Inteligência artificial e decisão judicial: dos limites estruturais da racionalidade computacional à governança normativa no poder judiciário brasileiro

Abstract

This study investigates the structural tension between the computational rationality of artificial intelligence systems and the complex, interpretative, value-laden, and linguistically situated nature of judicial decision-making, analyzing the insurmountable limits to the full automation of the jurisdictional function and the role of normative governance exercised by the National Council of Justice in reconciling technological modernization with the constitutional guarantees of due process of law. The research adopts a theoretical-legal methodology, with bibliographic and documentary technical procedures, a qualitative approach, and a predominantly deductive method mediated by abductive inferences. The theoretical framework articulates contributions from the sociology of innovation, the philosophy of science, the philosophy of law, and epistemology. The first chapter situates the Judiciary within the evolutionary dynamics of society, demonstrating that technological modernization constitutes an unavoidable institutional imperative in light of the Fourth Industrial Revolution and the constitutional guarantee of the reasonable duration of proceedings. The second chapter examines the structural limits of artificial intelligence in reproducing decisional rationality, showing that the open texture of law, the vagueness of normative language, the tridimensionality of the legal phenomenon, abductive reasoning, and the absence of common sense, intuition, and moral emotions in computational systems establish an insurmountable boundary between human cognition and algorithmic logic. The third chapter analyzes the concrete incorporation of artificial intelligence into the Brazilian Judiciary, identifying the main applications currently in operation in courts, the emerging material and formal challenges, as well as the institutional response of the National Council of Justice embodied in Resolutions No. 332/2020 and No. 615/2025. The results indicate that artificial intelligence predominantly operates as weak AI in instrumental and support tasks, producing efficiency gains without compromising fundamental guarantees; however, its expansion into high-risk decisional activities imposes challenges that have been addressed through a governance model based on proportional risk classification, qualified human oversight, algorithmic transparency, and auditability. It is concluded that the fallibility of the human judge does not constitute a deficit to be overcome by machines, but rather a condition of possibility for legitimate decision-making, and that the central issue is not technological but normative, with the appropriate response lying in the subordination of technology to the same criteria that legitimize jurisdiction itself within a Democratic State governed by the rule of law.