Predictive coding links perception, action, and learning to emotions in music: Comment on "The quartet theory of human emotions: An integrative and neurofunctional model" by S. Koelsch et al.

The review by Koelsch and colleagues [1] offers a timely, comprehensive, and anatomically detailed framework for understanding the neural correlates of human emotions. The authors describe emotion in a framework of four affect systems, which are linked to effector systems, and higher order cognitive functions. This is elegantly demonstrated through the example of music; a realm for exploring emotions in a domain, that can be independent of language but still highly relevant for understanding human emotions [2]. Emotion is fundamental to human life, survival and well-being [3] and music is one of the strongest and most universal sources of human emotion and pleasure [4–10]. Music also highlights a particularly interesting aspect of human emotions: the dynamical interplay between perception, action and learning, and emotion. Here, the review is less explicit in describing the mechanism by which the interaction between these systems takes place. Novel models of brain function have emerged such as the predictive coding theories [11–17] proposed to be general theories of brain function [18], explaining how brain areas exchange information. Such models offer a novel perspective on how specialized brain networks can identify and categorize causes of sensory inputs, integrate information within other networks, and adapt to new stimuli. They propose that the quartet of perception, action, learning and emotion occurs in a recursive Bayesian process by which the brain tries to minimize the error between the input and the brain’s expectation. Within this framework Perception can be described as the process of minimizing prediction errors between higher-level “prediction units” and lower-level “error units” in the hierarchically organized brain; Action is the active engagement of the motor system to resample the environment in order to reduce prediction error; Emotion acts as weight or modulator of the prediction error itself, guiding behavior, action and learning through neurotransmitters such as dopamine [19]; while Learning is the long-term influence on the prediction units [17,18].