Watson and Jeopardy

[Lennart Sorensen]

On Wed, Mar 02, 2011 at 11:56:38PM -0500, Yanni Chiu wrote:
> IIUC, the technology/intelligence is to parse the answer/question, use  
> that parse to narrow the search of a huge knowledge-base, assign  
> probabilities to possible answer/question. Then decide whether the  
> probability is high enough to ring the buzzer. A "learning" phase is  
> used to "teach" the system how to choose the best answer/question.
>
> It sounds a lot like what humans would do to be good at Jeopardy. So  
> they've created a super Jeopardy machine. But IMHO, humans are still  
> capable of much more. It's kind of like computers are really good at  
> saving data and making calculations, compared to humans. Now they're  
> also good at recalling trivia.
>
> Funny that their envisioned uses of the technology is for medical  
> diagnosis. That kind of says that doctoring is an "easy" task that  
> computers could encroach upon.

Actually what they said is, it is hard for any doctor to read and remember
all medical literature.  One methods humans have invented to try to deal
with this problem so far is consulting with other doctors, or even more
like it the "Best Doctors" thing that is now being promoted on the radio
and such where complicated cases can be analysed by many leading doctors
over a wide area in case one of them recognizes something that is in their
specialty that the treating doctor isn't aware of.  Watson would be able
to potentially take all the medical journals and text books and case
studies and search them using the terminology doctors use to describe
the symptoms and return possible things for the doctor to look at.

> Maybe there needs to be a new challenge (or AI gauge) - namely, build a  
> machine that can build a really good Jeopardy player. Then, us software  
> guys/gals would be looking over our shoulders.

The problem then is "Please define your problem in a way that can be
implemented".  You statement is completely vague and ambiguous.  What is
"good" for example?

-- 
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists