Posted on 02/17/2011 12:46:37 PM PST by SeekAndFind
After conquering puny humans Ken Jennings and Brad Rutter and winning a total of $77,147 over three days and two full games on Jeopardy!, IBM's know-it-all new supercomputer is going to med school.
On Wednesday, IBM, along with Nuance Communications Inc. and the Columbia University and University of Maryland medical schools, announced that they are developing Watson as a diagnostic tool that can help doctors identify diseases and recommend treatments. They hope to begin lab tests as early as next year, with real world testing later in 2012.
"What makes Watson unique is that it can rip through massive amounts of information and give a small amount of possible answers with levels of confidence," says Dr. John Kelly, IBM's senior vice president of research.
Doctors have long relied on technology to help them manage patient care — electronically stored patient histories, digital lab results and machines that regulate medication are all commonplace in today's hospitals. Indeed, the first attempt to create a machine that could help diagnose human illness came back in the 1970s, when Stanford University researchers developed MYCIN — a computer designed to indentify different types of bacteria responsible for infections. But even the most up-to-date systems, which were developed in the 1980s, still require physicians to spend costly time typing in test data and patient information, and still only cover a limited number of diseases.
That's why doctors like Eliot Siegel, a professor and vice chair at Maryland's department of diagnostic radiology, says Watson's capabilities are necessary now. Imagine a supercomputer that can not only store and collate patient data but also interpret records in a matter of seconds, analyze additional patient information and research from medical journals and deliver possible diagnoses and treatments, with the probability of each outcome precisely calculated.
(Excerpt) Read more at time.com ...
Which brings me to my query.
At what point in the reading of the question was the 'enter' key pushed. If at the start of the reading, Watson would have a milliseconds advantage.
Or were they typed in as they became visible?
How about a computer for the thugs in government...
You put in everything they have done and said and then
have it compute if they have been “helpful” or “NOT”...
Hear the question.
Recognize that I know the answer.
Push the button.
Recall the actual answer.
So, humans could have an advantage in letting their thinking go on after the button has been pushed.
Reading is faster than hearing, even for humans. If the human contestants don’t get the questions in both written and oral/aural formats, it isn’t “fair”.
Who knew?
The double whammy of getting the answer wrong ( deduction for you, and lots of time for opponents to think of correct answer ) makes this a very bad strategy.
“Toronto”
What about mental illness. I hope it can recommend some effective treatments for librardism.
If they had allowed each player to click in and answer any question they wanted, without hearing each other, I'm sure the humans would have done much better. I would like to know how much better.
As opposed to Watson going through all the answers first and then calculating a confidence.
YMMV.
Yes - but Watson Junior couldn't be . . . Watson would only get better . . . if they were to continue development for Jeapardy! competition.An article in the WSJ today said that part of what it would take for Watson to pass a Touring test would be that it would have to be dumbed down a little, in some ways, because actual people don't know everything.
I was shocked when Watson guessed "Toronto" in final Jeopardy. Seems like it would have been able to find the cities w/ more than one airport, and compare their names with the names of WWII heroes and WWII battles. That sounds like a pretty short list to search. No real subtlety in that.
That's a basic filtering error, and I saw Watson exhibit this kind of error in at least three different contexts. It's odd too, since this kind of filtering problem was solved by elementary search/selection algorithms, several decades ago.
I would expect this to be a tool used by doctors, rather than a way for patients to bypass doctors. A doctor would know how to normalize the inputs to get more accurate results.
This is a failing with most Jeopardy contestants. Few have any understanding of game theory, and it applies in Daily Double and Final Jeopardy bets. Watson seems to have been given separate game theory programming, for making these bets, and it was done correctly.
This is why these bets looked like such odd numbers. An optimal bet, made according to game theory, will be a specific number.
I agree. The technology for doing that already exists, and it would have put things on a more equal footing.
Some questions in Jeopardy would seem to require two or more filtering operations. Questions of that type seemed to give Watson the most trouble. Watson was quite fast when given relatively simple fact questions (e.g. who wrote X).
One of the categories that real gave Watson trouble was ‘Computer Keys’.
One question I remember,(paraphrasing)...
‘A woman’s straight, loose dress’
ans. Shift
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.