Coevolution of learning ability and neuro-cognitive organization

Abstract

The environmental complexity thesis states that environmental complexity is the driving force behind the evolution of cognition. Herbert Spencer held a particularly strong version of this view, and believed that life and mind can be understood as reflections of the environment they evolved in. However, Spencer's view does not account for the possibility of fit but diffusely implemented behaviour. As connectionist AI has amply demonstrated, fit behaviour does not by itself necessitate any isomorphism between a species' neuro-cognitive organization and the environment. We suggest supplementing Spencer's view with an account of the selection pressures that would cause evolution to organize cognition after the environment, and identify selection pressure on learning ability as a candidate. We argue that the more a species' neuro-cognitive organization resembles the organization of the environment, the easier it is to make appropriate updates in behaviour. We discuss various types of learning ability as it occurs in nature, and identify latent learning as the type most likely to constrain neuro-cognitive organization. We then build an Artificial Life model of the evolution of latent learning, and compare the structures of networks evolved under selection pressure for latent learning with networks evolved in absence of such selection pressure. Unlike the latter, the former repeatedly evolved the same compact behaviour system, which innately encodes some of the spatial relations of the environment. Our results indicate that selection pressure on learning ability can indeed guide evolution towards forms of neuro-cognitive organization that reflect environmental features.