Precision Weighting in Predictive Coding Frameworks: A Neurocognitive Account of Musical Expertise in Autism

Abstract

Exceptional musical ability has repeatedly been reported among autistic individuals. Absolute pitch occurs in under 1% of the general population but in approximately 5–11% of autistic individuals, and enhancements have been documented in pitch discrimination, melodic memory, and the recognition of emotion conveyed by music. In some autistic individuals, these abilities exceed those of non-autistic listeners, yet their neural mechanism remains unresolved. This study develops that mechanism within the predictive coding framework. Precision accounts of autism converge on a setting in which the inverse-variance weight, or precision, assigned to sensory prediction errors is elevated and insufficiently modulated by context. This setting is proposed to produce musical expertise along a chain running from the gain on prediction-error units in auditory cortex, through the population-level precision of pitch representation and locally hyperconnected network organization, to perceptual-categorical learning through veridical mapping. Because precision is a continuous parameter and different musical abilities depend on different levels of the chain, from enhanced recognition of emotion in music at the first level to absolute pitch at the fourth, outstanding musical ability would be expected in only a subset of individuals. The chain is likely to operate only where the statistical structure of the input is stable. Music, being hierarchical, repetitive, and governed by a closed rule system, satisfies that condition more fully than speech, whose grammatical regularities coexist with variation across talkers and contexts, and social interaction, so musical strength and social difficulty can both be understood as consequences of the same precision setting operating in two statistical environments. The same setting would impose a cost on rhythmic entrainment, which requires sensory attenuation. Four testable predictions are derived from the account.