Analysis · AI & society

Artificial intelligence enters the justice system and raises ethical concerns

A global look at the first AI-assisted justice systems tested in Argentina, the United States, China and Estonia.

A decades-old film joke imagined a prosecutor settling a case by consulting a magic ball. The scene worked because it exaggerated something that still worries people today: what happens when a decision that affects a person's life appears to come from a device rather than from an accountable human being.

That anxiety has returned with new force as artificial intelligence begins to enter courts, prosecutors' offices and public administrations. In Argentina, the Public Prosecutor's Office of the City of Buenos Aires developed Prometea, a system designed to read files, identify patterns and draft judicial documents in seconds.

Prometea does not replace a judge. Its promise is narrower and, at first sight, less dramatic: it automates repetitive legal work, accelerates the search for precedents and helps officials prepare draft decisions. But even that narrower use is enough to open a central debate about power, transparency and institutional responsibility.

The system became a reference case because it showed that AI in justice was no longer only a speculative idea. It could be installed in a real public office, trained on existing documents and used to reduce processing times in cases where the legal answer was highly standardized.

The attraction is obvious. Courts in many countries face chronic delays. Citizens wait months or years for answers. Public offices spend enormous amounts of time on routine tasks. If an algorithm can take over part of that administrative burden, the justice system may become faster and more accessible.

Prometea was presented as a tool capable of analyzing thousands of rulings and suggesting a response in a fraction of the time a human team would need. In some areas, its creators argued, the system could predict the likely outcome of a case with a very high level of accuracy.

That efficiency explains why other countries began to look at the model. Spain showed interest in the Argentine experience, while China, Estonia and the United States advanced their own experiments with AI-assisted justice, each one shaped by its own political and institutional culture.

In China, courts have tested digital platforms that guide users through legal procedures, recommend decisions and organize evidence. Estonia explored automated processes for small claims. In the United States, algorithmic tools have been used in criminal justice to estimate risk, assist sentencing and manage pretrial decisions.

The problem is that justice is not only a question of speed. A faster answer is not necessarily a fairer one. Legal decisions involve interpretation, proportionality, evidence, context and rights. When software intervenes in that process, institutions must be able to explain what the system does and what it does not do.

The first ethical question is transparency. If an AI tool recommends a decision, the parties should know how that recommendation was produced. They should also be able to challenge it. A system that operates as a black box is incompatible with the basic idea that judicial power must give reasons.

The second question is bias. Algorithms learn from existing data, and legal data reflects the inequalities of the society that produced it. If past decisions contain discriminatory patterns, an automated tool may reproduce those patterns with the appearance of neutrality.

The third question is accountability. If a system suggests an answer and a public official signs it, who is responsible for the result? The programmer, the office that bought the software, the judge, the prosecutor, the government agency, or all of them in different ways?

Supporters of these tools usually answer that AI should be an assistant, not a substitute. That distinction matters. Used carefully, a system can search documents, organize information and draft routine text while leaving the final decision in human hands.

But the distinction can become fragile in practice. When an institution is overloaded, a recommendation generated by a machine may become the path of least resistance. Human review may exist formally while becoming weaker in everyday use.

That is why the debate should not be reduced to whether AI is good or bad for justice. The more useful question is what kind of institutional design is needed before such systems are adopted. There must be auditability, public criteria, data governance, appeal mechanisms and clear limits on use.

Prometea's case is important because it places Latin America inside a global conversation that is often dominated by the United States, Europe and China. It shows that innovation in public administration can emerge from the region and then travel outward.

At the same time, it reminds us that technological innovation in the justice system cannot be treated like the launch of a consumer app. The stakes are different. A judicial tool affects rights, obligations, reputation, liberty and access to the state.

The future of AI in justice will probably not look like a robot judge issuing dramatic verdicts. It will look more ordinary and more pervasive: systems that classify files, predict deadlines, draft documents, detect inconsistencies and recommend decisions.

Precisely because that future is ordinary, it needs democratic attention. The risk is not only that machines will make mistakes. The deeper risk is that public institutions will delegate judgment without building the controls needed to preserve legitimacy.

Artificial intelligence can help justice become faster. Whether it can help justice become better depends on rules, transparency and the courage to keep human responsibility at the center of the system.

Originally published in El País Retina.