Proceedings of the 2013 Winter Simulation Conference R. Pasupathy, S.-H. Kim, A. Tolk, R. Hill, and M. E. Kuhl, eds. EMERGENCE BY STRATEGY: FLOCKING BOIDS AND THEIR FITNESS IN RELATION TO MODEL COMPLEXITY Michael Wagner Wentong Cai Michael Harold Lees School of Computer Engineering Parallel and Distributed Computing Centre Nanyang Technological University 50 Nanyang Avenue, SINGAPORE 639798 ABSTRACT In this paper we aim to examine emergent properties of agent-based models by using evolutionary algorithms. Taking the model of flocking boids as an example, we study and try to understand how the pressure of natural selection towards strategic behavior can result in emergent behavior. Furthermore we investigate how an increase of complexity in the model affects those properties and discover some counter-intuitive behavior in the process. 1 INTRODUCTION Multi-agent based simulation (ABM) that is designed under a bottom-up approach defines behavior on a low or microscopic level, supplying every agent with a certain set of rules to follow. The high level, or macroscopic behavior in a system results from an accumulation of low level interactions between the individual agents. The striking feature of macroscopic behavior is its vastly increased complexity in comparison to the microscopic properties that cause them. Emergent behavior resulting from those interactions is usually hard to predict, difficult to understand and - most of all - very challenging to create on purpose. This stems from the nonlinear relationship between micro- and macroscopic properties. Slight changes in the environment or agent rules may lead to completely different outcomes in the simulation. A good example of this complexity is the Game of Life by Conway, where little differences in the initialization decide whether a population sustains itself, dies out or grows forever. Another one of the most well known examples are the flocking boids, first described and developed by Reynolds (1987), where agents resemble flocking animals such as fish or birds. Understanding the powerful phenomena of emergence is very desirable as it can provide valuable insight about the real life counterparts of the processes represented in the simulation. Stonedahl developed an evolutionary approach for finding emergent behavior in multi-agent based environments (Stonedahl and Wilensky 2010). In this case the emergent properties to search for (e.g., converging to flocks or flocking in vee-shapes) are described beforehand, by using their geometric properties. These properties, which mostly describe relationships between movement vectors, are quantifiable. As such they can be used to specify error measures, allowing to detect occurrences of this behavior and determine a metric distance to the ideal form. This approach was successful in repeatably creating convergent and volatile flocking of boids. Our first goal is to extend the work done by Stonedahl to investigate how variations in model complexity may affect the evolution of model parameters. Increasing the complexity of a model can be helpful by offering a larger set of possible directions to reach a goal. The negative trade-off here is an also increased 1479 978-1-4799-2076-1/13/$31.00 ©2013 IEEE