Gaming the System
Economists have long used game theory to make sense of the world. Now engineers and computer scientists are using it to rethink their work.
This conundrum, known as the prisoner’s dilemma, is the most familiar example of a game, in the technical sense employed by game theorists. Game theory is a mathematical way to describe strategic reasoning, and the prisoner’s dilemma illustrates the three basic requirements of the situations it encompasses: the game must involve multiple agents (here, the two accomplices); each must make a decision (squeal or don’t squeal); and every decision must carry a quantifiable payoff (the prison terms) that varies according to the other agents’ decisions.
Game theory has been a staple of economics research since 1950, when John Nash, who taught at MIT from 1951 to 1959 and is the subject of the movie A Beautiful Mind, published the seminal paper that would win him the Nobel Prize in economics. As game theory has matured, it’s become even more central to that field. In just the last eight years, the Nobel Prize has gone to game theorists three times, for shedding light on, among other things, the logic of nuclear deterrence, the circumstances in which free markets can and cannot maximize public welfare, and the best solutions to “matching problems”—organs and patients, medical residents and hospitals, and the like.
But recently game theory has been drawing attention in engineering and computer science, too. Researchers are using it to analyze thorny problems such as optimizing traffic flow or preventing blackouts.
Asuman Ozdaglar, SM ‘98, PhD ‘03, a professor of electrical engineering and computer science, says the rise of the Internet has made this necessary. Historically, the engineers of communication networks had to contend with a wide range of technical questions—such as power constraints and the relative merits of centralization or decentralization. But with the Internet, they suddenly had to deal with human agency, too.
If a Comcast subscriber in Boston and an EarthLink subscriber in San Francisco are exchanging data, their transmissions are traveling over networks maintained by several different providers: Comcast, EarthLink, and others in between. “The whole operation relies on both collaboration and competition of these different parties,” Ozdaglar says. “How do you design protocols that will actually yield the right incentives for people to collaborate?” In other words: why does the Internet work even though it is made up of individual networks? Game theory provides a way of answering that kind of question.
As engineers began bringing game theory to bear on questions within their field, however, they also realized that the tools of their trade were applicable to outstanding questions of game theory. Indeed, of the handful of researchers in the Department of Electrical Engineering and Computer Science (EECS) who work extensively on game theory, all have spent substantial time on questions more typically addressed by the social sciences.
EECS professor Constantinos Daskalakis is a good example. In 2008, he won the Association for Computing Machinery’s dissertation prize by showing how techniques drawn from theoretical computer science could shed new light on one of the central concepts in game theory: equilibrium.
Dahleh has also collaborated with Ozdaglar and her husband, the MIT economist Daron Acemoglu, to analyze how information propagates through populations. The “game” in this case is one in which people weigh the truth or falsity of information that reaches them, as they strive to maximize the accuracy of their own beliefs.
“These are questions that have been studied in both sociology and economics,” Ozdaglar says. Traditionally, however, these investigations have assumed that any person in a given population can receive information directly from any other. What engineers offer, Ozdaglar argues, are well-honed tools for analyzing the underlying network structure of the population. Most people, for instance, in fact receive most of their information from just a few immediate neighbors in the network—and they assign different probabilities to the accuracy of different neighbors’ claims.
“In the past, I think that social science and economics were dealing with problems differently than engineers,” Dahleh says. “Now we’re all talking about social networks—decisions in social networks, dynamics on networks—so I think the two fields are converging.”
MIT News Magazine