The authors investigate the evolution of individual behavior in dynamic distributed systems populated by two sorts of individuals, P and Q, linked by read-only (that is, as opposed to message-passing) local communication. This will be the framework for a discrete predator/prey (P/Q) model. The environment is a 2D torus upon which the individuals have absolute locations that initially are random. Each individual can see its immediate neighbors in locations relative to that individual (this is the read-only communication). Behavior programs are used by individuals to move to a neighboring position. Further, individuals of type P can occasionally consume those of type Q--individuals consumed at time t are not in the system at later times.
States, for individuals not yet consumed, are a description of what they see in their neighborhood (that is, the relative locations, not the absolute locations). The investigation is limited to experimental simulations in which individuals of type P attempt to consume individuals of type Q, and those of type Q attempt to avoid consumption.
Q’s behavior program is occasionally mutated. Mutations that increase the expected survival time are considered successful. These successful mutations are used to build future Q-programs. In this way, a given P-program drives the evolution of Q-programs to a successful survival strategy, which may be seen globally as swarming.
The authors hope to use this method to evaluate Hamilton’s selfish herd hypothesis. In particular, the directionality of a predator attack (as defined in the formula>P-program) is shown to play a critical role in the evolution of swarming behavior versus purely dispersive behavior.