Computational Imprecision in Value-Based Learning and Decision-Making: Investigating the Potential Role of the Noradrenergic Locus Coeruleus System

Human decision-making in the context of value-based learning considerably deviates from the premise to always maximise short-term reward. This sort of behavioural variability has been attributed to exploration during the choice process and has been linked to the locus coeruleusnoradrenaline (LC-NA) system. A recent computational account implementing a “noisy” reinforcement learning (RL) model provides evidence that computational imprecision during the learning process contributes to behavioural variability in decision-making as well. In the present study, the role of the LC-NA system in modulating learning imprecision has been investigated, using a multi-modal approach including the newly developed noisy RL model, a double-blind pharmacological manipulation, and EEG. The responses of thirty participants in a decision-making task were investigated during two experimental sessions in which either atomoxetine, a NA reuptake inhibitor, or placebo was administered, and the model fit of...Show moreHuman decision-making in the context of value-based learning considerably deviates from the premise to always maximise short-term reward. This sort of behavioural variability has been attributed to exploration during the choice process and has been linked to the locus coeruleusnoradrenaline (LC-NA) system. A recent computational account implementing a “noisy” reinforcement learning (RL) model provides evidence that computational imprecision during the learning process contributes to behavioural variability in decision-making as well. In the present study, the role of the LC-NA system in modulating learning imprecision has been investigated, using a multi-modal approach including the newly developed noisy RL model, a double-blind pharmacological manipulation, and EEG. The responses of thirty participants in a decision-making task were investigated during two experimental sessions in which either atomoxetine, a NA reuptake inhibitor, or placebo was administered, and the model fit of the noisy RL model was tested against the classical RL model without learning imprecision parameter. Contrary to our expectations, we did not find evidence for a modulatory role of the LC-NA system of learning imprecision. However, in line with our expectations, increased NA levels did not impact exploratory behaviour either. Crucially, the noisy RL model outperformed the classical RL model assuming exact learning and learning imprecision led to a significant number of non-maximising decisions contributing to behaviour variability. Moreover, the exploratory EEG analyses on potential underlying mechanisms of learning imprecision suggest learning noise to emerge broadly during feedback processing, rather than within a certain time window. The current study provides further evidence for the importance of computational imprecision during value-based learning.Show less

Leiden University Student Repository

Documents

In Collections

Computational Imprecision in Value-Based Learning and Decision-Making: Investigating the Potential Role of the Noradrenergic Locus Coeruleus System