xiphias: (Default)
xiphias ([personal profile] xiphias) wrote2011-02-16 08:02 pm

On Watson on Jeopardy

In case you haven't see the results yet,

I bet Brad Rutter and Ken Jennings had some pretty mixed feelings at that match.

In order to be a Jeopardy champion, you have to have a good competitive streak, so they can't have liked losing.

On the other hand, in order to be a Jeopardy champion, you also have to be a pretty serious geek, and, to a pretty serious geek, a computer that can win at Jeopardy is really freakin' cool.

[identity profile] ykats.livejournal.com 2011-02-17 02:40 pm (UTC)(link)
There were a few occasions where Watson's confidence was above 90%, yet it lost the timer. I think human contenstants of that calibre are trained to anticipate the buzzer, based on audio and visual cues from Alex. They don't just wait for the signal.

[identity profile] xiphias.livejournal.com 2011-02-17 05:28 pm (UTC)(link)
My guess is that, in those cases, Watson was still processing until a millisecond after the signal to buzz in. Just a guess, of course.

[identity profile] ykats.livejournal.com 2011-02-17 05:44 pm (UTC)(link)
That may have always been the case. After all, humans press the button while they continue thinking, and even take some time to think while they start to answer. It's only logical for a computer of that scale to press the button even when it did not fully finish computing, winning a second or two, let alone milliseconds.

[identity profile] xiphias.livejournal.com 2011-02-17 08:33 pm (UTC)(link)
Ken Jennings talked about that in an interview -- Watson didn't buzz in until it had a confidence level high enough to justify it, while humans do buzz in before they've completely figured it out, sometimes. He was talking about how that was part of his strategy against Watson.