Tonight IBM’s Watson goes against the Jeopardy stars Ken Jennings and Brad Rutter.
The system is pretty amazing with the amount of technical aspects that happen in 3 seconds for it to recognize the speech and also come up with an answer.
Some technical video on it:
http://www-943.ibm.com/innovation/us/watson/what-is-watson/a-system-designed-for-answers.html
my DVR is already set
lol man vs computer on a trivia show on Valentines day. I wonder what their target demographic is?
it’s on tomorrow and Wednesday too, Fry. Besides, Valentines day is a crock of shit. Man Up.
interesting… perhaps i’ll watch.
---------- Post added at 04:59 PM ---------- Previous post was at 04:59 PM ----------
What time and what channel?
you cant be serious?
LOL, they tied! i was so impressed at first, then like all of a sudden it became a dumbass
It was pretty impressive. Near the end they definitely thought they had like a 90%+ confidence and was wrong but then again, the human mind does the same thing.
I hope they really break it down somehow at the end of this and discuss their thoughts, changes, etc. Its a really cool project and something that be a major change in the world.
The technology, algorithms and computing power is amazing. That said…this is like a IBM infomercial, nothing more. I won’t be recording the remainder. What’d you guys think?
Nothing more than an infomercial? Without them explaining what’s behind Watson, it would be boring as shit.
I think the end of the first show was interesting. Watson started to suffer with the more difficult questions. I think he’s going to get crushed in double jeopardy.
Also, I think it’s unfair that Watson is being fed the questions via text file. I think it would take the human contestants WAY longer just to read the actual question. Watson is probably half-way through his algorithms by the time the human even finishes reading. Not really any way around it though.
The most interesting ones are when Watson doesn’t get the question right at all. The most interesting answer was when the correct response was supposed to be “What is a missing leg?” and Watson responded with “What is leg?”. It shows that Watson was onto the correct answer, but that the word association was a little off, and he just couldn’t get the correct response by the rules. It really shows just how hard it can be to compute the responses to the clues.
Ya I am glad they talked about how this thing works. Its not a infomercial at all. Infomercials try to sell you something… this is a show case of how much computing has come. Think Top Gear of computing.
To think about getting information and answers is one thing but Jeopardy you need to form the question based on the answer and how they analyze each word and learn what words go together usually to best form the answer. Also, based on the results it found for possible answers, it built a confidence level for the answers and would only buzz in if it broke the level. One of the coolest parts was when they were trying to form recognition between words and talked about how do you tell a computer that ran in the president race and ran in the olympics are different forms of run.
I was wondering about the text file too. It is broken into parts as its being read or is it given to them after the question is read.
The whole non-dairy creamer example was kind of funny too. It really showed how difficult it was to teach watson about the differences in words and how to view the data based on those annoyances. It was really funny when they all laughed at the answer, definitely funny.
yeah, I’m curious about when the text file is pushed in
It was interesting. I’ll be watching the next two nights.
The one that surprised me was how they apparently did zero speech recognition. Ken answered incorrectly and then Watson buzzed in and gave the same wrong answer. If I can speak queries to google via a phone app I’m sure they could have programmed it to listen to the competitors answers so it doesn’t repeat something that was wrong. It already has logic programmed to buzz in after an incorrect answer.
My only thought on this is that its not easy to do since they are people interacting. They would need to teach it what voice to listen to (Alex, Brad, Ken, etc.) and then also differentiate between someone stumbling over a answer by say “ah…”, “Oh i dont know”, “i think…”, etc. Very rarely do they buzz in and say the direct answer so it would be pretty complex and take away from the actual computing going on inside Watson for answers alone.
What was the non-dairy question/answer?
clue was something like “This is the most popular non-dairy coffee creamer”. Watson went with “What is milk?”. Obviously it ignored ‘non-’, the correct response was “What is coffee-mate?”
That was funny. I know the coding that goes into the algos has got to be insane so I’m not being critical. I guess without doing speech recog there would be no way for it to know the answer was already given.
The real limitation is human code really. Computing power has exceeded the human ability to program and take advantage of it for quite some time I would say. One would think they would have accounted for repeating a wrong answer…but I suspect they thought it would have far less wrong answers and it would be a non issue.
this thing is pullin away like a freight train
watson FTMFW…
