Testing predictive coding theories of autism spectrum disorder using models of active inference

Abstract

Several competing neuro-computational theories of autism have emerged from predictive coding models of the brain. These accounts have a common focus on the relationship between prior beliefs and sensory inputs as a mechanism for explaining key features of autism, yet they differ in exactly how they characterise atypicalities in perception and action. We tested these competing predictions using computational modelling of two datasets that allowed us to probe both visual and motor aspects of active inference: manual gripping forces during object lifting and anticipatory eye movements during a naturalistic interception task. We compared estimated belief trajectories between autistic and neurotypical individuals to determine the underlying differences in active inference. We found no evidence of chronic deficits in the use of priors or weighting of sensory information during object lifting. Differences in prior beliefs, rates of belief updating, and the precision weighting of prediction errors were, however, observed for anticipatory eye movements. Notably, we observed autism-related difficulties in flexibly adapting learning rates in response to environmental change (i.e., volatility). These findings suggest that aberrant encoding of precision and context-sensitive adjustments provide a better explanation of autistic perception than generic attenuation of priors or persistently high precision prediction errors.

The relationship between environmental statistics and predictive gaze behaviour during a manual interception task: Eye movements as active inference

“…This hierarchical model can be further extended so that the rate of change of the volatility can itself change over time (creating a 4-level model), and so on. These additional hierarchical levels enable us to model more complex human learning in shifting environmental contexts (see Arthur et al, 2023 for an example with sensorimotor tasks).…”