Neural Noise- Reward and emotion prediction errors are separable at the neural level.1
- The finding is relevant for understanding neural coding of affect and could inform BCI decoding of emotional states.1 1
Weekly enrichment (2026-07-20)
- The study with this exact title was published in Nature Communications (2025, vol. 16) by Joseph Heffner, Romy Frömer, Matthew R. Nassar, and Oriel FeldmanHall (DOI 10.1038/s41467-025-63135-5).2
- The core question was whether emotion (affective) prediction errors are neurally separable from reward prediction errors; the authors recorded electroencephalography (EEG) while participants engaged in a social-learning task.2 3
- At the neural level the two error signals engaged distinct event-related potentials: the feedback-related negativity (FRN) was largely and consistently indexed by reward prediction errors, whereas the P3b was more consistently tracked by affective prediction errors.2
- The P3b (a later positive deflection, reported over a cluster spanning roughly 326–790 ms) was the primary neural predictor of subsequent social choice, highlighting a mechanistic role for emotion in decision-making.2 3
- Behaviorally, affective prediction errors predicted choices most when uncertainty about a partner was high (early in learning), and their influence gradually diminished as reward prediction errors became more predictive over time.2 3
- The authors interpret the early processing of reward PEs and later processing of affective PEs as evidence for a neurobiologically viable emotion-learning signal partially distinguishable from reward.2
- Prior behavioral work across four studies (N = 1,016) found emotion prediction errors exert the strongest impact on punish/forgive decisions, and that individuals with significant depression show a selective impairment in using emotion—but not reward—prediction errors, underscoring clinical relevance for affective-state decoding.4
- A complementary fMRI study (n = 43, modified Ultimatum Game) found reward and emotional prediction errors are encoded by distinct circuits—left dorsomedial prefrontal cortex during reward receipt versus bilateral anterior insula for emotion—with separable brain-wide multivariate patterns, reinforcing the separability seen in EEG.5