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

Area of Science:

• Neuroscience
• Computational Psychiatry
• Autism Research

Background:

• Autism research increasingly utilizes predictive coding models.
• Competing theories offer different explanations for autistic perception.
• Distinguishing these theories requires precise computational modeling.

Purpose of the Study:

• To computationally model sensorimotor tasks in autism.
• To test predictions from competing neuro-computational theories.
• To clarify the nature of atypicalities in autistic perception.

Main Methods:

• Computational modeling of object lifting and anticipatory eye movements.
• Analysis of prior beliefs, sensory weighting, and prediction error processing.
• Investigating learning rate adaptation and environmental volatility encoding.

Main Results:

• No evidence of atypical prior use or sensory weighting in object lifting.
• Observed differences in prior beliefs, belief updating, and prediction error weighting for eye movements.
• Identified difficulties in adapting learning rates to environmental volatility in autism.

Conclusions:

• Atypical precision encoding and context-sensitive adjustments better explain autistic perception.
• Findings challenge theories of generic prior attenuation or persistent high prediction error.
• Results do not support the hypothesis of perceived persistent environmental volatility in autism.