Interesting. In the last piece I read on this, IBM wasn’t saying how the trial runs went.
Does anyone know if there’s a voice recognition front-end to the NLP (natural language processing), or are the “answers” delivered as text? The human contestants are given 2 modalities of input - hearing and reading - where Watson needs only one. There are subtle timing issues in that distinction as well.
I think Watson should be made to hold and press the Jeopardy buzzer too and write the Final Jeopardy answer on the monitor AND right after the first commerical break have some extremely lame personal story that splatters any coolness that may have been achieveable. Ill-fitting, unflattering, poorly chosen clothing should be required too.
Neither article really said, but I assume that Watson has speech-recognition capabilities, since Watson supposedly “can get jokes and have fun”. That’s what Rutter said the IBM guys told him. (That’s in the second article, last bullet point.)
According to the second article, Watson does have a mechanical attachment to press the buzzer, to help “level the playing field”, so to speak.
Watched the first video in the second article of Watson in action at the practice at IBM.
Good gravy, he actually correctly answered a question about the BIBLE!!! I thought that those IBM guys would have somehow… “forgotten” that part. But I might just be being cynical. I suppose they wouldn’t want to leave anything up to chance with this.
As an afterthought, aren’t you glad Watson’s voice doesn’t sound like a Dalek?
Watson listens to the questions in real time with a microphone, processes the data in its main memory, comes up with several probable answers, picks what it thinks is the best answer, and hits the buzzer with a pneumatic finger, and gives the question response to a statement given by the show host.
Another project that might be of interest is the Google poetry translation project. If they can ever get the antilogarithms to put E.E.Cummings work in Spanish, I’ll be impressed.
Thanks, Will, and thanks for that link, ID. The input method is still not clear to me. The second article says, “The Jeopardy questions hit Watson’s chips as soon as the human retinas get the data.” Which implies textual input as the equivalent of presenting the humans with cues in visual form. And that’s good if so … Because every player knows that you can read the cues far more quickly than it takes for Trebek to speak them, and often have the answer (er, question) even before he finishes speaking. That’s the reason for the artificial button lock-out until the verbal cue is done. Requiring VR input on top of the NLP, with problems of disambiguation etc., could put Watson at a significant unfair disadvantage. And it’s pretty clear that IBM’s motivation for the stunt is to promote their back-end knowledge engine, not VR (though the tech underlying ViaVoice is quite good). Of course mike input, and maybe VR matching of the text, is needed to identify with the spoken cue is ended.
It sounds like the back-end incorporates techniques from what we used to call a Blackboard system in AI - a central module to adjudicate among proposed solutions from different, competing (and possibly parallel) expert systems or knowledge sources.
Heck, why not? Or the Koran, Rigveda, Avesta, Huck Finn, The Hobbit, War and Peace. They’re all just books, and knowledge sources.
Just a reminder that Watson should be going on Jeopardy on February 14, 15 and 16th. If you want to watch him go on the show, might want to start making plans.
This was my favorite: Watson has lots in common with a top-ranked human Jeopardy! player: It’s very smart, very fast, speaks in an uneven monotone, and has never known the touch of a woman.
I thought that those IBM guys would have somehow… “forgotten” that part. But I might just be being cynical. I suppose they wouldn’t want to leave anything up to chance with this.
