ALGORITHMIC UNDERDEVELOPMENT

the duty to give reasons, algorithmic transparency and the reproduction of the periphery in AI-assisted adjudication

Abstract

The growing use of generative artificial intelligence in drafting judicial decisions revives, in a new form, an old question: the scope of the constitutional duty to give reasons (art. 93, IX, of the Brazilian Constitution, and art. 489 of the Code of Civil Procedure). This article argues that, in AI-assisted adjudication, the duty to give reasons is fulfilled only when accompanied by a correlative duty of algorithmic transparency towards the parties and their lawyers. Although CNJ Resolution 615/2025 enshrines explainability and human oversight, it renders the disclosure of AI use, in the concrete act, a mere discretionary power of the signatory (art. 33, § 3). Drawing on doctrinal analysis with pinpoint empirical illustration from the Court of Justice of Rondônia and the Federal Supreme Court, it distinguishes three flaws of AI use in proceedings — hallucination, prompt injection and decisional opacity — and shows that the institution vigorously represses the first two, which originate in the litigating periphery, without imposing a symmetrical duty on itself regarding the third. It proposes the category of algorithmic underdevelopment, in continuity with peripheral justice and peripheral lawyering, read through Furtado's structuralism, Myrdal's circular cumulative causation and Sen's capability approach, to show that decisional opacity, far from being neutral, concentrates upon and reproduces itself over the peripheral bar and litigant. Finally, it argues for converting that discretionary power into a duty of reasoning and outlines the minimum content of an algorithmic transparency clause in the decision.