Gaston Rey has posted Generative AI and Cognitive Delegation in Legal Practice on SSRN. Here is the abstract:
Recent decisions around the world—Mata v. Avianca in the United States, the Sullivan & Cromwell apology in the U.S. bankruptcy courts, and an emerging Argentine line of cases—have made clear that the problem posed by generative artificial intelligence in legal practice is not, properly speaking, the technology itself. It is the structural transformation in how legal text is produced: an unprecedented capacity to outsource professional cognitive functions (the search for authority, the testing of arguments, the very drafting of pleadings), to probabilistic systems whose outputs cannot be authenticated from within. This article advances the thesis that the central legal-ethical problem is cognitive delegation, not artificial intelligence; that the danger of generative tools in litigation is not their tendency to produce absurd outputs but their fluency in producing plausible ones; and that the lawyer’s signature is best understood not as a procedural formality but as a mechanism of institutional imputation. Generative AI does not merely automate drafting. It destabilizes the institutional relationship between legal text, professional judgment, and responsibility. The argument proceeds through a comparative reading of the international cases, an analysis of the architecture of plausibility in large language models, a critique of the ‘human in the loop’ formula, a survey of emerging professional standards, and a discussion of when negligent reliance on AI may cross into procedural bad faith.
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum