A Understanding Incentives in Social Computing

My research is at the intersection of computer science and economics. I study systems where multiple self-interested agents interact and I aim to design such systems to achieve desirable outcomes. Lately, researchers are increasingly making use of large and distributed groups of participants to solve complex problems. For example, businesses including Amazon, Yelp, and Expedia strive to collect truthful opinions about products and services. Prediction markets such as Inkling markets (home.inklingmarkets.com) are used to elicit accurate estimates about political, business, and entertainment events. Sites like All Our Ideas (www.allourideas.org) and openIDEO (openideo.com) aim to collect creative and useful solutions to important social challenges. My research is first and foremost driven by the desire to understand the strategic interactions of self-interested participants in a complex system. Given a certain system design, the participants determine their strategies to maximize their welfare, but their behavior is not necessarily desirable for the designer’s objective. For instance, the state-of-art prediction market mechanism only motivates a participant to reveal truthful information if she trades once in the market. Beyond understanding the participants’ behavior, I would like to better design these systems to incentivize desirable participants’ behavior for a wide range of problems. In pursuing these goals, I use techniques and insights from artificial intelligence, microeconomics, statistics, human computer interaction and the social sciences. Economic theory has promising ways to incentivize desired behavior from strategic agents, and computer science provides algorithms to coordinate and to analyze the participants’ behavior. Furthermore, I enjoy tackling problems using a mix of theoretical and experimental approaches, in order to address the significant gap between theory and practice in this field. For instance, peer prediction methods have nice theoretical guarantees but have not been evaluated in experiments. Prediction markets have stellar practical performance, but lack theoretical explanations for its success. My eventual goal is to design these systems to have great practical performance and solid theoretical foundations.