Michael Littman

Image
Michael Littman
Professor of Computer Science
Mike Cohea/Brown University
Must it be this difficult? Michael Littman once asked each member of an honors seminar to find three digital clocks and write down the steps required to set the time. Thirteen people documented 39 different approaches to changing a clock.

There is no better crossword puzzle assistant than a powerful computer. Need an 11-letter word with two “t’s” and a “p” in the third position? Not a problem. Just crush the entire Oxford English Dictionary, and you’ll have a short list of possibilities in a matter of milliseconds.

That, of course, is not the point of working on a crossword puzzle. “Finding all the matches — that’s old, old stuff,” said Michael Littman, newly appointed professor of computer science at Brown and an expert in artificial intelligence. “We designed a crossword solver that actually uses the clues, which it can understand because it has seen lots of crossword puzzles. It’s pretty confident.”

The “we” was a trio of bright, young AI computer scientists at Duke University in the late 1990s: Littman, then an assistant professor; a graduate student who created Google’s spelling-correction program (“Did you mean ... ?”); and another graduate student who became chief technical officer at Rosetta Stone. Their creation “understands the clues in an artificial intelligence way,” Littman said. “It can make semantic associations, and it can recognize some clues by their structure. If the clue is ‘Hanks of the movies,’ it understands that what is being asked for is the first name of an actor. It hunts through a database of movies, finds Tom Hanks, and bang. It’s got it.”

Littman, a graduate of Yale (B.S./M.S, 1988), earned his Ph.D. in computer science at Brown in 1996, working in artificial intelligence with Leslie Kaelbling. He began his academic career at Duke (1996-2000), spent two years in the Artificial Intelligence Principles Research Department at AT&T Labs, then joined the Rutgers faculty. When he accepted the position at Brown, he was a full professor and department chair at Rutgers.

“When I first came onto the scene as a junior research person, there was a struggle between the machine-learning, neural-net people and the people who were interested in classical artificial intelligence,” Littman said. “There was a sense that they would never get along.” The growth and development of “Big Data” has brought the two camps together, and now classical AI questions are being addressed by bringing large-scale statistics to bear.

“We have unprecedented amounts of computer power to process unprecedented amounts of data, and we’re getting very clever about how to draw interesting stuff out of it,” Littman said, “but people who are interested in AI want to talk about the meaning of something. Numbers are just numbers; they don’t give you meaning. How can we merge notions of data and quantification with what we recognize as meaning?”

That’s no longer just a research question. There are few areas of business that aren’t touched by the combination of AI and Big Data. Almost any company can now seek answers to significant product information questions by studying the vast record of user experience that’s gathering in the cloud: What instructions appear to be giving people the most difficulty? What features are missing? Why do people abandon shopping carts? Finding those answers requires expertise in data as well as in learning.

Littman is also interested in making the user interface of day-to-day technologies more intelligent, even programmable.

“At Rutgers, I taught a class for non-computer science people. I found that the students were game. They listened to me. It was fun, but they didn’t really see much point to it,” he said. “One of the reasons is that a lot of the computer science stuff is packaged up and hidden in their devices. There’s lots of incredible stuff going on inside smart phones. They’re all computers, so they can accept programs. It’s just that we’re not allowed to program them. I’d like to find a way to make everyday devices programmable by the end user not just the systems designers.”

User-friendly languages available now, Littman said, could help end users gain a measure of control even if the languages aren’t robust enough for software engineers. He hopes to offer courses at Brown for non-computer science concentrators, aimed at doing just that.

“There is so much to know about computer science that doesn’t involve programming,” Littman said. “What are these technologies? How can you command them? How can you use technology without being its pawn? That’s important for everybody, but we mostly teach it to computer scientists. I want everyone to have access to that knowledge.”

Browse New Faculty by Year