An Active Inference Model of Meter Perception and the Urge to Move to Music

TL;DR

This paper explores why certain rhythms make us want to move by using a Bayesian model to show that prediction errors and their reduction through movement drive the urge to move to music.

Contribution

The paper introduces a Bayesian model that operationalizes the predictive processing account of meter perception and movement urge.

Findings

Surprisal increases linearly with rhythmic complexity.

Delta surprisal correlates with urge to move ratings confirmed in an online study.

The urge to move is driven by expected reduction in prediction errors via movement feedback.

Abstract

Why do some rhythms make us want to move and not others? A predictive processing account suggests that prediction errors drive this phenomenon, but this hypothesis remains underspecified. Here, we operationalized this account as a Bayesian model that infers whether a rhythmic sequence is caused by a metered or unmetered template, coupled with an active inference rule in which movement occurs if the sensory feedback from movement would reduce the prediction errors generated by this inference process. Surprisal, as an index of prediction error, was calculated for each rhythm with and without a metronome (a proxy for the feedback from moving along), with delta surprisal as the difference. Surprisal increased linearly as a function of rhythmic complexity, while delta surprisal showed a similar pattern with urge to move ratings shown in previous studies, and this relation was confirmed in an online study. These results suggest that the urge to move to music is driven by the potential to reduce meter‐based prediction errors via the expected feedback from moving along to the beat. This work provides a crucial update to the predictive processing account and highlights a key role of active inference and prediction‐based learning in our musical experiences.

Prominent theories suggest that the urge to move along to rhythmic music is driven by precision‐weighted prediction errors. We operationalized this account as a Bayesian model which outputs surprisal as an index of prediction errors based on posterior probabilities calculated over metered and unmetered priors. Our results suggest that it is the expected reduction in prediction errors due to feedback from on‐beat movements that drives the urge to move.

Figures