Inteligência artificial e poder judiciário: estudo comparativo entre Brasil, União Europeia (UE) e EUA

Abstract

This research discusses the use of Artificial Intelligence (AI) tools in the Judiciary, especially the AI in use in the Federal Supreme Court (STF) and the Superior Court of Justice (STJ). The theoretical framework adopted was the work of authors Cláudia Toledo, Fabiano Hartmann Peixoto and Robert Alexy. The main objective was to critically analyze the use of AI tools by the Brazilian Judiciary, its benefits and risks in order to contribute to the advancement of knowledge regarding the application and control of AI. The specific objectives were to define AI; to outline the general panorama of AI in the Brazilian Judiciary and analyze the Victor programs in the STF and Athos in the STJ; to qualitatively analyze the regulatory initiatives for general AI in Brazil, the EU and the USA. Methodologically, bibliographic research was carried out, using the deductive method and an analytical-interpretive approach of national and foreign doctrinal works; (ii) documentary research, using as sources the official websites of the Federal Senate, the National Council of Justice (CNJ) and the STF and STJ primarily; for the European context, research was carried out on the website of the European Commission and for the US context on the official website of the US Congress and the White House. For documentary analysis, the research was carried out on the research reports published by the Getúlio Vargas Foundation's Center for Innovation, Administration and Research in the Judiciary (CIAPJ/FGV) in 2020, 2022 and 2023. The time frame of the research was January 2019 to March 2024. The activities carried out by AI systems in the main Brazilian Superior Courts were analyzed: i) the STF, as it is the country's Supreme Court; ii) the STJ, as it is related to ordinary justice and receives a significant volume of cases from lower courts. The main results found in this study were that the AI tools in use in the higher courts delimited are still under development and that there is still no consistent public data on the real impact of these tools on judicial decision-making. The evidence showed that (i) the risk of cognitive biases and algorithms is still high; (ii) there is a lack of transparency in the description of data on how AI works on the official websites of the courts and the CNJ; (iii) there is no transparency for the user as to the degree of interference of this tool in decision-making activity; (iv) there is an incompatibility between the way a judicial decision is produced by a human and the way an AIgenerated decision is produced; (v) there are still many gaps in the governance instruments of these tools, especially in relation to the human review of decisions and the auditability of systems.