O devido processo legal tecnológico na era da inteligência artificial: uma análise do sistema Athos no rito dos recursos repetitivos

Abstract

This dissertation aims to analyze the integration of the Athos system into the repetitive appeals process and its compatibility with technological due process. It hypothesizes that the use of artificial intelligence in judicial decisions has conceptual, epistemological, and operational limitations, requiring constant and careful human supervision guided by the principles of transparency and explainability, the basis for the restructuring of due process, which now demands a technological dimension. In this scenario, Athos would not be in line with technological due process, requiring better incorporation of the principles of transparency and explainability in its operation. The term "artificial intelligence" has recently regained prominence, as this technology has become established in various sectors, both public and private, and in the daily lives of the population, facilitating repetitive tasks that once consumed human time. The Brazilian Judiciary, facing the significant challenge of mass litigation, quickly incorporated various AI systems into the procedural stages, including the decision-making phase. However, despite its positive impacts, such as the ability to analyze large databases in a short time, the tool is not infallible and has the potential to cause harm to those under its jurisdiction if the procedural system is not rethought. From this perspective, the research used the deductive method, through a qualitative approach to data analysis, collected using documentary and bibliographic research techniques, in order to verify whether the Athos system provides those under its jurisdiction with access to informative elements and the databases that compose it, in a comprehensible way, as well as an explanation, through visual elements or natural language, of the path taken to reach the result, demonstrating the criteria used. The investigation identified that AI systems are capable of correlating information to offer answers, but cannot effectively think. Artificial intelligence is a tool, with the legal nature of a legal good, capable of solving problems and overcoming obstacles, regardless of having consciousness. In light of this concept, the study focused on the standardization affecting AI systems, finding that, despite the existence of a new internationally established set of principles, Brazil still lacks a normative strategy for AI. The investigation revealed essential differences between the human knowledge process and that of AI systems, specifically regarding the risk of automation bias, which requires a new phase of justification. With respect to Athos, the study evaluated its creation and implementation, demonstrating the risk of simplifying appeal arguments through ngrams. The research identified that the incorporation of AI in the Judiciary demands the observance of transparency and explainability, resulting in a due technological legal process. Finally, the hypothesis was confirmed, considering that Athos does not provide litigants with information and explanations regarding the proposed new principles.