Jeff Shamma fondly remembers his first major academic Eureka moment. He was in grad school at LIDS -- he pushed his chair back and started pacing back and forth around his Tang Hall apartment. When he presented his findings to his adviser, LIDS Professor Michael Athans, he was almost breathless with excitement.
His adviser heard him out, let him feel the excitement, and then said: "Oh so you derived a better bound? What else have you done?" And Jeff thought, "Wow. I guess this is going to be a lot harder than I thought." But he got a taste of what it felt like to crack a case. And more than two decades later the New York City born father of four still lives for it.
Jeff has been based at Georgia Tech for the past seven years where he's working on cross-pollinating his background work in Control Systems with Game Theory for multi-agent systems. He juggles several projects in addition to his teaching and service activities. His projects highlight his journey from both feet planted firmly in the engineering world to his rather unexpected migration into the social world. Still, most of his projects have a multi-agent theme -- whether that's many controlled devices interacting, or a hybrid of machine and humans interacting, or humans interacting with humans.
To get a sense of his current focus, think about unmanned vehicles, as in mobile sensors or search and rescue robots. The current model is that there are one or several pilots for one vehicle. Shamma is working on finding a way to reverse that so there's one pilot - or one operator - for multiple vehicles. He likens it to a coach and their team of players. If a coach had to be moving the players on the team individually, it would take several coaches to get anything done. But if the coach could issue some high level commands from the sideline, the players can then try to execute them in concert, and the team needs only one coach for all the players on the field.
While in many ways more efficient, the challenge of the one coach-many players model is that there can be miscues and miscommunication. Not everyone has all of the information, and top-level communication is limited -- a coach can't yell out everything that he's seeing in the field from his larger perspective. The brilliant thing is that plays get executed anyway, largely as the coach instructs, and the game moves forward.
There are many systems that function in this coach-team fashion. So, Jeff reasoned, why couldn't it work that way in even more systems -- for instance, a manufacturing or an engineering situation? This line of inquiry led Jeff to Game Theory -- a pursuit that changed his research direction completely.
A major component of Game Theory is incentivization. The assumption is that humans interact, play, or execute tasks both with their own motives or desired outcomes in mind, as well as with the actions or reactions of others factored in. Balancing the two can lead to very effective collaborations between people with complementary incentives. Jeff is working on programming robots, drones, and machines with these ideas in mind. He looks at how to incentivize these machines in such a way that they act as a collective effectively, while still using the one operator to multiple vehicles model. "But I'm getting more excited by taking the controls concepts and things we did at LIDS and applying them toward problems in Game Theory in societal settings," says Jeff. "I'm finding that control engineering is relevant to problems in the mathematical social sciences and in game theory."
To understand how much of a shift this is for Jeff, you have to understand that Jeff purposely didn't go into the social sciences as an undergraduate because he felt like it wasn't well grounded. "You didn't have mathematical models that you could assess in repeatable controlled experiments as with physics. So there's this grounding that you can build on. That's what turned me off from social sciences, and now it's completely flipped," he says.
"Now what I find most exciting about the social sciences is that it's not first principles based on physics. And yet you still want to design systems and/or policies that are effective. But you don't have the luxury of these well-understood mathematical models. And that's what makes dynamics and controls all the more relevant! So it's completely flipped, what turned me off before is what makes it all the more exciting now," he says. Interestingly, just as Jeff was starting to look into Game Theory, he read a paper by leading researchers in the field that said: If players are not informed of the incentives of other players, then dynamic coordination to equilibrium is impossible.
He used some controls ideas to show that they didn't have the full story. A player or actor can react not just to another's actions, but to that person's recent trends, as well. For instance, if Player A knows that Player B has a trend of being altruistic by giving more money to someone who is in apparent need, then Player A can use that information to predict whether Player B will donate to charitable causes. Essentially, Jeff realized individual players didn't necessarily have to know the incentives of other players, they could take actions in relation to them just by knowing trends in their behaviors, and react in a more sophisticated way than what the paper allowed.
"I remember where I was in the office when I had this idea and I programmed it and saw that it was working," says Jeff. "Now this is some 20 years later than the pacing incident at LIDS, but I reacted the same way I did as a graduate student. I got very excited and became short of breath, I stood up and started pacing." When he realized what he had derived, he started going to Game Theory conferences -- where he knew no one -- and met Game Theorists. It pushed him out of his comfort zone, and it took a bit of persistence, but he was able to have the conversation he wanted, present his findings, and get recognized for it. "It was pretty gratifying," he says.
Jeff's greatest challenge these days seems to be an embarrassment of riches with multiple projects, as it becomes more and more difficult to find time to focus, read, and think. "I still don't feel I'm progressing unless I feel like I'm personally involved...that's a persistent tension."
He may yet get a reprieve when at the end of July, Jeff, his two teenage sons, two younger daughters, and his wife pack up and head to the Middle East for an extended leave of absence. Jeff's new faculty position will provide funding to develop and sustain robotics and controls research at King Abdullah University of Science and Technology (KAUST), a government research university an hour north of Jeddah, on the Red Sea in Saudi Arabia. Here he will continue using the fundamental ideas of information and decision sciences to shape new research trends in exciting, floor-pacing ways.