How Learning and Evolution Interact: The Case of a Learning Task which Differs from the Evolutionary Task

It has been reported recently that learning has a beneficial effect on evolution even if the learning involved the acquisition of an ability which is different from the ability for which individuals were selected (Nolfi, Elman & Parisi, 1994). This effect was explained as the result of the interaction between learning and evolution. In a successive paper, however, the effect was explained as a form of recovery from weight perturbation caused by mutations (Harvey, 1996, 1997). In this paper I provide additional data that show how the effect, at least in the case considered in the paper, can only be explained as a result of the interaction between learning and evolution as originally hypothesized. In a recent article Jeffrey Elman, Domenico Parisi, and I reported the results of a set of simulations in which neural networks that evolve (to become fitter at one task) at the population level may also learn (a different task) at the individual level (Nolfi, Elman & Parisi, 1994). In particular, individuals which were selected for their ability to find food in their environment were asked during their lifetime to learn to predict the sensory consequences of their motor actions. Each individual animat lives in a two-dimensional grid world where 10 food tokens are randomly distributed. Each food token occupies one cell; if the animat happens to enter one of these cells, the food token is automatically “eaten'' and the animat's fitness is increased. Figure 1. Neural network architecture. All connections are encoded into the genotype and inherited; however, connections represented with thin lines are also modified by prediction learning during the lifetime of the individual while connections represented with thick lines are not. Each individual is equipped with a neural network interfaced to a sensorimotor system that provides input information on the distance and angle (with respect to the facing direction of the animat) of the nearest food token and on the planned motor action. Two input units encode the angle and the distance of the nearest food token and two other units (thresholded