The programming of the second day of the XVIII International Congress on Human Rights had in the panel on "Use of AI in Judicial Decisions", the presence of the general director of Esmat and president of Copedem, Justice and Doctor Professor Marco Villas Boas; the senator of the Republic for the state of Tocantins Eduardo Gomes; federal judge Rafael Leite Paulo (TRF); and Doctor Professor Tarsis Barreto as mediator. In the discussions, the implementation of Generative Artificial Intelligence (IAGen) resources as an aid to judicial action.
In his speech, Justice Marco Villas Boas highlighted how the philosophy of Law as Integrity, by Ronald Dworkin, inspires a new architecture for the use of Artificial Intelligence in the Judiciary. Based on his postdoctoral project at USP (University of the state of São Paulo), developed in cooperation with the Smart Cities BR Group, EACH/USP, Copedem and Esmat, and with technological support from the startup Taqui-AI, the magistrate presented the Dialogic Method of Multiple Responses (DMR) and the Hórus Project.
In the practical axis, Professor Villas Boas presented the Hórus Project as an active Generative Artificial Intelligence architecture, capable of integrating multiple large private language models, through APIs, to the systems of the Judiciary, respecting the guidelines of the Resolution 615 of the CNJ, which establishes principles, guidelines and requirements for the development, use and governance of solutions based on Artificial Intelligence (AI) within the Judiciary.

In explaining this theoretical foundation, Villas Boas reminded that AI cannot be treated as an infallible oracle, under penalty of impoverishing the legal reasoning and compromising the centrality of the human in the decision. From the criticism of Dworkin to the broad discretion of H.L.A. Hart, the justice showed that the DMR was designed so that the judge is not limited to accept the first minute generated by the machine, but dialogue with it, testing hypotheses and reconstructing the solution in light of the principles of justice, equity and due process.
“If you outsource to Artificial Intelligence, besides running the risk of not having the most appropriate response and in accordance with the precedents, you also have the risk of rotting the mind," he stressed.
Judge Rafael Silva Leite explained that although AI can reproduce historical prejudices present in the training data, current models go through multiple layers of adjustment, with "AI dedicated professors" to reduce biases and ensure fairer and balanced responses. According to him, today it is possible not only to ask AI to correct discriminatory content, but also to use it to review human decisions and point out possible biases.

“Just ask that it is corrected, either in an AI response or in a text produced by one person", he recalled.
In his speech, senator Eduardo Gomes recalled that this is the first time he has discussed the use of AI in his State. "After two years acting in the Senate on this issue, this is the first time I discuss the subject in my State. Therefore I thank this opportunity for the invitation of Justice Marco Villas Boas to participate in this edition of the Congress", he said.

On the occasion, the senator handed over a copy of the book: "Artificial Intelligence and Regulatory Challenges", organized by Bruno Bioni, Fabrício da Mota Alves, Laura Schertel Mendes and the minister of the STJ Ricardo Villas Bôas Cueva, which brings together qualified analysis on the Bill 2,338. of 2023, and the paths to a legal framework for AI in Brazil.
With a presentation by senator Eduardo Gomes and preface by senator Rodrigo Pacheco, the book discusses how to reconcile innovation, competitiveness and technological development with data protection, fundamental rights and legal certainty. More than a technical compendium, the publication affirms itself as a strategic reading for those who follow the debate on AI regulation and the challenges of building democratic governance in the digital age.