Agent-based modeling is a widely used computational method for studying the micro-macro bridge issue by simulating the microscopic interactions and observing the macroscopic emergence. This thesis begins with the fundamental methodology of agent-based models: how agents are represented, how agents interact, and how the agent population is structured. Two vital topics, the evolution of cooperation and opinion dynamics are used to illustrate methodological innovation. For the first topic, we study the equilibrium selection in a coordination game in multi-agent systems. In particular, we focus on the characteristics of agents (supervisors and subordinates versus representative agents), the interactions of agents (reinforcement learning in the games with fixed versus adaptive learning rates according to the supervision and time-varying versus supervision-guided exploration rates), the network of agents (single-layer versus multi-layer networks), and their impact on the emergent behaviors. Regarding the second topic, we examine how opinions evolve and spread in a cognitively heterogeneous agent population with sparse interactions and how the opinion dynamics co-evolve with the open-ended society's structural change. We then discuss the rich insights into collective intelligence in the two proposed models viewed from the interaction-based adaptation and open-ended network structure. We finally link collective emergent intelligence to diverse applications in the realm of computing and other scientific fields in a cross-multidisciplinary manner.
Read
