Bayesian decision-making as a model of implicit motor adaptation

Abstract

The sensorimotor system is constantly tasked with deciding which errors to learn from and which to ignore. Recent work has shown that humans are able to accurately parse internal versus external sources of motor errors as small as 2 deg: Participants demonstrate robust implicit adaptation to the error component caused by a visuomotor rotation but no adaptation to size-matched errors that are due to intrinsic motor noise (Ranjan & Smith 2018). Following a successful replication of the main results of Ranjan and Smith involving 16 neurotypical adults, we formalized our understanding of this behavior by developing a novel Bayesian ideal observer model. The Parsing of Internal and External Causes of Error (PIECE) model frames implicit adaptation as a process of causal inference regarding the source of error, with the magnitude of motor corrections reflecting state estimation and the observer's belief that their movement was externally perturbed. This framework presents a challenge to an entire class of computational models that frames adaptation as a process of aligning the perceived hand position with the movement goal (Wei & Kording 2009, Tsay et al 2022, Zhang et al 2024). Only PIECE can accurately capture the precise parsing of internal versus external errors observed, as objective model selection criteria (Bayesian Information Criterion scores) favored PIECE model fits over the other three models for all 16 participants. Combined, our results provide a normative explanation of how the nervous system discounts intrinsic motor noise and adapts to small perturbations in order to keep movements finely calibrated.