English Lessons; Jeopardy-winning computer aimed at Holy Grail of AI
By JASON MAGDER, The Gazette April 8, 2011
First came Jeopardy. The next step may be Star Trek.
One of the designers of IBM's Watson super-computer said the goal for the next version of the machine is to seamlessly understand human language and answer questions, like the ever-present computer that responds to the commands of Captains Kirk and Picard on the Star Trek television series.
"That's the holy grail of AI (Artificial Intelligence)," said Aditya Kalyanpur, a researcher at IBM's Thomas J. Watson Institute.
Kalyanpur, one of the engineers who designed Watson's brain, said the computer has achieved a remarkable breakthrough in the ability to answer questions phrased in complex English. It paves the way for a computer that would understand all human phrases and produce answers to any question, by sifting through huge stores of data.
Watson was designed to play Jeopardy - as a way to showcase advanced language analytical technology and problem solving. In February, the computer faced off against two of the greatest Jeopardy champions in the show's history and won cleanly. But it was a lot of work to get to that point. Kalyanpur said the first version of the computer took two hours to produce just one answer. During the quiz show, the average response time was three seconds.
"People have been really amazed," said Kalyanpur, who spoke Thursday at the Palais des congrès as part of the Crystal Ball conference on IT, organized by the Centre de recherche informatique de Montréal. "Jeopardy was actually a really good demo of this technology. It was a way to engage the average person so they can understand what the challenges are with language recognition."
He said the computer's greatest achievement was to be able to analyze human wordplay, and figure out what was being asked in the tricky wordplay used in the show. The computer then had to determine how sure it was of its answer, so that if it was not so confident, it would not buzz in and risk answering wrong and losing money.
Now that the Jeopardy experiment was a success, Watson is looking for work, Kalyanpur said. The supercomputer has been touted as a tool to help doctors make accurate diagnoses.
"We're looking at a host of different application areas," Kalyanpur said. "Health care is one big one, but also enterprise intelligence, government, legal and tech support. Any domain where there is a lot of information that is unstructured, and you want to make sense of that information.
"We're still working out the mode of deployment. We're working on making a miniaturized version of Watson. Something else that might work out is a cloud model. Where all the hardware is in the cloud, and the user just has access to the question answering and the text analytic capability."
jmagder@ montrealgazette.com
One of the designers of IBM's Watson super-computer said the goal for the next version of the machine is to seamlessly understand human language and answer questions, like the ever-present computer that responds to the commands of Captains Kirk and Picard on the Star Trek television series.
"That's the holy grail of AI (Artificial Intelligence)," said Aditya Kalyanpur, a researcher at IBM's Thomas J. Watson Institute.
Kalyanpur, one of the engineers who designed Watson's brain, said the computer has achieved a remarkable breakthrough in the ability to answer questions phrased in complex English. It paves the way for a computer that would understand all human phrases and produce answers to any question, by sifting through huge stores of data.
Watson was designed to play Jeopardy - as a way to showcase advanced language analytical technology and problem solving. In February, the computer faced off against two of the greatest Jeopardy champions in the show's history and won cleanly. But it was a lot of work to get to that point. Kalyanpur said the first version of the computer took two hours to produce just one answer. During the quiz show, the average response time was three seconds.
"People have been really amazed," said Kalyanpur, who spoke Thursday at the Palais des congrès as part of the Crystal Ball conference on IT, organized by the Centre de recherche informatique de Montréal. "Jeopardy was actually a really good demo of this technology. It was a way to engage the average person so they can understand what the challenges are with language recognition."
He said the computer's greatest achievement was to be able to analyze human wordplay, and figure out what was being asked in the tricky wordplay used in the show. The computer then had to determine how sure it was of its answer, so that if it was not so confident, it would not buzz in and risk answering wrong and losing money.
Now that the Jeopardy experiment was a success, Watson is looking for work, Kalyanpur said. The supercomputer has been touted as a tool to help doctors make accurate diagnoses.
"We're looking at a host of different application areas," Kalyanpur said. "Health care is one big one, but also enterprise intelligence, government, legal and tech support. Any domain where there is a lot of information that is unstructured, and you want to make sense of that information.
"We're still working out the mode of deployment. We're working on making a miniaturized version of Watson. Something else that might work out is a cloud model. Where all the hardware is in the cloud, and the user just has access to the question answering and the text analytic capability."
jmagder@ montrealgazette.com
Read more: http://www.montrealgazette.com/entertainment/Superbrain+Watson+looking+work/4579446/story.html#ixzz1J79t5bvC
No comments:
Post a Comment