Specialization under social conditions in shared environments

Abstract

Specialist and generalist behaviors in populations of artificial neural networks are studied. A genetic algorithm is used to simulate evolution processes, and thereby to develop neural network control systems that exhibit specialist or generalist behaviors according to the fitness formula. With evolvable fitness formulae the evaluation measure is let free to evolve, and we obtain a co-evolution of the expressed behavior and the individual evolvable fitness formula. The use of evolvable fitness formulae lets us work in a dynamic fitness landscape, opposed to most work, that traditionally applies to static fitness landscapes, only. The role of competition in specialization is studied by letting the individuals live under social conditions in the same, shared environment and directly compete with each other. We find, that competition can act to provide population diversification in populations of organisms with individual evolvable fitness formulae.