Applicability of the Future State Maximization Paradigm to Agent-Based Modeling: A Case Study on the Emergence of Socially Sub-Optimal Mobility Behavior

Abstract

Novel developments in artificial intelligence excel in regard to the abilities of rule-based agent-based models (ABMs), but are still limited in their representation of bounded rationality. The future state maximization (FSX) paradigm presents a promising methodology for describing the intelligent behavior of agents. FSX agents explore their future state space using “walkers” as virtual entities probing for a maximization of possible states. Recent studies have demonstrated the applicability of FSX to modeling the cooperative behavior of individuals. Applied to ABMs, the FSX principle should also represent non-cooperative behavior: for example, in microscopic traffic modeling, there is a need to model agents that do not fully adhere to the traffic rules. To examine non-cooperative behavior arising from FSX, we developed a road section model populated by agent-cars endowed with an augmented FSX decision making algorithm. Simulation experiments were conducted in four scenarios modeling various traffic settings. A sensitivity analysis showed that cooperation among the agents was the result of a balance between exploration and exploitation. We showed that our model reproduced several patterns observed in rule-based traffic models. We also demonstrated that agents acting according to FSX can stop cooperating. We concluded that FSX can be useful for studying irrational behavior in certain traffic settings, and that it is suitable for ABMs in general.