The neurophysiological processes related to reward and feedback processing have been extensively investigated for more than 30 years now. It has been shown that dopaminergic neurons encode a reward prediction error. In humans, feedback processing is reflected in a cascade of event-related potential (ERP) components. The frontocentral feedback-related negativity (FRN) is presumably generated by computations of the so-called reward system, specifically prediction error signals in midbrain structures and the frontal cortex, while other ERP components may reflect feedback processing outside of the reward system. In this symposium, we seek to integrate recent advances in feedback processing research to better understand how cognitive and affective contextual manipulations influence feedback processing and utilization. Hans Kirschner will describe neuro-computational mechanisms responsible for the counter-normative influence of task-irrelevant variability in reward magnitude and feedback salience on probabilistic learning. Maren Giersiepen will talk about how the sense of agency affects neural correlates of feedback processing within and beyond the reward system. Franz Wurm will discuss research investigating how prediction errors and their neurophysiological representation affect happiness. Christine Albrecht will present how the event associated with feedback can lead to a recruitment of alternative neural mechanisms during feedback processing as reflected in the N170 ERP component. Finally, Constanze Weber will show how feedback timing affects neurophysiological representations of prediction errors for both presented and omitted feedback. Taken together, this symposium will contribute to a better understanding of the factors that influence feedback processing and its downstream effects.

Optimal decision-making requires organisms to adaptively adjust their sensitivity to new information. While numerous studies demonstrate that humans can adaptively weight task-relevant information based on the stochasticity and volatility of the environment, less is known about the influence of task-irrelevant factors on choice behavior. Here, we used computational modeling and EEG - as a brain measure with high temporal resolution - to better understand mechanisms responsible for the influence of task-irrelevant variability in reward magnitude and feedback salience on probabilistic learning. Specifically, we investigated learning behavior in a variant of a probabilistic reversal learning task with different levels of noise, that introduced two types of task-irrelevant events: pay-out magnitudes were varied randomly and, occasionally, feedback presentation was enhanced by visual surprise. We found that participants’ learning performance was biased by distinct effects of these task-irrelevant factors. On the computational level, we show that both factors modulated trial-by-trial learning rate dynamics. In the EEG, these learning rate dynamics were reflected in a feedback-locked centroparietal positivity that also predicted behavioral adaptations. These results were replicated in an independent sample using a version of the task with reduced levels of noise. Interestingly, higher sensitivity to task-irrelevant factors was only negatively related to overall task performance in the task with high level of noise. Collectively, these data help to clarify the impact of task-irrelevant factors on probabilistic learning and suggest that these factors have a counter-normative influence on trial-by-trial learning rate dynamics.

Freedom of choice enhances the sense of agency over our actions and their effects, which has been suggested to facilitate learning. Yet, how agency experience influences feedback processing during goal-directed behavior is not clearly understood. We performed two EEG studies to examine how freedom of choice influences feedback processing during reinforcement learning. Participants performed self- (free-choice) or externally determined (forced-choice) binary item choices, followed by monetary gains or losses. Study 1 (N = 30) revealed augmented midfrontal theta power for losses compared to gains, suggesting an increase in cognitive control when processing negative outcomes. Importantly, theta power was heightened for free compared to forced choices, irrespective of feedback valence, indicating enhanced outcome monitoring for self-determined actions. Study 2 (N = 37) demonstrated that free choices also increase feedback-related ERPs, indicating wide-ranging effects of choice on outcome processing. Free compared to forced choices elicited a larger N100, reflecting intensified sensory feedback processing. Furthermore, only free choices elicited a Reward Positivity, suggesting a selective coding of outcome value for self-determined actions. Crucially, these effects emerged despite comparable learning between free and forced choices and irrespective of whether forced-choice outcomes were relevant for participants’ future actions. This suggests that outcome processing is driven by the association with self-determined choice rather than the relevance of outcomes for future actions. Our findings highlight the pivotal role of self-determination in tracking the consequences of our actions and advance our understanding of the cognitive processes underlying the choice-induced facilitation in outcome monitoring.

