Learning Explanatory Coherence Models from Agent-Based Simulation Experiments

In Agent-based models of complex adaptive systems, macro-level behavior is not engineered but results from local interactions among agents. Due to the consequences of complex, distributed interactions among decentralized agents, the causal chain of cross-cutting processes that give rise to behavioral regularities is difficult to explain. To provide a context for the explainability of agent-based models, a systematic review of philosophical and cognitive models of causal explanation is provided. For illustration purposes, the theory of explanatory coherence is used as a computational framework for learning explanatory cognitive maps of increasingly refined and broadened model features. The framework offers a perspective that signifies strategies for learning coherence-driven explanatory models with implications for simulation model development environments.