University of Southern California People At USC

View All
Milind Tambe
Milind Tambe

Milind Tambe

Professor of Computer Science
USC Viterbi School of Engineering

Milind Tambe, who came to USC in 1993 after earning his Ph.D. at Carnegie Mellon, grew up in Bombay (now Mumbai) with a keen, lifetime interest in science fiction. He has incorporated this into some of his classes, such as the freshman seminar titled “Artificial Intelligence Agents and Science Fiction.” Tambe explains how a random finding about randomization that he and his colleagues observed is leading to changes in airport security practices. He spoke with USC Trojan Family Magazine’s Allison Engel.


You’re an expert in artificial intelligence, specifically multi-agent teamwork. What are multi-agents?

Agents are entities with some degree of autonomy. They can be software-bots, which are software programs, or they can be robots or they can be real people or sensors. Multi-agents are systems where multiple agents interact.

You helped develop a program involving randomization that is changing airport security. What were the seeds of that project?

The way it worked out, it was never meant to be anything about randomization. In fall 2002, my Ph.D. student Praveen Paruchuri and I started work on teamwork with resource constraints, which is, let’s say, if you have two robots that want to collaborate but have only a fixed amount of fuel. It turned out in solving these problems that one way to attain coordination is to have the robots randomize their actions. We didn’t want that! The randomization seemed like an annoying byproduct. And when we presented our results at a conference in 2004, the audience members didn’t like it either. They said you could be much more systematic without randomization.

So what did you do?

USC’s CREATE [Center for Risk and Economic Analysis of Terrorism Events] was being formed about that time, and somebody gave a talk and said the terrorists had used the fact that our society runs precisely on time against us. The fact that everything works like clockwork was exactly the reason that they could target everything so precisely. I had this idea bubbling in my head about what to do with randomization. I thought, “This is what we use it for.”

Your annoying byproduct became your principal goal?

That’s right. … So we were no longer looking at resource constraints, we were just looking at randomization. We were very happy that we were pursuing something that seemed to us quite novel. And it seemed an important problem.

How old were you when you became a Trekkie?

In India, we got to see Star Trek the 1980s. Star Wars was 1977 or 1978. Before that, there were Isaac Asimov books that I borrowed and read from the library.

Is it true you sent a letter recently to the writer of a favorite Star Trek episode?

With one of my Ph.D. students, I put together a class that puts sci-fi and AI [artificial intelligence] teaching together. I wrote to a Star Trek writer, Melinda Snodgrass, who wrote about a robot who has rights. In class, we conduct a trial of that robot based on that episode. On a whim, I wrote and asked if she would come to the class. I was surprised she actually responded.

What other classes do you teach?

The one for undergraduates is called “Intelligent Agents and Science Fiction.” The other class is on the graduate level and is called “Software Multi-Agent Systems.” There’s also a freshman seminar I taught with Anne Balsamo from the School of Cinematic Arts, again on intelligent agents and science fiction.

Is there anything about teaching that you didn’t expect?

I really enjoy working with Ph.D. students a lot. This has been an interesting discovery. … The bonds that are formed seem very long-term.