Understanding Is Not in the Text. Meaning, Interpretation, and Artificial Intelligence

This article engages critically with a recent argument by Walter Quattrociocchi concerning the epistemic risks posed by Artificial Intelligence and Large Language Models. Quattrociocchi suggests that the real danger is not that these systems are becoming intelligent, but that we are gradually redefining the concept of intelligence in order to accommodate their outputs. While agreeing that language models generate linguistic fluency without genuine access to the world-and therefore without epistemic responsibility-this paper questions the assumption that knowledge necessarily requires direct embodied contact with reality. Much of human knowledge already operates through symbolic coherence, inference, and socially validated linguistic networks rather than firsthand experience. From this perspective, AI does not simply distort the concept of intelligence but reveals its distributed structure: intelligence emerges through language, tools, and collective validation practices rather than residing entirely within an individual mind. The crucial novelty introduced by Large Language Models is the separation between the generation of meaningful linguistic structures and epistemic commitment to reality. The central claim of the paper is therefore that understanding is not located in textual fluency itself but in the responsibility for meaning within epistemic practices.