Influencing Massive Multi-agent Systems via Viral Trait Spreading

This paper describes a method by which a massive multi-agent system can be influenced without resorting to micromanagement. This method could be utilized in the development of meta-reasoning components of individual agents. Agents in the system adopt the traits of their successful peers. The administrator guides this spread of traits through selectively injecting influential agents with modified traits. These key agents are identified via social network analysis techniques. Experimentation is described in which the system is tested for its ability to automatically adopt an acceptable configuration as well as testing the ease in which the administrator is able to guide the system to a better configuration.