Modeling agent decisions and actions with the aid of artificial intelligence

Keywords: Agent-based modeling, complex systems, artificial intelligence, modeling agents’ behavior
Abstract Type: Paper Abstract

Abstract

Increasingly, applications and recognitions have been given to agent-based modeling (ABM) and complex systems (CS) science that backs up ABM when handling many human-environmental or sustainability problems. At the same time, ABM has been doubted and criticized in many instances. This presentation summarizes several advances and challenges in agent-based modeling over the last 2-3 decades, highlighting that understanding and modeling agents’ behaviors is a major challenge, and should be one of top priority research areas for ABM and CS scholars. Towards this goal, artificial intelligence and data science will likely generate tremendous impacts on understanding agents’ behaviors in complex systems. Specifically, we show an example of using reinforcement learning and convolutional neural networks to dig out the rules and mechanisms behind many decisions and actions of agents directly from data. Without downplaying the importance of domain knowledge and theory when modeling agent decisions and actions, we call for more efforts to be invested on the unique potential of artificial intelligence approach (with the aid of data science) to detecting internal, theory-relevant mechanisms in many complex human-environment systems.