Abstract

Everyday work with large language models (LLMs) normalizes a practice in which a generated formulation supports judgment before the user has checked its grounds. This crisis of rationality arises not from a single technical defect in the system, but from a shift in justificatory practice in which fluent textual coherence is read as evidence of semantic understanding and rational judgment. The method brings conceptual analysis into contact with 4E cognition and a postphenomenological account of technological mediation. Within this framework, LLMs are described neither as autonomous rational subjects nor as neutral instruments, but as multistable moral-epistemic mediators of human rationality. The analysis distinguishes textual competence from world-involved understanding and relates interface mediation to trust and responsibility. On this basis, the article proposes a four-cluster protocol for the attribution of rationality, which introduces an epistemic pause as a route of verification between a generated formulation and its authorial acceptance as a claim. The central risk lies not in whether machine consciousness has been proven, but in the normalization of practices in which ready-made text acquires the status of a ground before the user has reconstructed its sources and accepted responsibility for what is asserted.

IPC Classification

Keywords