[ExI] Watson On Jeopardy

[Kelly Anderson]

On Tue, Feb 15, 2011 at 2:41 PM, BillK <pharos at gmail.com> wrote:
> No, because Watson doesn't have time to do any learning or
> optimisation while the game is actually in progress. Watson doesn't
> take any notice of opponents answers. That's why it gave the same
> wrong answer as an opponent had already given.

On the NOVA show, Watson made the same mistake (giving an answer
already given) and the programmers talked about having solved that
problem a bit later. I would *guess* that the mechanism they used
somehow violated the rules imposed by the Jeopardy producers. It seems
like it would be an easy fix if they had a speech recognition
algorithm feeding back into the system, but they don't have that
capacity (yet). Alex T indicated that Watson wasn't "listening" in the
first show.

Again, according to the NOVA show, Watson does have a module that
learns during the game, related to the interpretation of the Category.
I did not get the idea that the real-time learning was very
sophisticated or extensive.

The IBM materials on DeepQA indicate that there are a number of
modules making up the architecture. In other words, you can plug in
new algorithms. On the NOVA show they were talking about plugging in a
"Gender" module. I would think that each of these modules contributes
to an overall score for a good or bad answer. The Spam Assassin
algorithm works like this, and it wouldn't surprise me if DeepQA used
a similar approach.

-Kelly