University Researchers Find Business Applications for Watson

After IBM Watson won a "Jeopardy!" competition against two of the show's major champions in February, the technology that powers the super computer has been getting more attention.

And now universities are researching and developing business applications for Watson to help improve the technology.

Through natural language processing or more specifically, deep question answering technologies, the computer analyzes complex questions and responds with answers in English. It even includes a pun detector algorithm.

In order for computers to interact with humans through language, we have to acknowledge the fundamental challenge that computers don't live in the human world, said David Ferrucci, the lead architect and principle investigator for the Watson Project at IBM.

"Ultimately we're training them to give a response that has meaning to us," he said at a symposium held at the University of Pittsburgh and Carnegie Mellon University on Wednesday.

Natural language is implicit, highly contextual, ambiguous and often precise. But computer programs are natively explicit, fast and exacting in their calculations of numbers and symbols. That's why this is such a challenge, Ferrucci said. However, the "Jeopardy!" competition gave the team a crisp metric to measure how well they were doing.

By training the machine in context, researchers learn what algorithms help the computer come up with the right answer. To get the right answer, they used many different methods instead of relying on one type of algorithm.

"We didn't assume any one method was the right way," Ferrucci said.
 

Universities contribute to Watson technology

A number of universities have been working with IBM to develop the Watson technology as well as build potential business applications for it.

Carnegie Mellon Professor Eric Nyberg from the Language Technologies Institute consulted with IBM throughout the Watson project. And his PhD students, Nico Schlaefer and Hideki Shima, both contributed algorithms and components. In addition to that, they worked together to develop an open-source architecture.



At the end of the symposium, Nyberg along with other experts from IBM, Carnegie Mellon and the University of Pittsburgh talked about Watson's potential for business applications. Currently, a team from Columbia University is working with the company on applications that would allow doctors to use Watson as a tool to make medical decisions.

In the symposium, the panelists stressed that the computer doesn't know everything, isn't designed to make decisions for humans and has weaknesses that lead to incorrect answers. But it can empower humans by assessing, synthesizing and organizing information, Ferrucci said.
 

Educational applications reveal potential of Watson technology

As far as educational applications go, Watson could be used as a computer tutor agent, said Diane Litman, faculty director of the Intelligent Systems Program at the University of Pittsburgh who participated in the panel. The university's interdisciplinary Learning Research & Development Center has been making investments for years in natural language processing that could be applied to education technology.

While most of the work in the field of intelligent tutoring systems hasn't been language-based, more people are showing interest in incorporating natural language processing into tutoring systems, Litman said in a phone interview. But due to technical limitations, the tutor generally controls the conversation and asks questions that it knows the answer to.

“Although everybody knows that student questioning is really important, from a technical perspective that’s much harder to implement," Litman said.

Technology such as Watson could support general questions that the system designers haven't explicitly programmed into it, and that opens up another range of possibilities for this technology.  Watson displays correct and alternative answers to questions, along with a numerical rate of confidence in the answer.

"Being able to explain your answers to some extent is a really important capability," Litman said.

While right now Watson's process is limited to a single question and answer, this technology could be useful in larger applications if it could have some sort of conversation.

If Watson isn't confident in the answer, the computer could engage in a conversation to get more information. On the other hand, it might be confident, but because it exposes other alternatives, the user might want to explore why the computer suggested those answers.

A lot of Litman's research has been on conversational capabilities, and IBM is pursuing the integration of that line of research with Watson.

In the context of a tutor, the student might have a question that the computer tutor wasn't prepared to answer. But if it was essentially hooked up to the Web, it could use its conversational techniques to go off into other areas for answers.

Partly because Watson doesn't always get the right answer, it could be used as an assistant to human tutors. That way the computer can give real-life tutors information to help them make decisions.

In her natural language processing class, students do a big project at the end of the semester where they develop a question-answering system.

“It was really fun for the students as well as kind of illustrated the important technical issues in natural language processing,” Litman said. “In fact, I’m teaching it next semester, and I’m trying to rethink how I can recast my old assignment to be in this Watson/Jeopardy camp because it just generates so much excitement for the students."