Abstract: Agent-based modeling has been criticized for its apparent lack of establishing causality of social phenomena. However, we demonstrate that when coupled with evolutionary computation techniques, agent-based models can be used to evolve plausible agent behaviors that are able to recreate patterns observed in real-world data, from which valuable insights into candidate explanations of the macro-phenomenon can be drawn. Existing methodologies have suggested the manual assembly and comparison or automated selection of pre-built models on their ability to fit patterns in data. We discuss the cons of existing manual approaches and how evolutionary model discovery, an evolutionary approach to explore the space of agent behaviors for plausible rule-sets, can overcome these issues. We couple evolutionary model discovery with concepts from the Agent_Zero framework, ensuring social connectivity, emotional theory components and rational mechanisms. In this study, we revisit the farm-seeking strategy of the Artificial Anasazi model, originally designed to simply select the closest potential farm plot as their next farming location. We use evolutionary model discovery to explore plausible farm seeking strategies, extending our previous study by testing four social connectivity strategies, four emotional theory components and five rational mechanisms for a more complex human-like approach towards farm plot selection. Our results confirm that, plot quality, dryness and community presence were more important in the farm selection process of the Anasazi than distance, and discover farm selection strategies that generate simulations that produce a closer fit to the archaeological data.
Introduction
Agent-based modeling (ABM) is a powerful tool for "growing" artificial societies to see if local rules can recreate macro-scale historical patterns. However, ABMs often face a "black-box" problem: researchers typically hand-craft a rule (e.g., "move to the closest farm") and then simply tune parameters. This risks subjective and inaccurate representation because it fails to explore the vast space of other possible behavioral rules.
This study targets the landmark Artificial Anasazi model of the Long House Valley (800–1400 AD). While the original model assumed households moved to the closest available plot, the authors argue this is not entirely human-like. They introduce Evolutionary Model Discovery (EMD)—a framework combining Genetic Programming and Random Forest regression—to automatically evolve and interpret the rules governing ancient human behavior.
Methodology
The EMD framework operates in two stages. First, Genetic Programming (GP) evolves decision-making rules, treating them as trees of primitives (factors like land quality or distance) and mathematical operators. The GP searches for combinations that minimize the Root Mean Squared Error (RMSE) between simulated population time-series and archaeological household data.
Second, Random Forest (RF) Importance analysis is used for "white-box" interpretation. Once the GP identifies successful rules, an RF regressor is trained on the presence of factors within those rules to quantify their contribution to model accuracy. Agents' utility functions were built from three dimensions inspired by the Agent_Zero framework: social/emotional factors, rational mechanisms (land quality, dryness, yield), and information flow configurations.
Key Results and Insights
The EMD approach significantly improved accuracy, with 75% of GP runs producing lower RMSE values than the best results previously reported in literature using manual calibration. A primary finding was the debunking of the "closeness" myth; the original reliance on distance was found to be less effective than more complex social and environmental considerations.
As shown in Figure 1, statistical analysis confirmed a clear hierarchy of importance: Quality > Social Presence > Distance > Migration > Dryness. This suggests that ancestral Pueblo farmers were more deliberative than previously modeled, often moving further away to find better land or larger communities.
Furthermore, the scope of agent knowledge proved critical to model performance. The researchers tested four information flow strategies: knowing only neighbors ($S_{Neigh}$), knowing based on performance ($S_{Perf}$), knowing via family networks ($S_{Fam}$), and full valley awareness ($S_{All}$).
Figure 6 illustrates that models performed best when agents had "Full Information" ($S_{All}$) of the valley. This suggests the ancestral Pueblo likely maintained robust information-sharing networks to optimize their farming locations. Additionally, while factors like dryness ($F_{Dry}$) had low individual importance, they showed high joint contribution when paired with land quality, highlighting that environmental factors are not evaluated in isolation.
Conclusion
By coupling evolutionary computation with agent-based modeling, this research provides a data-driven methodology for inferring the underlying logic of social phenomena. The discovery that land quality and community presence were paramount to the Artificial Anasazi challenges long-standing assumptions in archaeological modeling. While computationally intensive, the EMD framework serves as a general-purpose tool for behavior inference in complex systems, from historical migrations to modern epidemiology.
Summary updated