Bayesian Behavioral Systems Theory

Abstract

Behavioral Systems Theory suggests that observable behavior is embedded in a hierarchy. A CS elicits behavior because, after learning, it activates a pathway through this hierarchy. Much of Timberlake's body of work on Behavioral Systems Theory focuses on the conditions that support the conditioning of these pathways. Most notably, his work shows that the identity of the CS, US, and the CS–US interval all help support conditioning of the system. Here, we use recent experiments in the interval timing literature to motivate a Bayesian implementation of Behavioral Systems Theory. There is a probability distribution over possible pathways through the hierarchy, and the one that maximizes reinforcement is elicited. This probability distribution is conditioned on background information, like the CS–US interval and the animal's motivational state. Lower level actions of the hierarchy, like tracking prey, are conditioned on higher level goals, like the general search for food. Our implementation of Behavioral Systems Theory captures the essential features of Timberlake's verbal model; it acts as a glue, integrating sensory, timing, and decision mechanisms with observed behavior.