Giulia Santelli
Affiliation: Università degli studi Roma tre
Category: Philosophy
Keywords: Akrasia, Addiction
Date: Tuesday 2nd of September
Time: 17:00
Location: Gen. Henryk Dąbrowski Hall (006)
View the full session: Akrasia & Self-Control
Akrasia, or weakness of will, describes the paradoxical condition in which individuals act against their better judgment, persisting in behaviors they explicitly recognize as harmful. While traditionally examined in philosophical ethics, akrasia also presents a significant challenge for computational and neuroscientific models of decision-making, particularly in the context of addiction. Substance use disorders epitomize a case where agents persist in self-destructive behaviors despite an awareness of their detrimental consequences, raising fundamental questions about self-control, agency, and rational action. Why do rational agents knowingly choose actions that contradict their evaluative judgments? How do cognitive and neural mechanisms give rise to such inconsistencies? Recent advances in computational neuroscience have framed addiction as a dysfunction in predictive and reinforcement-based learning systems, yet existing models struggle to reconcile how impaired decision-making arises while the agent retains the ability to evaluate the consequences of their actions (Mollick & Kober, 2020). This paper argues that the Arbitration Model (Lee et al., 2014) provides a compelling explanation for addiction as a case of akrasia by describing the dynamic regulation between model-free and model-based decision-making. By integrating insights from Predictive Processing (Friston, 2012; Hohwy, 2013) and Reinforcement Learning (Sutton & Barto, 1998), the model captures how maladaptive learning dynamics override rational deliberation, leading to a persistent failure to act in accordance with one’s better judgment. Predictive Processing conceptualizes addiction as a dysfunction in precision-weighting, where excessive confidence in top-down priors reinforces compulsive behaviors, making them resistant to new information (Schwartenbeck et al., 2015; Pezzulo, Rigoli, & Friston, 2015). Conversely, Reinforcement Learning frames addiction as a shift from goal-directed to habitual decision-making, driven by the dysregulation of dopaminergic reinforcement mechanisms (Everitt & Robbins, 2005; Redish, 2004). While both models provide valuable insights, neither fully accounts for the conflict inherent in akrasia -how agents can explicitly recognize the suboptimality of their actions yet remain unable to correct them. The Arbitration Model offers a solution by introducing a regulatory mechanism that determines whether behavior is governed by model-free or model-based strategies at any given moment (Kim et al., 2024). This dynamic allocation depends on factors such as the reliability of predictions, computational cost, and reinforcement contingencies. In addiction, the arbitration system misattributes excessive uncertainty to model-based learning, causing a persistent reliance on habitual drug-seeking behaviors (Lee et al., 2014). Even when the agent acknowledges the long-term harm of substance use, their decision-making is dominated by low-cost, high-certainty model-free mechanisms. Neurobiologically, impairments in the arbitration system are linked to weakened connectivity between the ventrolateral prefrontal cortex and the putamen, regions responsible for balancing goal-directed and habitual learning (Kim et al., 2024). This dysfunction mirrors the way akrasia is traditionally conceived as a failure of rational governance over desire—the agent possesses the correct evaluative judgment but fails to act accordingly (Erginel, 2016). By framing addiction as a modern instantiation of akrasia, the Arbitration Model bridges the gap between normative philosophical theories of rationality and empirical neuroscientific models of decision-making. Unlike traditional accounts that emphasize a simple dichotomy between impulsive and reflective systems, this model presents a more dynamic understanding of how decision-making strategies are continuously reweighted based on uncertainty and cognitive demand. It challenges the conventional notion that addiction is a mere failure of willpower and instead suggests that akrasia emerges from systematic misattributions of uncertainty, which bias the arbitration process toward habitual responses (Radoilska & Fletcher, 2016). This framework has significant implications for addiction treatment, suggesting that interventions should target the arbitration mechanism itself -by strengthening model-based learning through cognitive training or neuromodulatory interventions- rather than merely suppressing cravings (Friston, 2012; Miller et al., 2020). Furthermore, this approach invites a reconsideration of classical philosophical questions about agency and self-control. Aristotle’s distinction between akrasia rooted in bodily pleasure and akrasia arising from other passions offers a rich conceptual parallel to the model-free versus model-based distinction in decision-making (Donini, 1977). This interdisciplinary perspective highlights how the ancient puzzle of akrasia continues to illuminate the complexities of human behavior, morality, and rationality. By integrating computational models with philosophical insights, we propose a comprehensive account of addiction that respects both the mechanistic underpinnings of decision-making and the lived experience of those struggling with self-control. In doing so, we demonstrate how bridging the ethical concerns of akrasia with the computational models of modern neuroscience can offer new pathways for theoretical inquiry and practical intervention. Key references: Donini, P. (1977). Incontinenza e sillogismo pratico nell'Etica Nicomachea. Rivista Critica di Storia della Filosofia, 32(2), 174-194. Erginel, M. M. (2016). Akrasia and Conflict in the Nicomachean Ethics. British Journal for the History of Philosophy. DOI: 10.1080/09608788.2016.1176890. Everitt, B. J., & Robbins, T. W. (2005). Neural systems of reinforcement for drug addiction: from actions to habits to compulsion. Nature Neuroscience, 8(11), 1481-1489. Friston, K. (2012). Policies and Priors. In B. Gutkin & S. H. Ahmed (Eds.), Computational Neuroscience of Drug Addiction (pp. 237-248). Springer. Hohwy, J. (2013). The Predictive Mind. Oxford University Press. Kim, J., Lee, S. W., & O'Doherty, J. P. (2024). A Bayesian Arbitration Model for Dynamic Decision-Making Between Habitual and Goal-Directed Systems. Nature Communications. Lee, S. W., Shimojo, S., & O’Doherty, J. P. (2014). Neural computations underlying arbitration between model-based and model-free learning. Neuron, 81(3), 687-699. Miller, M., Kiverstein, J., & Rietveld, E. (2020). Embodying addiction: A predictive processing account. Brain and Cognition, 138, 105495. Mollick, J. A., & Kober, H. (2020). Computational Models of Drug Use and Addiction: A Review. Journal of Abnormal Psychology, 129(6), 544-555. Pezzulo, G., Rigoli, F., & Friston, K. (2015). Active inference, homeostatic regulation, and adaptive behavioural control. Progress in Neurobiology, 134, 17-35. Radoilska, L., & Fletcher, K. D. (2016). Lessons from Akrasia in Substance Misuse: A Clinicophilosophical Discussion. BJPsych Advances, 22, 234–241. Redish, A. D. (2004). Addiction as a Computational Process Gone Awry. Science, 306(5703), 1944-1947. Schwartenbeck, P., FitzGerald, T. H. B., Mathys, C., Dolan, R., & Friston, K. (2015). The dopaminergic midbrain encodes the expected certainty about desired outcomes. Cerebral Cortex, 25(10), 3434-3445. Sutton, R. S., & Barto, A. G. (1998). Reinforcement Learning: An Introduction. MIT Press.