Predictive Formats and Informational Attenuation: A Unified Account of Autism Across Cognition, Phenomenology, and Social Interaction

Persistent differences between autistic and neurotypical cognition are widely documented across perception, reasoning, communication, and social interaction, yet existing explanations remain fragmented across disciplinary levels. This paper proposes a unified account grounded in predictive processing, information theory, representational format, and phenomenology. It explains these differences as systematic consequences of how cognitive systems regulate and selectively attenuate information prior to explicit inference. We argue that the critical distinction does not concern the capacity for prediction as such, but the manner in which predictive systems attenuate low-probability, high-information signals. Informational attenuation is introduced as a common organizing mechanism linking multiple explanatory levels. At the computational level, it defines a principled trade-off between informational fidelity and predictive efficiency, grounded in information-theoretic relations between probability, surprisal, and compression. At the algorithmic level, this trade-off is implemented through precision weighting and thresholding processes that determine which signals are propagated, suppressed, or preserved. At the representational level, differential attenuation shapes predictive format itself, biasing cognition toward either inductively compressed content or explicitly maintained formal structure. At the phenomenological level, the same mechanisms account for systematic differences in experiential organization, explaining why some forms of cognition are characterized by transparency and immediacy, while others involve heightened explicitness and structural salience. At the social and interactional level, differential attenuation explains how inductively stabilized norms support epistemic stability—and why structurally oriented agents may experience such normativity as opaque, underdetermined, or coercive. By linking these levels through a single mechanistic principle, the proposed framework avoids both reductionism and explanatory fragmentation. Autism is thereby understood not as a localized cognitive deficit nor as a loose collection of symptoms, but as a coherent reconfiguration of predictive instantiation and informational priorities.