Action Change Theory: A Reinforcement Learning Perspective on Behavior Change

Abstract

Traditional theories of behavior change rely mostly on influencing higher-order mental processes as a route to altering deliberate responses, whereas more recent theorizing postulates that interventions can also rely on using contextual cues influencing lower-order processes as a route to changing spontaneous responses. We propose an alternative mechanistic account based on reinforcement learning theory, which utilizes different action control systems in the brain. Therefore, this account works at a different level of analysis and description, which promises to lead to the development of a more general and integrative theory of behavior change. Reward systems generate specific affective states that influence behavior via 3 action controllers. Innate actions are stereotyped evolutionarily determined responses to stimuli. Habitual actions develop through stimulus-response learning without explicit outcome representations. Goal-directed actions are based on an explicit model of the structure of the environment, which utilizes computations of action-outcome contingencies. We describe how these mechanisms for action control parsimoniously explain behavior change theories and techniques.