By Marlene Cimons, National Science Foundation
Getting a computer to be intuitive and funny, or having it grasp the nuances of emotions and ideas, qualities more typically associated with humans, remains one of the most provocative and elusive goals in the artificial intelligence field. But scientists are getting closer.
“Computers lack this understanding of the relationship of ideas—the whys and wherefores—and we are trying to imbue the machines with that kind of understanding of the world,” said Kristian Hammond, whose career-long quest has been to capture these elements and instill them into computers. “At the end of the day, we are trying to model human behavior on the computer so it can better interact with us.”
Hammond, professor of electrical engineering and computer science at Northwestern University’s McCormick School of Engineering and Applied Science, is working on a set of projects aimed at enhancing computer ingenuity by generating products in a variety of forms.
These include giving computers the ability to create entirely new, personalized documents, news shows and news stories, comics and humor based on the juxtaposition of ideas. “It all comes from understanding the angle of a narrative, of what is important,” Hammond said. “Some things are funny, some are straightforward.”
For example, “give me a person’s name, and the computer will put together a web comic based upon that person,” Hammond explained. “Finding ideas that have an interesting juxtaposition; it turns out sometimes they are funny.”
Here’s how one of the projects’ systems would work: the computer first would “read” the document the consumer is working on, whether reading or writing it, and extract its key elements. Then it would compose a brand new document based on the threads it picked up, combined with information it gleans from other existing documents.
“We are building systems that can generate compelling, informative, enlightening experiences that people will want to consume,” said Hammond, who also directs the university’s Intelligent Information Laboratory. “In order to do that, we have to have a fairly deep understanding of how people think around a wide range of possibilities, and imbue the machine with that behavior.”
The research is funded over three years with $712,883 from the National Science Foundation as part of the American Recovery and Reinvestment Act of 2009.
Ultimately, the work could have an impact on how consumers and businesses use computers, and could change the nature and scope of the information that is delivered to the public.
“There’s a tremendous amount of information out there, but it is in a million different places,” Hammond said. “It’s feasible for a machine to look across a wide range of behaviors; we can take the data that’s out there, and find the story in it.”
It also likely would expand the field of artificial intelligence research, spawning new companies eager to take advantage of the sophisticated programs. Northwestern University recently licensed some of this technology to a start-up, Narrative Science, in Evanston, Ill., which grew out of one of its research projects. Narrative science technology generates sports stories, industry reports, headlines from data, and without any human authoring or editing.
“The company can use the data associated with a baseball game or softball game to generate stories,” Hammond said. “The first piece it published was on the Big Ten Network for college baseball and women’s softball. It wrote a recap story on every game played in the Big Ten.”
Another prototype can create a virtual news show. It collects, edits and organizes news stories, which are presented by virtual anchors. The system uses existing web resources, but also retrieves relevant images and blogs with commentary, and even can generate dialogue and interaction between the anchors.
“It knows how to analyze a story, and go online to find people expressing opinions, and can organize them into an ‘op-ed’ piece,” Hammond said. “If you want to know more about a movie, for example, imagine that you can build a conversation between two people, one of whom loved the movie, one of whom hated it. The input is the name of the movie—it knows what a movie review should look like, and it can go to various sites and find the consensus about the movie—and create an entire dialogue around it.”