Feelings of happiness are thought to be driven by “how well things are going”. Interestingly, previous research suggests that momentary fluctuations in subjective well-being are also driven by outcome expectations and predictions errors – i.e., the difference between expected and received outcomes. Besides the replication of these behavioral observations, the aim of the current study is to complement earlier fMRI findings and investigate the role of well-known feedback-processing ERPs in subjective well-being. To this end, 65 participants performed a gambling task and a simple bandit task. In both tasks they make choices to receive outcomes and report on their momentary happiness across trials. Using computational modelling, we confirm that both outcomes and prediction errors drive happiness across tasks. We show that the reward positivity (RewP) and P300 components are modulated by current happiness and predict future happiness.

The preliminary findings of our pre-registered study demonstrate a clear connection between rewards, happiness and neural activity. Future analyses will focus on causal relations, testing if ERPs act as mediators between outcome (expectations) and happiness, and the role of learning in this process. Linking affective, cognitive and neural processing of rewards, this research could have profound implications for our understanding of the genesis and maintenance of mood disorders and related psychiatric symptoms.

Patient and fMRI studies suggest that the timing of performance feedback crucially determines where in the brain feedback is processed. While immediately delivered feedback relies on the striatum, delayed feedback appears to depend on hippocampal involvement. Challenging this distinction, we showed that computationally modeled prediction errors (PEs) are similarly encoded in neurophysiological responses (RewP and P300) to both immediate and delayed delivered rewards—but not losses. Building on the seminal observation of a dip in dopaminergic firing when expected rewards are omitted and on the assumption that omission-related ERPs can reveal correlates of the PE, we hypothesized that the omission of rewards and losses might have distinct neural representations. To investigate this, we examined EEG responses to the reception and omission of rewards and losses, which served as feedback during a learning task. Crucially, omissions conveyed either the omission of a reward or a loss depending on learning context. To examine differences in omission-related responses with feedback timing, participants completed the task with either immediate or delayed (omitted) feedback. Analyses of frontocentral and centroparietal omission-related responses indicate PE effects on both reward and loss omissions for immediate feedback. However, centroparietal effects of reward omissions were more tightly linked to the time at which a reward could have been delivered. For delayed feedback, only responses to reward (but not loss) omissions showed an encoding by PEs, especially centroparietal. While these findings also uncover differences evoked by feedback timing, they again emphasize striking similarities, e.g. with regard to their sensitivity to rewards.

Feedback learning probably engages two learning systems: the striatal reward system, as reflected in the FRN/RewP ERP component, and the medial temporal lobe (MTL). Recent research suggests that MTL activity during feedback learning may be reflected in the N170 ERP component, traditionally associated with higher-order visual processing in the fusiform gyrus. In terms of credit assignment, the MTL might play a role in reactivating a relevant event at feedback presentation, linking event and feedback, particularly when feedback is delayed. We hypothesized that the N170 reflects reactivation of visual stimuli at feedback presentation, assuming it to be most strongly involved in learning associations between visual stimuli and feedback. In contrast, due to its role in motor learning, the striatum, reflected in the FRN/RewP, may be more adept at forming action-feedback associations. In three studies, we examined whether (immediate or delayed) feedback processing depends on the type of association being learned, by manipulating the type of the associated event (actions vs. visual stimuli and visual vs. auditory stimuli) or the individual’s belief about what the feedback relates to. The results demonstrated that both FRN/RewP and N170 components are influenced by the type of learned association, although the data suggest that the striatal and MTL systems also cooperate. In particular, we found first evidence that the N170 reflects the reactivation of specifically visual material during feedback presentation. Additionally, the N170 was consistently modulated by prediction errors across all studies, which further suggests that the MTL and/or fusiform gyrus contribute to feedback learning.