E-mail updates
Follow on Twitter
Subscribe to RSS
Back in 1997, IBM made history by fielding a supercomputer that beat world chess champion Garry Kasparov at his own game. For the past three years, the company has been working on a super-duper-computer to follow up on Deep Blue's triumph of the machine. Now the computer touted as the world's best question-answering machine, dubbed Watson, is almost ready for prime time. Or at least syndicated TV.
To put Watson to the test, IBM's programmers have been pitting the machine against human rivals for months. This time, the human-vs-machine battle isn't played over a chessboard, or even a poker table. The competition is in the form of a "Jeopardy" game, in which players have to buzz in quickly to provide the questions that go with answers displayed on a screen. For example, "In 2003, Mahmoud Ahmadinejad became mayor of this city." The correct response (stated in the form of a question!) is "What is Tehran?"
The test isn't just a game: Being able to provide answers to questions using natural language analysis is the multibillion-dollar trick done by search engines, voicemail robots and future artificial-intelligence systems.
This week, an article in The New York Times Magazine traces how IBM selected "Jeopardy" as the standard for designing a better question-answering machine, how the company's engineers designed and fine-tuned Watson, and how the machine can often trounce us puny humans.
One of the big tricks is to cross-check a list of possible answers against additional searches and see which answer gets the highest ranking. Which is kind of what I do when I'm using the Web to answer a particularly tricky question.
The producers of "Jeopardy" have promised to put the machine to the test on national TV as early as this fall, in competition with some of the show's best veteran players. IBM expects to sell the Watson question-answering package to institutional customers in the next year or two. But you don't have to wait that long to get an idea how Watson works. This New York Times interactive lets you play against Watson in a trivia challenge, and you can even pick which questions you want to answer.
I, for one, welcome our new question-answering overlords. I might even have one of them attached to my brain one of these days. But what do you think? Feel free to weigh in with your comments below.
The YouTube video at the top of this item was produced by IBM. Join the Cosmic Log corps by signing up as my Facebook friend or hooking up on Twitter. And if you really want to be friendly, ask me about "The Case for Pluto."
Leave a Comment:
Answer: A supercomputer named Watson.
Question: What is the result of information apathy in the human race.
We are in a headlong rush to a future no one can predict and we are moving there faster and faster. New things are coming up faster than we can begin to understand them, much less debate them. One way to look at it is we are in the same situation as the native Americans were when confronted with European culture, technology and diseases, except this alien culture confronting us is us ourselves invading from the immediate future.
I, for one, am a bit of a doom-sayer. I honestly believe we will one day face a robot apocalypse.
Yes, but will they be able to finally answer which came first, the chicken or the egg?
It is interesting to see this machine in action...I'd like to watch Watson in action, for sure!
I wish I knew what answers were given that Watson 'answered' (or questioned) incorrently. It's still amazing to see computer technology to come this far.
how does it fare in the turing test?, deep blue was tweaked over a long period to beat a human chess player, but it could easily be beaten by a dog (simply because dogs can't play chess, to wait expectantly for the dogs move is an admission of it's lack of intelligence)....somewhere there is a product in all this, webbots already serve commercial functions. The need is surely there for a program that could answer the incoming message and query the sender in order to set it's priority, ie know to delete the sales call for the unwanted widget but put the call for free concert tickets from a freind RIGHT thru!!...."What is free Aerosmith tickets for saturday night alex?, GREAT! one minute whilst I connect you to the human"!!
Interesting different tests. On the one hand, deep blue had 64 squares and 6 types of units to handle, the method by which it was presented with a problem was very simple and computer friendly, but the problem to solve was very difficult, involving advanced strategy and I imagine even prediction of human behavior.
On the other hand, Jeopardy should present Watson with problems so simple that they're trivial. Put Wikipedia onto an N-gb table, spend a bit of time cross-indexing it, and you'll have any mere mortal heavily beat on general trivial knowledge. The difficult part must be understanding the problem and formatting a response, and I agree that that's a pretty interesting problem, it's just on the opposite end of Deep Blue.
Deep Blue did an incredible amount of lookahead, but it's brute force lookahead comes short compared the deeper and more focused looking ahead a Chess Master does. Where is shines is in it's "book", in other words memory of the actual play by masters in actual games.
It didn't come up with moves better than a human, it just remembered how the best humans played and did that.
That's not true.
After the opening (standard books have this) It systematically explored and pondered all options, that is what a master does. However a master will tend to stay inside the box of chess theory while big blue will come up with some crazy stuff..
candie, there were eggs long before there were chickens. QED.
"IBM expects to sell the Watson question-answering package to institutional customers in the next year or two."
Once again Big Blue is placing one of humanity's most widely useful achievements out of reach of most of its citizens. I'm still surprised that the IBM PC was sold to the general public.
Just think of the result of some child genius who discovers this system on the net early in their life. I am absolutely sure that their (self) education would rapidly outpace their classmates. Wouldn't that inevitably benefit society?
Interesting question! Would it?
Depends on which philosophy you buy into.
God knows parents would be happy..
Daddy why is the sky blue?
why do horses have four legs?
why, why why..
Basically every eight year old would wear themselves out...
The real wonder of this machine is the "Understanding" of the questions. Once the question is broken down in to a search term, getting the response is much easier. Google has been doing this for years and has revolutionized the Internet with it. The video makes reference to this when it talks about puns and jokes in some of the questions. It could be that the machine did not understand how the questions were spoken that lead to wrong answers. That will probably be one of the leading problems to resolve with Watson in the next few years.
IBM has long had an interest in understanding natural language processing. Given that most people don't really understand how to talk to a computer, thier only possible choice for working with them is the natural language interface. Most science fiction that depend on computers for plot use natural language interfaces as well as natural language responses. I see this as the real break through with this machine.
This raises another interesting question. How computer like are we willing to accept we are?
A computers response is nothing more than the sum of the logical rules and memory programmed into it. Does this apply to you and I too?
Even though our answers are based on logical rules and such. Unlike computers, our EMOTIONS play a vital role in day to day life. Thus computers may give the right answers all the times, but humans are able to give answers that will benefit the present time or the future
eg. Your daughter is dying, She ask, "Daddy will I be ok". A computer will give the real answer "NO!!!!!"
You as a father would never do that instead you'll say, " Everything is fine sweetie, You're going home soon".
Thus we give the right answer for that specific moment in time.
Might emotional responses also be programmable as just another set of logical rules that put situations into context?
Internet search based answers pose some interesting issues :)
I wonder how this would go..
"It is the only place in the United States that allows cock fighting."