EmTech: IBM Tries to Make Watson Smarter

Three years after its artificial-intelligence engine Watson made its high-profile win on Jeopardy!, IBM is adapting the technology as it seeks practical commercial uses, an IBM executive explained yesterday at EmTech, a conference organized by MIT Technology Review.

The original version of Watson was built around a question-and-answer format optimized for the game show, but that turned out to be “just the first building block” in an emerging artificial-intelligence system, said Mike Rhodin, senior vice president of the IBM Watson Group.

Rhodin said IBM is refining Watson to make it more adept at providing the correct answer to a specific question in a specific domain—for example, by learning from previous queries. “Just as [computer operating system] platforms emerged in the 1960s, we are seeing the beginnings of that kind of a system emerge in the information age,” he said.

In the past year, despite skepticism in some quarters, the company has announced it was making more investments in the Watson platform (see “Facing Doubters, IBM Expands Plans for Watson”). It has also opened the platform to app developers (see “Trained on Jeopardy!, Watson Is Headed for Your Pocket”).

Some results of that strategy are now emerging. In one recent competition run by IBM, a startup called Majestyk came up with an app that added Watson functionality to a stuffed animal so that it could converse with a child and give feedback to educators—potentially revealing whether the child might have learning disabilities.

“We never would have thought of it; we don’t have that DNA,” he said. “It validated the idea that we needed to open up the platform and make it available to the startup marketplace.”

Ultimately, Rhodin said, IBM will pursue a revenue-sharing model for any effort that reaches market.

The company also continues to pursue applications in the medical, financial, and legal sectors. Using Watson to examine thousands of documents could, for example, help doctors see different diagnoses in order of probability and “rule out things they didn’t think of,” Rhodin said.

IBM has also been working with USAA, a company that provides financial services to U.S. military personnel. The Watson engine analyzes more than 3,000 USAA documents about financial or health-care benefits and answers questions from USAA users.

At the EmTech conference, a questioner from USAA pointed out that the system was having some trouble, in part because it would not allow users to ask follow-up questions. Each request was treated as a standalone problem, making Watson awkward to use at times. Rhodin said he would meet with USAA later this week.

Even with such stumbles, IBM has high hopes for getting Watson to work well in medical settings (see “Watson Goes to Work in the Hospital”). And Rhodin said significant effort is now being expended on applying the technology to automated call centers.