Large language models are increasingly designed to maximize coherence, fluency, and user satisfaction. This paper argues that such optimization creates a structural epistemic risk: the premature narrowing of the interpretive field available to human agents before they decide and act. It names this process the Coherence Paradox. The paper introduces Plurity as the field of interpretive possibilities available prior to decision and proposes the Ontological Thermodynamics of Knowledge as a conceptual framework for analyzing the relation between Plurity, Coherence, Cognitive Energy, and Knowledge Richness. From this perspective, the problem of contemporary AI is not only misinformation or misalignment, but the progressive compression of semantic and deliberative diversity under conditions of cognitive and economic optimization. The article develops a normative account of human–AI symbiosis in which AI should function as an explorer of possibility spaces rather than as a machine for premature closure, while humans remain responsible for contextual integration and action. The contribution is theoretical: to articulate a framework for evaluating AI systems not only by correctness or efficiency, but also by their capacity to preserve epistemic plurality and support the long-term ecology of knowledge.
Alexis Arellano (Wed,) studied this question.