Learning and adaptation strategies for evolving artifact capabilities

In this study we address enhancing the ability of social agents embedded in multi-agent based simulations to achieve their goals by using objects in their environment as artifacts. Reformulated as a discrete optimization problem solved with evolutionary computation methods, social agents are empowered to learn and adapt through observations of their own behavior, others in the environment and their community at large. An implemented case study is provided incorporating the model into the multi-agent simulation of the Village EcoDynamics Project developed to study the early Pueblo Indian settlers from A.D. 600 to 1300. Eliminating the existing presumption that agents automatically know the productivity of the landscape as part of their settling and farming practices, agents use the landscape as an artifact, learning to predict its productivity from a few attributes such as the area's slope and aspect. Given the dynamic nature of the landscape and its inhabitants, agents also evolve various combinations of learning strategies adapting to meet their needs. The result is the demonstration of a mechanism for incorporating artifact use learning and evolution in social simulations, leading to the emergence of favorable learning strategies