Exploring adaptive agency III: Simulating the evolution of habituation and sensitization

Abstract

Sensitization and habituation, we postulate, both serve the adaptive function of cluster-tracking: entraining and exploiting the basic spatio-temporal regularities in the environment. To better understand the adaptive pressures shaping cluster-tracking, we used a genetic algorithm to evolve simulated creatures controlled by neural networks. The creatures make decisions about when to eat in simple simulated environments containing ‘food’ (which raises fitness) and ‘poison’ (which lowers it) based on sensory cues. Food and poison were distributed in randomly-occurring clusters of a certain scale fixed for each environment. Sensory input had a limited accuracy level fixed for each environment. When sensory accuracy is moderate and food and poison come in fairly large clusters, certain time-delay feedback connections evolve to allow cluster-tracking. We ran several simulations for each of 6 cluster-scales and each of 7 levels of sensory accuracy. As expected, the average number of generations required to evolve cluster-tracking follows a U-shaped curve as a function of sensory accuracy, and generally declines as cluster scale increases. But an asymmetry in this ravine-like surface illuminates some previously unsuspected complexities of sensitization and habituation.