Search Papers | Poster Sessions | All Posters
Poster B32 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Positive Reward Bias on Human Reinforcement Learning under Increased Dopaminergic Neurotransmission
Arnaud Zalta1,2 (), Vasilisa Skvortsova3, Samuel Hewitt3, Michael Moutoussis3, Matthew M. Nour3,6,7, Raymond J. Dolan3,6, Charles Findling4, Tobias U. Hauser5,8, Valentin Wyart1,2,8; 1Laboratoire de Neurosciences Cognitives et Computationnelles, Institut National de la Santé et de la Recherche Médicale, 2Département d’Études Cognitives, École Normale Supérieure, Université PSL, Paris, France, 3Max Planck UCL Centre for Computational Psychiatry and Ageing Research, University College London, London, UK, 4University of Geneva, Switzerland, 5Department for Psychiatry and Psychotherapy, German Center for Mental Health (DZPG), University of Tübingen, Germany, 6Wellcome Centre for Human Neuroimaging, University College London, London, UK., 7Department of Psychosis Studies, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, UK., 8equal senior authorship
Learning action values is key to maximize their effective payoff in uncertain reward environments. But how does dopamine affect this reinforcement learning (RL) process in humans? To test the hypothesis that increases in sustained dopamine concentration levels trigger a positive reward bias on human RL, we administered dopamine precursor L-DOPA to healthy adult volunteers performing a restless two-armed bandit task during a double-blind randomized placebo-controlled study. We found that L-DOPA decreases switching between volatile choice options. Using computational modelling, we show that L-DOPA decreases the learning rate and precision of RL but does not affect the policy used to choose between options. These learning effects of L-DOPA are best explained by a positive reward bias on recurrent neural networks (RNNs) trained to perform the same task.
Keywords: human reinforcement learning dopamine computation noise recurrent neural network