Analysis of Cognitive Architecture in the Cultural Geography Model.

摘要

The Cultural Geography (CG) Model is a multi-agent discrete event simulation developed by the U.S. Army Training and Doctrine Command (TRADOC) Analysis Center-Monterey. The GC model seeks to enhance existing DoD efforts to model the responses of populations and social networks to operations conducted by the military in operations other than war (OOTW) and irregular warfare (IW) campaigns. The model is based on social science theories. In particular, agent decision-making algorithms are built on Exploration Learning (EL) and Recognition-Primed Decision Making (RPD), and trust between entities is modeled to increase the realism of the interactions. This study analyzed the effects of these components on behavior and scenario outcome. It aimed to identify potential approaches for simplifying the model, and improving the traceability and understanding of entity actions. The effect of using EL/RPD with and without trust was tested in basic stand-alone scenarios to assess its impact in isolation on entities' perception of civil security. Further testing investigated the influence on entity behavior in the context of obtaining resources from infrastructure nodes. The findings indicate that choice of decision-making method did not significantly change scenario outcome, but variance across replications was greater when both EL and RPD were used. Trust was found to delay the rate of change in population stance due to interactions, but it did not affect overall outcome if given sufficient time to reach steady state.