Why "underweighting priors" is too simple
Predictive processing describes perception as a negotiation between what the brain expects (prior beliefs) and what the senses report. A popular account of autism proposed that autistic people give too little weight to expectations and too much to raw sensory input, which would make the world feel less predictable. The evidence does not support this as a blanket description. In a duration-reproduction task, autistic adults integrated prior knowledge into their perceptual estimates similarly to typical participants; what differed was how heavily they relied on sensory input when updating their priors over time [2].
Tactile experiments point the same way. Autistic adults showed no difference in average performance or sensory sensitivity, and both groups learned an implicit prior that biased their perception. What varied was the adjustment of that prior to context: depending on the condition, autistic participants showed no adaptation, normal adaptation, or slower adaptation [4]. That pattern fits a difference in how flexibly expectations are revised, not a fixed refusal to use them.
What brain responses reveal about anticipation
Electrophysiological studies add a sensory-specific wrinkle. When a button press produced a tone, typical participants showed a reduced early auditory response to self-initiated sounds, a sign that the brain anticipated the sound's consequences; autistic participants did not show this attenuation, suggesting weaker anticipation of the auditory results of their own actions [1]. When a visually predicted sound was unexpectedly omitted, autistic participants produced a larger early omission response, consistent with a stronger prediction error when an expectation was violated [1].
These findings do not say that autistic people lack predictions. They suggest that the anticipation of upcoming sound may be altered, and the authors argue this may reflect domain-specific changes in auditory processing rather than a general failure of predictive coding [1]. That distinction matters: a general underweighting account would predict broad differences across senses, whereas the auditory results sit alongside intact visual-speech deviancy detection [1].
The difference may be in timing, not weight
The updating account helps reconcile studies that seemed to disagree. In the duration task, autistic participants initially underweighted priors, producing a weaker pull toward the average early in the session, but by the end of the session their integration weights matched those of typical participants [2]. In other words, the same final weighting can be reached by a different route, and a snapshot taken early could look like a permanent deficit.
A similar theme appears in tactile learning, where a higher expectation for sameness was linked to slower or absent adaptation in some contexts [4]. The authors describe this as atypical perceptual learning rather than altered perceptual inference itself [4]. That is a meaningful shift in what needs explaining: not whether priors exist, but how quickly and under what conditions they are revised.
What rejecting the hypothesis means for predictive processing
Rejecting the simple underweighting hypothesis does not reject predictive processing. It narrows the claim. The framework still offers a language for describing autism in terms of precision, prediction error, and hierarchical signaling, but the evidence favors accounts built around iterative updating and context-sensitive adjustment rather than a single global imbalance [2][4].
Some neural evidence is consistent with an imbalance between sensory evidence and predictive feedback, though the measurement is indirect. During rhythmic visual stimulation, typical adults showed more backward (frontal-to-occipital) waves, proposed to carry top-down predictions, while autistic adults showed a pronounced increase in forward waves at the entrained frequency, interpreted as stronger bottom-up sensory drive [3]. The authors frame this as support for predictive coding accounts of autistic perception, but it is a pattern of brain oscillations during entrainment, not a direct measure of how much a person weights a prior.
The practical upshot is that predictive-processing theories of autism need to specify which stage is atypical, in which modality, and over what timescale. A theory that predicts uniform underweighting across all contexts is not well supported; a theory that predicts slower or context-dependent updating, with possible domain-specific effects in audition, fits the current evidence better [1][2][4].
Sources used in this answer
Predictive coding in autism spectrum disorder: Electrophysiological alterations in early auditory predictive processing as potential markers for autistic symptomatology
Autistic participants showed reduced attenuation of the auditory N1 for self-initiated tones and a larger early omission response to unexpectedly omitted predicted sounds, suggesting altered anticipation of auditory stimulation that may be domain-specific rather than a general predictive-coding impairment.
Predictive Processing in Autism Spectrum Disorder: The Atypical Iterative Prior Updating Account
Autistic adults integrated prior knowledge similarly to typical controls in perceptual estimates but relied more on sensory input when updating priors, causing early underweighting that resolved to typical integration weights by the end of the session.
Oscillatory traveling waves during visual entrainment in autistic and neurotypical adults
During visual entrainment, autistic adults showed increased forward (occipital-to-frontal) traveling waves while typical adults showed increased backward waves, a pattern interpreted as stronger bottom-up sensory drive relative to predictive feedback.
Disentangling sensory precision and prior expectation of change in autism during tactile discrimination
Autistic adults showed no difference in average tactile performance or sensory sensitivity and could learn an implicit prior, but their adaptation of precision weights to context was absent, normal, or slower depending on condition, consistent with atypical perceptual learning rather than altered inference per se.
Maladaptive Precision: A Channel-Specific Predictive-Processing Framework for the Neurodevelopmental Consequences of Early Life Adversity
The paper proposes a channel-specific predictive-processing framework for early life adversity and contrasts it with global precision accounts of autism, rejecting the idea that global underweighting of prior expectations explains autistic symptoms.