Why information theory offers a different way to understand autistic brains
Information theory emerged in the mid-twentieth century as a mathematical framework for analysing communication systems. Claude Shannon’s foundational work addressed how signals are transmitted, how much information a channel can carry, how noise degrades transmission, and how redundancy can protect against error. The framework is value-neutral by design. It describes the properties of systems without judging them.
A review published this month in Informatics and Health applies information theory to autism. The University of Haifa researchers synthesise existing neuroimaging and electrophysiological research through this lens, proposing that autistic neural processing can be understood in terms of entropy, mutual information, channel capacity, and noise. The result is a unified framework explaining connectivity differences, sensory atypicalities, and cognitive patterns — described in terms of signal properties, rather than deficits.
This represents a shift in the explanatory register. The dominant frameworks for understanding autism’s neurobiology have historically been clinical — connectivity “impairments,” processing “deficits,” developmental “abnormalities.” Information theory offers an alternative vocabulary. A brain with reduced channel capacity between cortical regions is not a broken brain. It is a brain with different transmission characteristics. A brain with altered entropy profiles is not a disordered brain. It is a brain whose signal complexity differs from population averages.
The distinction matters because explanatory frameworks shape what questions get asked and what interventions get proposed. A framework built on deficit and disorder naturally generates questions about remediation and intervention. A framework built on signal properties generates questions about how those properties manifest functionally — what they enable, what they constrain, how environments might be structured to work with rather than against them. So: coherence-first. Which is why I’ve taken interest in this research, and decided to highlight it.
Entropy, noise, and channel capacity — what these terms actually measure in neural systems and what it means in autistic brains
The review synthesises findings across several information-theoretic measures. Each describes a different property of neural signal processing. In this section, I’ll explain the core ones.
Entropy measures signal complexity — the degree of randomness or unpredictability in neural activity. Research consistently finds reduced entropy in autistic brains, particularly in frontal regions and the right hemisphere. Lower entropy indicates more predictable, less variable neural dynamics. The brain’s activity patterns are more constrained: less adaptable to changing conditions. This manifests functionally as reduced cognitive flexibility and difficulty adjusting to novel or unpredictable environments.
Mutual information measures how much information is shared between brain regions — the degree to which activity in one region predicts activity in another. Autistic brains show altered patterns of inter-regional coupling. The balance between segregation (distinct regions operating independently) and integration (regions sharing information) differs from neurotypical patterns. The result is less efficient information transfer across the network as a whole.
Channel capacity refers to the maximum rate at which information can be reliably transmitted between regions. Reduced channel capacity in autistic brains means less information passes between cortical areas per unit time. This affects predictive coding — the brain’s capacity to generate predictions about incoming sensory input and update those predictions based on error signals. When channel capacity is constrained, the prediction-error loop operates differently.
Noise and redundancy describe signal degradation and protective mechanisms. The review notes excessive neural noise in autistic processing — random fluctuation that obscures signal. Redundancy can protect against noise by encoding the same information multiple times, but maladaptive redundancy patterns may compound rather than compensate for transmission difficulties.
These measures converge on a picture of autistic neural processing characterised by reduced complexity, altered inter-regional communication, constrained transmission capacity, and different noise profiles. None of these descriptions require the concept of deficit. They describe properties of a system. Neutrally.
The cerebellum as a prediction engine — why it matters for autism beyond basic motor control
A significant highlight in the review for neurodivergent insight regards the cerebellum’s contribution to autistic neural processing — a finding underrepresented in popular discourse. Most people associate the cerebellum pretty much exclusively with motor coordination. Its role in cognition and prediction is less widely understood.
Structural and functional differences in the cerebellar vermis — particularly lobules VI and VII — appear consistently in autism research. These regions connect to cortical areas involved in higher-order cognition, not just motor control. The cerebellum functions as what the researchers term a “regulatory hub for temporal prediction and adaptive control.” It generates internal models of expected sensory input and motor output, enabling the brain to anticipate rather than merely react.
Altered cerebellar function affects this predictive capacity. The brain becomes more reactive, less anticipatory. Sensory input arrives without adequate prediction to contextualise it. Motor sequences require more conscious attention because automatic prediction fails to scaffold them. Social interaction — which relies heavily on predicting others’ behaviour milliseconds ahead — becomes more effortful.
The cerebellar findings integrate with the broader information-theoretic framework. Prediction is fundamentally an information-processing operation. It requires storing past information (active information storage), transmitting predictions across regions (channel capacity), and updating models based on error signals (mutual information between prediction and input). Cerebellar differences affect the substrate on which these operations run.
This positions sensory sensitivity, motor coordination difficulties, and social processing differences as related manifestations of the same underlying mechanism — altered predictive processing — rather than separate symptoms requiring separate explanations.
What happens when explanatory neurodiversity frameworks do not require the concept of assumed, inherent deficit
The review’s most significant contribution may be inadvertent. By adopting information theory as its explanatory framework, it demonstrates that autism’s neurobiological characteristics can be described without deficit language. The framework simply doesn’t require it. (This is not typical with typical neurodiversity statistics and research, whereby default navigating principles are medical model, deficit, etc.)
Information theory asks “what are the properties of this system?” not “what is wrong with this system?” Reduced entropy is a measurement. Altered channel capacity is a measurement. Different noise profiles are measurements. These describe how a system operates, not whether it operates correctly. The concept of “correct” operation has no place in information theory. There is only the question of what a system does.
This reveals something about deficit framing: it is a human choice embedded in explanatory models and complied with, not a human fact discovered in human brains. The same neuroimaging data can be described as “impaired connectivity” or as “different connectivity patterns.” The same EEG findings can be framed as “abnormal signal complexity” or as “reduced entropy.” The deficit is not in the data. It is in the vocabulary chosen to interpret the data. Thankfully, and significantly, the researchers did not use that vocabulary, which deserves a mention.
The researchers do not make this observation themselves. The paper occasionally lapses into clinical language — “abnormalities,” “disruptions” — no doubt naturally inherited and carried across from the studies it synthesises. But the theoretical apparatus they deploy does not depend on these terms. Information theory provides a complete description of autistic neural processing without ever requiring the claim that something is broken.
This matters for how autism is understood and how autistic people understand themselves. A framework that describes difference without pathology opens space for questions the deficit model forecloses. Not “how do we fix this?” but “how does this work?” Not “what is this person lacking?” but “what are the properties of this person’s processing?” The answers may point toward the same practical accommodations. But the framing determines whether those accommodations are understood as remediation of deficiency or as alignment between system and environment.
