Does ChatGPT Have Law Sense? Reflections on an Experimental Study of Argument Mining in Judicial Decisions of the Italian Court of Cassation

Abstract

This article examines the potential and limits of Large Language Models in judicial practice, taking as its test case the reconstruction of legal argumentation within judicial reasoning. Through an argument mining experiment on decisions of the Italian Supreme Court of Cassation, conducted with GPT-4o and interpreted through Philip Bobbitt’s theory, the paper asks whether an LLM can grasp not only the textual dimension of a judicial decision, but also its rhetorical form. The results show a concrete usefulness in preliminary tasks of segmentation and textual organ-ization, but also a structural limitation in recognizing ethical argumentation and, more broadly, the law sense required by legal judgment.