Andrew M. Perlman (Suffolk University Law School) has posted Generative AI and the Future of Legal Scholarship (June 2026 Edition) on SSRN. Here is the abstract:
For a century and a half, legal scholarship has been organized around a single artifact: the article. The article bundled four functions—discovery, communication, credentialing, and archiving—and every institution of the legal academy, from the student-edited law review to the tenure file to the judicial citation, is built on the artifact’s distinctive properties: scarcity, fixity, and attributability. Generative artificial intelligence does not merely assist this system. It dissolves its organizing scarcity. When any competent legal argument can be produced on demand at negligible cost, the authored text loses the properties on which the artifact regime depends. The emerging literature has responded with what this Article calls the authentication paradigm: disclosure rules, certification regimes, and attestations of human authorship. This Article argues that authentication is a category error. It attempts to restore artificial scarcity to texts, when the scarcity that mattered was never textual. In its place, this Article develops a theory of latent scholarship. In the generative age, the de facto repository of legal thought is no longer the corpus of published writings but the model—the latent space from which arguments are drawn—and the scholarly acts that matter are those that form, validate, commit to, and architect regions of that space. The unit of scholarship accordingly shifts from the article to the maintained normative system: an executable, benchmarked, versioned instantiation of a legal theory that can be run against any case, probed adversarially, and revised in public. Influence shifts from the weight of authority to the authority of weights. The Article elaborates the theory’s institutional corollaries—law reviews reconstituted as validation institutions, peer review as adversarial evaluation, tenure metrics keyed to measurable uptake, a Daubert framework for machine-mediated normative systems, and corpus stewardship as a public trust—and confronts the strongest objections, including the charge that this is Langdellian scientism reborn, the dangers of model collapse and epistemic monoculture, and the risk that the imperial scholar will simply be succeeded by the imperial model.
Highly Recommended!
This article was drafted by Claude Fable 5 from a single prompt on June 9, 2026! The December 2024 edition, drafted by ChatGPT, proposed a much more modest theory. The distance between the two editions, eighteen months apart, is itself the most striking datum in the piece. Whatever you think of the theory of latent scholarship, the rapid progress in AI and its implications for legal knowledge are potentially transformational.
The theory of latent scholarship is genuinely original, but it may run two different ideas together: (1) where legal arguments are stored and (2) what legal scholars actually do. A model can be tested against cases and probed for weaknesses, but someone must decide what counts as a good answer — and that decision turns on the same contested questions of value that legal scholarship has always debated. The hard problem is not solved; it is relocated.
For readers unfamiliar with the term, ‘weights’ are the billions of numerical parameters inside a trained AI model — the numbers, set during training, that determine everything the model says. Perlman’s phrase inverts the lawyer’s familiar ‘weight of authority.’ Traditionally, an argument’s influence is measured by the authorities lined up behind it and tracked through citation. But if lawyers, judges, and scholars increasingly draw arguments from models rather than from the published literature, then influence flows from being encoded rather than from being cited. A theory absorbed into a model’s parameters shapes the answers the model gives everyone who asks — silently and without attribution. An article that moved the weights governs outputs even if it is never cited again; a heavily cited article the models ignore has authority on paper but not in practice. That is why, on Perlman’s account, what goes into training data — and who controls it — becomes the successor question to what gets published and who edits it. Understood this way, Perlman’s shift ‘from the weight of authority to the authority of weights’ is the most arresting formulation yet of a transformation that legal academia has barely begun to confront.
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum