Last nights game featured IBM’s “Watson” computer holding its own.
I was floored by Watson answering complex questions not written in natural English. Many of the questions use words with a double meaning.
So far we are safe. Our brains are less than 1/2 the volume of a single server, run quieter and require less heating * cooling and interface with far more systems.
In terms of social media “Watson” starts to pass the Touring test. “Watson” answers and responds just as a human. Using Twitter it would be hard to tell “Is it live or is it memory chips”
Would you like this for eGov portal? As a friend?
Watson computer: http://en.wikipedia.org/wiki/Watson_(artificial_intelligence_software)
Touring test: http://en.wikipedia.org/wiki/Turing_test
Quick update – The show is on tonight and tomorrow night.
Picking the daily double on the very first try? Hmmmm
Yes that was an odd choice. It was the $800 question in a un touched catagory.
Hopefully they address that in tonights episode – along with another behind the scenes look.
I did watch Jeopardy last night, and I am looking forward to the remaining episodes. It is very interesting to see it in action.
Yeah I can’t help but think this was a little rigged. I mean Watson won by a massive margin.
Watson was pretty cool! Sterling Whitehead posted on All Things Sterling about Watson, saying that something similar could be used by Doctors to provide a diagnosis…criss-crossing relevant words with complex diseases that med professionals may not usually think of…I see this as a way to really reduce the cost of healthcare on the individual..if we don’t have to go through a bunch of worthless guessing tests…then why not?
I see Watson having a role in the acquisition and legal fields. Call it Watson, J.D. Customers ask Watson, J.D. for advice and out comes an answer. It won’t be as nuanced as a legal analysis; however, it would provide quick, reasonably accurate answers for customers when an acquisition or legal professional isn’t aroud.
As is always the case in A.I. exercises, the really interesting parts are the nature of the errors made. In the first show, Watson had difficulty with one category in particular that required it to group together events within a decade. The “Toronto???” answer last night was wrong on so many levels, though. The only thing right about it was that the confidence level displayed was low. Lest I gloat too much, I still think it is a pretty impressive accomplishment to have a computational device evaluate an answer as not particularly likely.
Another thing I think people forget is that Jeopardy is a game of speed as much as knowledge. Rutter and Jennings may well know as much or more than Watson, but where humans waste precious milliseconds prior to a button-press thinking ‘”Was it that one or was it that one? No, that can’t be right.”, Watson verifies against some algorithm, and hits the key. What we incorporate as “doubt”, or simply a check against feeling-of-knowing, takes far more time (perhaps because it involves self-statements) for us than it does for Watson. Did Rutter and Jennigs have the same answer at the ready 20msec too late? We’ll never know.
Finally, if you have ever developed a multiple-choice test of any kind, you’ll know that there is always a question underneath the surface question. Indeed, some may feel that one of the reasons we so often see such profound adverse impact on multiple-choice employment tests is precisely because not everybody shares the same sort of understanding that would allow them to intuit the “hidden question” equally. That Watson could interpret what the hidden question was was impressive itself. Finding stuff is trivial. Knowing what you need to be looking for, THAT’s hard.
I wonder if this A.I. work could help us by helping to point citizens to applicable programs, laws, forms, government offices, etc. ? I worked on website efforts that attempted to provide “no wrong door” into government, kind of like those greeters at DMVs that try to help people select the proper line for their transaction. Given that more and more services are online, perhaps this type of A.I. could develop into better assistance in matching people to their needed government agency or service.
Watson almost lost to Ken Jennings in today’s episode… in other news Ken Jennings will be playing the role of John Conner in the upcoming Terminator movie.
You have the right idea. Most people are looking for information – what time the store is open, where do I get a permit, what is required for a permit.
Most websites and IVRs are “descision tree” based. Press 1 for English, then Press 2 for store hours, then press 3….” The AI would far more social. You could Twitter your question and get a response.
By asking in Greek, French or German no one would have to “Press one for English”
+1 to Stephen’s comment – that made me laugh out loud.
As for Watson on Jeopardy, my one complaint was that I feel like he was winning more due to the fact that he was a machine and was programmed to be able to buzz in at the precise moment when he could beat Ken/Scott to the punch. I think that’s what was frustrating to me (someone who wanted to see Ken win) was that Watson didn’t necessarily know any more than Ken did – he was just able to buzz in more quickly. I think that’s an area where Watson had a distinct advantage over the humans.
Rumor has IBM adding voice recognition to Watson.
@Steve, Watson may not have known more than Ken. However the ability to process the queston and retreive a good answer is impressive.
Speed and the ability to run for long hours – days or weeks are two advantages. IBM never said how fast Watson could be updated with new information. Watson might win today and loose tomorrow when questions are based on current events.
News reports the medical field is looking to use “Watson” in diagnosis.
IMO Watson, the spike haird blue man, may have a better bedside manor than “House” with just as much knowledge.
Could I have medical procedures for $1,000 please 🙂