Putting Watson to work

The question-answering system that bested two human champions on the game show “Jeopardy!” is preparing to move out of the lab and onto your phone

SCS's Eric Nyberg (photo courtesy IBM)

Technical competitions can spur world-changing innovations. Frenchman Nicolas Appert was awarded 12,000 francs for developing an airtight food-preservation process that allowed Napoleon to feed his far-flung troops, and canned food was born. Charles Lindbergh received a $25,000 cash prize for his solo flight across the Atlantic, beating his competitors (six of whom died trying) and long-distance air travel soon became routine.

SCS professor Eric Nyberg is hoping the spirit of competition will inspire students in a new course to develop an innovative, award-winning application for mobile devices using IBM’s Watson technology, the same technology that beat two Jeopardy! champions in an on-air match in 2011.

The course, “Intelligent Information Systems featuring IBM’s Watson,” was offered for the first time in the fall 2014 semester and is being taught by Nyberg (CS’92), a professor in the Language Technologies Institute; Alan Black, an associate LTI professor with expertise in mobile speech interfaces; and Norman Sadeh (CS’91), a professor in CMU’s Institute for Software Research and an expert in mobile devices. Members of the IBM Watson Group research team that developed the Watson technology also are presenting guest lectures.

“The students will marry Watson’s technology with the personalized and more geographically contextualized information you get from a mobile device,” says Nyberg, a leading researcher in question-answering, or QA, computers, a discipline focused on developing systems that can answer questions posed in natural language. In 2008, Nyberg led the team of CMU researchers who helped IBM create the Open Advancement of Question Answering initiative.

CMU computer scientists also contributed two important pieces of Watson’s software—an algorithm which identifies the best written resources for answering a question about any given topic, and another algorithm which scores possible answers, helping Watson to identify which answer has the highest likelihood of being correct.

According to Nyberg, the students have access to a copy of Watson via the Watson Developer Cloud, but not to its source code. During the course, they are selecting an appropriate domain for their application; ingesting relevant data from that domain; testing, evaluating and training its responses; developing and designing a prototype; and creating a business plan for marketing the app.

“This is the first time IBM has given universities access to an instance of Watson via the Cloud for a course,” says Pam Induni, who leads the Watson University Engagement team, part of IBM’s Watson Group. Her team is working with CMU and nine other North American universities to offer a cognitive computing course leveraging Watson. (The other universities are New York University, Northwestern, Ohio State, Rensselaer, Stanford, the University of California at Berkeley, the University of Texas at Austin, the University of Michigan and the University of Toronto.)

On Jan. 9, 2015, students from the participating universities will pit their apps against one another in a competition at IBM headquarters in New York. They will be evaluated on the overall idea, the prototype, the quality of the dataset, their training of Watson, the business model and their presentation. The winning team will receive $100,000 in seed funding to create a startup for the prototype.

In a brainstorming session during Carnegie Mellon’s first week of classes this fall, ideas for possible applications using Watson’s technology spanned many domains: from the legal, where users might ask non-complicated questions, like what kinds of weapons can legally be carried in the city of Pittsburgh; to sports, where someone learning to hit a baseball could ask how best to execute a swing; to navigational, where one could plan a romantic getaway to a city at specified time—the user would receive recommendations for places to stay and sites to visit, and the computer might even suggest a schedule that optimizes the couple’s time together.

SCS junior Elan Rosenfeld, who says he signed up for the cognitive computing course for a chance “to learn a ton about machine learning,” says he likes the idea of developing a navigational app, which could utilize a user’s personal data, such as geographic location and search history.

“The challenge with developing  (the trip-planning) app would be the size of the corpus, the amount of data Watson would be ingesting,” Rosenfeld says. He favors a prototype that could plan an individual night out in a single city, which would allow for more effective testing and training of the dataset.

“Then, if we decided we wanted to turn the prototype into an application, we could ingest data from other cities,” Rosenfeld says. “Certainly if it were deemed good enough to win $100,000 in seed funding from IBM, it would be worth making an app that could be used internationally.”

Whatever application the students will be presenting in January, they are being guided in their decision by CMU faculty and the IBM mentors who are working with the participating universities to provide technical and other support for the software. Ken Barker, a researcher in IBM’s Watson Group, is the course mentor for the University of Texas at Austin, where he was a research scientist for eleven years before joining IBM’s Watson Group in 2011.

“We want to make sure that the faculty and students are set up to succeed, including choosing an appropriate problem,” Barker says. “Watson is pretty amazing technology, but it does very specific things, and if you try to get it to do something that it is really not built for, it would be frustrating.”

Barker admits there are going to be “some hiccups throughout the semester,” but also believes that the course is “a huge opportunity for the students. This is leading edge. It’s not like the software at Best Buy. We’re really doing something that hasn’t been done before, and the students are getting access to software that isn’t really available commercially yet.”

Although the syllabus centers on developing innovative apps using Watson technology, the goal of the course is to build the cognitive skills necessary to create “an ecosystem of innovators” with expertise in such fields as machine learning, content lifecycle management and natural language processing, says IBM’s Induni.

“Cognitive is the next era of computing, and these courses are about developing the skills of the next generation of software developers,” Induni says. “Lots of industries will be implementing Watson technology, and they’ll need these very skilled employees. Out of CMU and the other universities will come the next generation of Watson developers, researchers and innovators.”

If history is a reliable guide, a technical competition is one way to make that happen. “It’s a friendly competition,” Nyberg says, “but it is a competition.” He and Ken Barker have been ribbing each other via e-mail, even before the semester began. The two have worked together professionally for about five years, since Nyberg led the CMU team that helped develop Watson’s QA capabilities for IBM.

Barker calls it “a brotherly rivalry,” but adds, “tell Eric we’re coming for them!”

—Linda K. Schmitmeyer is a freelance writer and editor, and teaches in the Point Park University School of Communications.

For More Information: 

The Link Magazine | 412-268-8721 | thelink@cs.cmu.edu