Embedding Artificial Intelligence in Agent-Based Models

Summary

Agent-Based Models (ABMs) have been increasingly used to study social phenomena, from the emergence of social norms to population dynamics or cultural transmission processes. Key to this method of computational simulation is the tension for explaining how macroscopic phenomena emerge from the interaction of agents behaving in a plausible manner. However, the behavior is too often encoded as a simple set of condition-action rules. We consider this kind of rule-based behaviour too simplistic, specifically when simulating human action and exploring the impact of decision-making processes on the simulation outcome. Therefore, we experiment with a novel type of agent whose decisions are based on casting the surrounding world as a Markov Decision Process, a decision-making model widely used in Artificial Intelligence. The performance of this type of agents is assessed on a simple ABM simulation and compared to that of traditional rule-based agents. We then analyse the interplay between decision-making processes and relevant simulation parameters such as the distribution of resources. Finally, we present the results of applying the insights thus gained to a concrete case study within the Simulpast project concerning the dynamics of Hunter-Gatherer populations in North Gujarat, India, in the mid and late Holocene.

SAA 2015 abstracts made available in tDAR courtesy of the Society for American Archaeology and Center for Digital Antiquity Collaborative Program to improve digital data in archaeology. If you are the author of this presentation you may upload your paper, poster, presentation, or associated data (up to 3 files/30MB) for free. Please visit http://www.tdar.org/SAA2015 for instructions and more information.

This Resource is Part of the Following Collections

• Society for American Archaeology 80th Annual Meeting, San Francisco, CA (2015)   •
• SimulPast -- Simulating the Past to Understand Human Behaviour

Keywords

Spatial Coverage