Abstract

This paper proposes an extension to Predictive Processing (PP) and Active Inference frameworks by introducing Cognitive Erosion as a mechanism of progressive degradation in precision estimation under sustained cognitive load. While standard PP assumes stable and adaptive regulation of precision weighting across hierarchical inference, we argue that sustained informational pressure induces a systematic reduction in precision differentiation capacity. This leads to hierarchical flattening, reduced representational granularity, and a shift from optimal Bayesian inference toward compression-based satisficing. The proposed Cognitive Erosion in Predictive Processing (CE-PP) framework generates testable predictions concerning semantic overgeneralization, attenuated prediction-error sensitivity, and impaired hierarchical updating under load. Keywords Predictive Processing; Active Inference; Cognitive Load; Precision Weighting; Hierarchical Inference; Cognitive Erosion