Traditional narrative explanations of prehistory have become increasingly difficult to operationalize as models and to test against archaeological data. As such models become more sophisticated and complex, they also become less amenable to objective evaluation with anthropological data. Nor is it possible to experiment with living or prehistoric human beings or societies. Agentbased modeling offers intriguing possibilities for overcoming the experimental limitations of archaeology by representing the behavior of culturally relevant agents on landscapes. Manipulating the behavior of artificial agents on such landscapes allows us to, as it were, "rewind the tape" of sociocultural history and to experimentally examine the relative contributions of internal and external factors to sociocultural evolution (Gumerman and Kohler in press). Agent-based modeling allows the creation of variable resource (or other) landscapes that can be wholly imaginary or that can capture important aspects of real-world situations. These landscapes are populated with heterogeneous agents. Each agent is endowed with various attributes (e.g., life span, vision, movement capabilities, nutritional requirements, consumption and storage capacities) in order to replicate important features of individuals or relevant social units such as households, lineages, clans, and villages. A set of anthropologically plausible rules defines the ways in which agents interact with the environment and with one another. Altering the agents' attributes, their interaction rules, and features of the landscape allow experimental examination of behavioral responses to different initial conditions, relationships, and spatial and temporal parameters. The agents' repeated interactions with their social and physical landscapes reveal ways in which they respond to changing environmental and social conditions. As we will see, even relatively simple models may illuminate complex sociocultural realities. While potentially powerful, agent-based models in archaeology remain unverified until they are evaluated against actual cases. The degree of fit between a model and real-world situations allows the model's validity to be assessed. A close fit between all or part of a model and the test data indicates that the model, albeit highly simplified, has explanatory power. Lack of fit implies that the model is in some way inadequate.
An agent-based model of sign language persistence informed by real-world data
