LEARNING TO ADAPT TO CHANGING ENVIRONMENTS IN EVOLVING NEURAL NETWORKS

To study learning as an adaptive process, one must take into considera tion the role of evolution, which is the primary adaptive process. in addition, learning should be studied in (artificial) organisms that li ve in an independent physical environment in such a way that the input from the environment can be at least partially controlled by the orga nisms behavior. To explore these issues, we used a genetic algorithm t o simulate the evolution of a population of neural networks, each cont rolling the behavior of a small mobile robot that must explore efficie ntly an environment surrounded by walls. Because She environment chang es from one generation to the next, each network must learn during its life to adapt to the particular environment into which it happens to be born. We found that evolved networks incorporate a genetically inhe rited predisposition to learn that can be described as (1) the presenc e of initial conditions that tend to canalize learning in the right di rections; (2) the tendency to behave in a way that enhances the percei ved differences between different environments and determines input st imuli that facilitate the learning of adaptive changes; and (3) the ab ility to reach desirable stable states.