Adaptive Advantage of Learning Strategies: A Study Through Dynamic Landscape

Abstract

Learning can be classified into two categories: asocial learning, e.g. trial-and-error; and social learning, e.g. imitation learning. Theory using mathematical models suggest that social learning should be combined with asocial learning in a strategic way (called learning rule or learning strategy), and that that combination should be scrutinised under different environmental dynamics, to see how advantageous the learning rule is. More interestingly, learning has been shown to be beneficial to the evolutionary process through the Baldwin Effect. This paper investigates the adaptive advantage of social learning when combined with asocial learning under a number of environmental variations. We propose a Dynamic Landscape as well as an algorithm combining both asocial and social learning in order to test our hypotheses. Experimental results show that if each individual in the population is either asocial or social, but not both, the average fitness of the population decreases when the proportion of social learners increases as the environment changes. Moreover, a population consisting entirely of asocial learners outperforms the previous type of population. If every individual agent in the population can perform both asocial and social learning depending on a strategic rule, the evolving population outperforms the two previous populations with respect to average fitness.