Reply to de Fleurian et al.: Toward a fuller understanding of reward prediction errors and their role in musical pleasure

We are pleased that our manuscript on the neural processes underlying musical pleasure and reward (1) has elicited commentary from de Fleurian et al. (2). Their first comment regards the distinction between reward prediction errors (RPEs), which concern predicted values like rewards/punishments, and sensory prediction errors, which concern predicted events in general (3). Whereas music aesthetic research often focuses on the latter (4), our findings indicate that value-based predictive processing also contributes to musical pleasure, in line with other recent data implicating the nucleus accumbens (NAc) in musical surprises (5). How sensory predictions might engage valuation processes is a pressing question in cognitive neuroscience, arising from evidence of RPE-exhibiting neurons facilitating error-driven learning even about nonrewarding sensory features such as unexpected auditory clicks (6), and reflecting RPE-like responses associated with learning nonrewarding information (7). These data suggest that reducing uncertainty might confer intrinsic value (8), such that RPEs might reflect the uncertainty-reducing value of unexpected sensory inputs, including certain musical events.

De Fleurian et al. (2) also question the source of RPEs in our experiment, since we cued predictions with visual rather than musical stimuli. We chose this approach to characterize the canonical RPE response in music because it has been well validated in many prior RPE-eliciting paradigms. For example, Gläscher et al. (9) used visual cues to elicit and provide feedback on predictions about monetary outcomes, finding a correlation between RPEs and NAc activity. These authors argue that the resulting RPEs arose from the money associated with the visual feedback, rather than the feedback itself, because the monetary outcomes were much more behaviorally salient. Likewise, our musical outcomes (harp/mandolin and consonant/dissonant versions of Bach chorales) were much more salient than the simple visual cues preceding them (2 colors/letters), and so our participants’ predictions pertained to musical rather than visual outcomes. Accordingly, our analyses examined the blood oxygen level-dependent response at the times of musical outcomes, when no new visual information appeared. Thus, while participants’ reward predictions began after visual cues, these predictions pertained to musical outcomes, and the RPEs we observed are hence best seen as responses to musical events.

Lastly, de Fleurian et al. (2) challenge the relationship we observed between neural RPE-related activity and learning. However, reanalyzing the data to circumvent the statistical issues of stepwise regression with forward and backward steps on all possible starting models (10), we have confirmed that learning is robustly related to neural RPE-like signaling. Moreover, the participant identified as an outlier due to a low RPE-related response and learning slope is the sole musical anhedonic in our sample—a trait associated with reduced NAc responses and decreased NAc−temporal lobe functional connectivity during music listening (11). We therefore chose to report this effect to motivate further research on this relationship, and invite others to contribute new empirical findings to the discussion. In conclusion, we are aware and pleased that our findings raise questions about the neural mechanisms of reward in music and other aesthetic stimuli, and we look forward to further data and debate to move our understanding forward.