lost with $2000. How does this compare to the computer game of jeopardy on a regular disc, because when I play Jeopordy at home or when my dad plays it, and we don't buzz in and the computer buzzes in, sometimes the computer contestants will get it wrong on the CD Rom game, and I think that is rigged or there is some kind of pattern on the CD Rom disc game? I don't know if Watson is actually rigged, or if I'm using the right words, but why did Watson get certain questions wrong during the show, and other times the contestants were able to buzz in before Watson? One time, Watson said the same answer as Ken Jennings did, and they both were wrong. I'm thinking Watson is like a speech recognition program or text to speech recognition program on a computer because when ever you talk into a speech recognition program to give it a command on the computer it will not always interpret what you say correctly. I'm far from being a programmer, I have decent knowledge about computers, I don't really know how to use programming language at all, so any info on how Watson works on Jeopardy will be helpful. These are just my opinions.Watson the IBM computer on Jeopardy?
Watson cannot recognize spoken words, that is why watson repeated the same wrong answer after Ken Jennings said it, even though it thought it was the right answer. It receives the question, in text form, as soon as Alex starts speaking it, and goes from there to interpret it and answer it. If watson does not think it has the right answer, meaning not a high enough percentage, it wont ring in or will be slow to ring in. Watson does have to physically ring in, just like the other contestants, by using a piston to press down on the buzzer button. The unfair part, from my point of view, is that a machine using a piston is going to be able to push down on the buzzer quicker than a human can.
Your game is designed to get questions wrong, it wouldn't be fun if it didn't. The game is programed with the right answer so if it wasn't programed to get some wrong it would be right every single time with a perfect reaction time. You would get mad and quit.Watson the IBM computer on Jeopardy?
you should be able to find something here http://www-943.ibm.com/innovation/us/wat鈥?/a>
1) Watson may just not have the answer in its database. In that case it won't buzz in. It can also be playing against a really fast human player who once in a while just buzzes in immediately because he's sure he knows almost anything in the category.



2) Watson can't really get an answer (okay, a question, if you want it by the language of the game) "wrong", but it can parse the question (the "answer") incorrectly, coming up with the correct answer - to the wrong question. Garbage in, garbage out.



Watson beats the other players in buzzing in more than 70% of the time. The best human player runs around 60%.



"I don't know exactly how algorithims work, I think it's like a code or somehting" Well, the algorithm is written as a code, yes. How well Watson does depends on how well that code is written. If the code were perfect, and Watson has an unlimited database, it would buzz in first 100% of the time, and always be right. the code isn't perfect, and Watson's database isn't unlimited.Watson the IBM computer on Jeopardy?
I don't know how Watson actually works. I would suspect that occasionally, it would not be able to parse the question well enough to get an answer with any sufficiently high degree of certainty, and in such cases would step back and allow the other contestants to answer the question (assuming that giving a wrong answer carries a greater penalty than giving none at all; I'm not too familiar with the rules of Jeopardy). Other times, it may parse the question perfectly, but may be so nuanced that only a very high search depth would narrow down the results to the point where it could provide an answer. There is nothing INHERENTLY impossible about a computer understanding human idioms; the issue is just that idioms, metaphors, puns, etc are highly arbitrary and do not fit well into the kind of structured grammar engines and knowledge bases that are easy for programmers to create.
There is a PBS program that was on last week that explains a lot about Watson. It does take some time to mull through its trillions of bits of data and it doesn't always take the most efficient path. Also it won't buzz unless it has a high percentage of confidence in its answer -- which often was the right answer. It isn't programmed with "answers", but billions of references and relationships. I don't agree with their approach (more brute force in computing speed than efficient logic), but it seems to get results.