Robustness of Multi-agent Models: The Example of Collaboration between Turmites with Synchronous and Asynchronous Updating

摘要

The robustness of multi-agent systems to simulation conditions is analyzed through a precise example, invented by Langton to investigate the foundations of artificial life. This system is composed of simple and memoryless agents, the turmites, which obey simple discrete local rules. While the local rules that govern each agent are kept constant, the interaction between agents is modified through nine variations. Our method consists in varying the updating scheme (synchronous vs. asynchronous) and the local conflict resolution policy (strong or weak exclusion rules). The effect of these modifications on three collaborative phenomena is experimentally estimated. The macroscopic robustness of the system is analyzed by examining how the conflicts that occur at the microscopic scale generate diverging trajectories of the system. Observations confirm that the definition of the individual agent's behavior is not the only setting that matters in the emergence of collaborative phenomena in complex systems: the way the agents are updated is also a key choice.