Flow Intelligence: Knowledge Resilience Under Conditions of Information Degradation - Storage and Compression Series FI-SC-02 - FI-SC-09-1

AbstractThis paper examines semantic degradation as a systemic failure mode in knowledgesystems, with particular attention to recursive contamination in artificial intelligence.The paper introduces the concept of knowledge prion risk to describe how meaningcan degrade even when syntax remains intact. Drawing carefully scoped analogiesfrom biological prion diseases, it describes how systems trained on their ownoutputs—or on outputs from similar systems—propagate structural corruption overtime. The paper argues that downstream interventions such as filtering, alignmenttraining, or improved prompting cannot cure substrate contamination. The onlyeffective response is strict control of what enters the epistemic substrate—what thepaper terms epistemic hygiene. This analysis extends beyond AI systems to anyrecursive knowledge environment where outputs become inputs. The paper does notname specific vendors or propose specific architectures; it offers a diagnostic frameapplicable across contexts.Keywords: semantic degradation; prion risk; recursive training; AI safety; meaning collapse;contamination; epistemic hygiene; model collapse; recursive contamination; substratecorruption