Learning in Dynamic Multi-layered Social Networks: A Mesa Verde Example

In this paper we take a multi-agent model of agricultural subsistence in the Mesa Verde region between 600 A.D. and 1300 A.D. and allow the emergence of a set of overlaid social networks over time in response to environmental dynamics. These overlaid networks include kinship, economic, and community level (hub) networks. Agents are able to participate in each of these networks and strategies for participation are learned using a framework for Cultural Evolution, Cultural Algorithms. Agents are embedded in the Mesa Verde environment. The environment is divided into cells. Each cell of the model contains a modelled agricultural, animal, hydrologic, forest and shrub components. The value for each of these components is derived from paleoproductivity and archaeological data from the region. The values can change over time as a result of migrations, changes in available water etc. These changes are dynamically programmed into the model. It is shown that changes in the environment can differentially affect the various layers and effects can ripple through to other networks. Thus, certain networks may be less resilient to environmental fluctuations than others and require additional maintenance from the population in order to maintain them. The extent to which the social system is able to learn strategies to adjust the network in response to environmental stress will be discussed. The results of the model are then compared to the known distribution and content of archaeological sites found in the regions.