UC San Diego

The modeling and prediction of collective human behavior has been one of the key challenges of social sciences for several decades. In this dissertation, we use an algorithmic approach to study the behavior of multiple agents that interact with each other. A typical scenario considers a group of agents driven by both individual and collective incentives who communicate with each other through a network whose links represent potential interactions among them. We consider both coordination tasks, where the incentives of the agents are aligned, and non-coordination tasks, where their incentives are conflicting. The tasks we consider include social differentiation, selection of a reciprocating partner, and information aggregation. One of the key questions is whether the agents can solve these tasks. To address this question, we model agent behaviors algorithmically and analyze both individual and collective outcomes. For each task, we assess which algorithmic models will explain observed human behavior and which are more efficient in accomplishing the task. Finally, we study how the network structure affects outcomes and performance