Computing Akrasia: Toward a New Framework for Understanding Addiction

Abstract

Akrasia, the failure to act according to one’s better judgment, originates from Aristotle’s Nicomachean Ethics and remains a key concept in understanding addiction. Addiction exemplifies a state where individuals persist in harmful behaviors despite recognizing their consequences, raising questions about agency and self-control. Computational neuroscience offers two dominant models of addiction: Predictive Processing (PP) and Reinforcement Learning (RL). PP explains addiction as a dysfunction in predictive coding, where maladaptive precision weighting reinforces rigid priors, making drug-seeking behaviors resistant to change. RL, in contrast, describes addiction as a shift from goal-directed (model-based) to habitual (model-free) learning, driven by dopaminergic reinforcement. However, PP struggles to account for compulsions, while RL overlooks inferential processes. The Arbitration Model addresses these gaps by regulating model-free and model-based learning based on reliability estimates. By integrating Aristotelian philosophy with computational models, this framework offers a comprehensive account of addiction as a dysfunction in decision-making and self-regulation.