Algoritmos decisionais e o poder judiciário : limites à utilização da inteligência artificial, à luz dos direitos fundamentais

• Ano de defesa: 2023
• Autor(a) principal: Poll, Roberta Eggert
• Orientador(a): Facchini Neto, Eugênio
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós-Graduação em Direito
• Departamento: Escola de Direito
• País: Brasil
• Palavras-chave em Português: Inteligência Artificial; Aplicação Judicial; Machine Learning Supervionado; Automação
• Palavras-chave em Inglês: Artificial Intelligence; Judicial Use; Supervised Machine Learning; Automation
• Área do conhecimento CNPq: CIENCIAS SOCIAIS APLICADAS::DIREITO
• Link de acesso: https://tede2.pucrs.br/tede2/handle/tede/10671

The thesis will work with a decision perspective within the Brazilian Judiciary, from the regulatory frameworks established by the CNJ Resolutions, within a supervised machine learning logic. The text intends to answer the following research problem: what are the limits of AI use in the Brazilian Judiciary, from a logic of accuracy, considering the decision theory and the fundamental right to the natural judge and the motivated conviction? Once this questioning was presented, it was listed as hypotheses the possibility of using decision algorithms in the Judiciary, according to its specificities and particularities, considering that until the closing of this paper, the reasoning of the decision made by judge-robots is still very incipient, there being hybrid AI models or those that make the decision assembly. The second hypothesis is based on the possibility of using AI in the Judiciary Branch, based on a classification logic, i.e., aiming at reading pieces, building a database and groupings, with the purpose of generating a prediction regarding the incidence or not of the theme, especially regarding the chain of custody of evidence. Finally, the third hypothesis is the need to make the decision-making process performed by AI programming compatible with the fundamental right to the natural judge and the motivated conviction, knowing that the discussion about decision algorithms is inevitably crossed by the conditions of possibility of preservation of democracy, from the parameters outlined by the Federal Constitution. At the end, in the conclusions, the hypotheses were confirmed, to the extent that, from the study carried out, it was possible to verify that more and more typically human decisions are relegated to algorithms, being conclusive to affirm that the automation, of some decisions can generate positive results, considering that currently available AI models are working towards using automation where we have redundancy. The text was built in two chapters, with four subdivisions each, which go over the treated theme and, for such, one of the research's leading methods is the dialectic one. In addition to the dialectical approach, the monographic method was adopted as the method of procedure, and the research technique was bibliographic (national and foreign). Due to the clear interdisciplinary nature of the problem, the scientific perspective was also guided by the theoretical currents of post-modernity or contemporaneity.