Social Learning in Neural Agent-Based Models
Résumé
Agent-based models (ABMs) are widely used in examining how interactions at the individual level shape the behaviors of collectives. It has recently been argued that ABMs tend to be too simple and abstract to capture the complexity and variability of real-world actors engaging in social interactions. We address this criticism by integrating artificial neural networks into ABMs, specifically focusing on enhancing the Hegselmann-Krause (HK) model. By replacing standard HK agents with multilayer perceptrons, we obtain a more realistic kind of ABM, more closely capturing the nature of actual agents. The approach yields more than one new model, given that, with multilayer perceptrons as agents, the core elements of the HK model can be defined in a number of ways. Through two computational studies, we compare the resulting models with each other and with a traditional individual-learning paradigm.
Agent-based models (ABMs) have become a popular tool for studying macro-properties of social systems which, although typically arising from simple micro-level interactions, cannot be fully understood by strictly analytical means. They are used across a range of domains, from economics and political science to epidemiology and urban planning (