Society Knows More Than It Can Use: Scientific Knowledge Integration as a Societal Steering Problem in the Age of AI

Science produces more knowledge than individual researchers, disciplines, and institutions can continuously process and integrate. This paper examines the resulting gap between knowledge production and knowledge integration as both a scientific and a societal steering problem. It distinguishes between visibility, citation, selective reception, deep reception, cumulative integration, and societal use of scientific knowledge. A central argument is that knowledge may be published, accessible, and even cited while remaining largely absent from the active cognitive maps through which science and society interpret complex problems. Generative AI changes this situation by enabling the synoptic analysis of large and widely dispersed bodies of text. AI is therefore considered not primarily as a producer of additional texts, but as a potential infrastructure for scientific knowledge integration, provenance reconstruction, longitudinal analysis, and collective memory. The paper also addresses major epistemic risks, including hallucination, false attribution, pseudo-coherence, retrospective teleology, and confirmation dynamics. Human source validation, conceptual judgment, and institutional responsibility therefore remain indispensable. Jean-Pol Martin’s body of work serves as a case study of how a theoretical architecture distributed across several decades may be reconstructed through human-AI cooperation. The guiding question is: Does humanity truly suffer from a lack of knowledge - or increasingly from a lack of integration of knowledge that already exists? What is new?The article reframes AI not primarily as a tool for producing new texts, but as a possible infrastructure for integrating existing scientific knowledge. Its central thesis is that modern societies may suffer less from a lack of knowledge than from a lack of integration of knowledge already available. Function within the seriesThe article extends the progression from LdL, basic needs, New Human Rights, the organism model, and human-AI cooperation toward a broader concept of collective societal cognition. It asks how AI-supported knowledge integration could help complex societies build coherent cognitive maps and improve their capacity for orientation and action.