Todd, P.M., and Miller, G.F. (1991). Exploring adaptive agency II: Simulating the evolution of associative learning. In J.-A. Meyer and S.W. Wilson (Eds.), From animals to animats: Proceedings of the First International Conference on Simulation of Adaptive Behavior (pp. 306-315). Cambridge, MA: MIT Press/Bradford Books.

Abstract

We consider psychology as the study of adaptive agency, investigating the processes and mechanisms resulting in fitness-increasing behavior in the world. A central issue in psychology so construed becomes: what are the relations between the primary adaptive process of evolution by natural selection, and the adaptive processes psychologists call "learning"? In particular, under what conditions would learning evolve? To explore this issue, we use genetic algorithms to simulate the evolution by natural selection of neural networks, which in turn control the behavior of simple creatures in virtual environments. We have developed what we consider the simplest possible environmental challenge in which unsupervised associative learning could prove adaptive: "bootstrapping" the learned use of one highly accurate, but individually varying, sensory modality by another less accurate, but evolutionarily stable, modality. We have found a possibly quite general U-shaped curve relating the time (in number of generations) to evolve the use of unsupervised learning on the varying "bootstrapped" modality, to the accuracy of perception in the stable modality which guides this learning. This U-shaped curve appears to represent a trade-off between the adaptive pressure to evolve learning (which peaks when perception accuracy in the stable guiding modality is at chance) and the ease of learning during a given lifespan (which peaks when this accuracy is perfect.)

See my other publications on Evolution of learning   Environment structure   Animal behavior   



   Back to main page.
   View publications categorized by topic.
   View publication abstracts searchable by keywords