Project maps artificial intelligence systems used by Brazilian judiciary
Tools help judicial employees with their everyday tasks, speeding up the progress of court cases in the country.

In 2022, there were around 78 million lawsuits in progress in Brazil’s 91 appeal courts, placing it top of the ranking of countries with the most lawsuits in the world, according to a recent report by the National Justice Council titled “Justice in Numbers.” This high number of lawsuits can often slow down the progress of cases, but artificial intelligence tools now offer a solution to speed up judicial routines.
In this context, researchers at Fundação Getulio Vargas carried out the largest study to date on how AI systems are used in Brazilian appeal courts and how they help with the day-to-day work of court employees. The goal was to understand what problems these systems seek to solve and how they are helping the work of courts.
The study found that most of the AI tools present in the Brazilian judiciary are capable of helping its employees classify and sort cases. “When it comes to the Federal Supreme Court, for example, it takes an employee 44 minutes on average to analyze whether an appeal is eligible to be heard by the court, but artificial intelligence can do this in 5 seconds,” points out Judge Caroline Tauk of the Regional Federal Court of the Second Region, one of the researchers at FGV Knowledge’s Center for Judicial Innovation, Administration and Research, who led this project.
In order to carry out this study, titled Artificial Intelligence: Technology Applied to Conflict Management in the Judiciary, questionnaires were sent, together a letter written by FGV, to Brazil’s 91 appeal courts, including courts of appeals, regional federal courts, regional labor courts, military courts and regional electoral courts, as well as the Federal Supreme Court, Superior Court of Appeals and National Justice Council. Out of these organizations, 44 appeal courts and the National Justice Council said they had some kind of AI system, ranging from simple automation programs to more complex ones.
Tauk points out that many of these courts use more than one AI tool. “The Superior Court of Appeals has more than one tool and it has one of the most advanced and specialized artificial intelligence teams in the country. Given the importance of this court, these tools are essential, as the Superior Court of Appeals receives a high volume of cases, and some of them are of a relatively simple nature, which is precisely where these systems can act to optimize the work of court employees,” she says.
These tools include ATHOS, a program that can identify and monitor recurrent themes that are judged by the court. “The Superior Court of Appeals hears many appeals on similar issues. The ATHOS system was developed by the court to automate the processing of these appeals and determine whether they are eligible to be heard by the court. ATHOS is able to aggregate cases based on semantic criteria, meaning words that are close to each other, to classify recurrent themes. This helps the court’s justices identify themes to be judged by the court in final decisions,” Tauk explains.
Different AI systems and their functions
In order to effectively understand how artificial intelligence systems are incorporated into the activities of the judiciary, the researchers divided them into four groups. The first group relates to “back office” activities, rather than ones that directly assist judges in judging cases.
“Some tools help with the administration of a court, such as chatbots that help employees answer questions about people management and human resources, like vacations and bonuses. Likewise, there’s a system called AMON that collects photographs of people who frequently enter the courts, so that these individuals don’t have to go through the X-ray machine every day to enter them,” Tauk says.
Most of the technological tools used by the judiciary in Brazil fall into the second category, which encompasses systems aimed at “front office activities,” aiding the provision of justice by carrying out administrative tasks such as transcribing hearings. These technologies directly support office management by performing administrative activities to assist judges in their decision-making processes.
The Superior Court of Appeals’ ATHOS system falls into this category and the Paraná State Court of Appeals has a system called LARRY, which groups requests under similar themes. This system is able to identify cases involving the same type of demand in the state, such as compensation for pain and suffering or requests for medications,” Tauk explains.
The third type of AI system also assists the provision of justice, but in addition to classifying and screening cases, these technologies are closer to judges and help them draft decisions and sentences.
“It’s important to mention that no tool will actually draft a judgment, let alone make a decision. In fact, ethical guidelines and also the systems themselves wouldn’t allow this in the first place, but these technologies can point out certain patterns to help judicial employees in their decision making,” Tauk stresses.
She illustrates how these tools can be useful, commenting on how they are able to identify court cases that have a chance of setting a general precedent. These are cases that go beyond the interests of the two parties involved in the trial, affecting a large number of people, with political, social, economic and possible media repercussions.
“Some systems, such as VICTOR, are able to analyze in five seconds whether a case has a chance of setting a general precedent, allowing court employees to focus their efforts on other activities instead of spending 44 minutes analyzing whether a case may set a general precedent,” the researcher says.
Another example of a tool that works in this way is ELIS, developed by the Pernambuco State Court of Appeals, which works with tax foreclosures and debt collection processes. It sorts cases, identifying whether the collection of a certain debt is still possible, whether the period has already exceeded the five years allowed by law for individuals to be charged, and so on.
“These systems support decisions but they don’t make them independently. They recognize tax statutes of limitation and general precedents, among other types of cases, but always under double human supervision, from a court employee and then from the judge himself,” Tauk says.
Finally, the fourth type of AI system in the Brazilian judiciary refers to a minority focused on the reconciliation stage, using information from previous court cases that are similar, to check which cases are more likely to end in conciliation between the parties involved in the process.
“When a judge receives a proposal, this system will analyze the type of appeal and the litigant, among other information, to indicate the cases that have the best chance of conciliation, in order to better inform the judge so that he can make his decision. The Court of Appeals of the 12th Region has a system of this type,” Tauk says.
Artificial intelligence and the UN Sustainable Development Goals
She believes it is important to bear in mind that 91% of these AI tools were developed by the courts’ own information technology departments, and not by the private sector. In addition, she mentions that the Brazilian judiciary has started to incorporate the UN Sustainable Development Goals into these systems.
“As part of the UN 2030 Agenda, all countries have committed to adopting 10 principles in the areas of human rights, labor, the environment and anti-corruption. The Brazilian judiciary has decided to participate directly in this agenda in order to make its institutions more effective, accountable and inclusive, and consequently to make judicial processes faster and to connect the activities of courts with the principles of the UN Sustainable Development Goals,” Tauk notes.
According to the judge, one of the biggest challenges for the Brazilian judiciary is its slowness. Therefore, one way to incorporate the UN Agenda is to include technology in the everyday routines of court employees. “These actions have been taking place since 2016, so FGV set out to understand what problems these systems seek to solve and what results they have achieved,” she says.
Next step: Measuring the real benefits of AI in the judiciary
Although one of the objectives of this study was to measure how beneficial these AI systems have been for the routine activities of court employees, the researchers found that it can be hard to measure these results in a practical way, in terms of faster processes and facilitation of the work of judicial workers.
Tauk points out that, intuitively, it is possible to say that these technological tools have improved the speed and quality of court professionals’ work, but further research will be needed to be able to measure this phenomenon more concretely.
“In the questionnaire that we sent to judicial organizations, accompanied by a letter, we asked about this, but most of the institutions replied that they did not have metrics to measure these benefits in practice, although they are looking to develop them. We need these results in order to know what investments are required to improve the performance of these courts,” says Tauk, adding that this measurement could be the objective of a future study carried out by FGV.
Leia também