Exploring social theory integration in agent-based modelling using multi-objective grammatical evolution

Tuong Manh Vu, Charlotte Buckley, Joao A. Duro, Robin C. Purshouse

Keywords: agent-based modelling, inverse generative social science, genetic programming, grammatical evolution, multi-objective evolutionary algorithm

Abstract: In Generative Social Science, modellers design agents at the micro-level to generate macro-level a target social phenomenon. In the Inverse Generative Social Science (iGSS), from a target phenomenon, the goal is to search for possible explanatory model structures. This model discovery process is a promising tool to improve the explanatory capability and theory exploration of computational social science. This paper presents a framework for iGSS and applies Grammatical Evolution to an empirically-calibrated agent-based model of alcohol use. Results of the model discovery process find many alternative rules for agent behaviours with different trade-offs. Future work should involve domain experts to evaluate the discovered structures in terms of theoretical credibility and knowledge contribution.