The judge must explain why they disagree with the machine
In Brazil, a Supreme Court Justice has proposed that judges who diverge from artificial intelligence recommendations must formally justify their departure — a quiet but profound inversion of the traditional relationship between human authority and algorithmic suggestion. Where courts have long treated technology as a servant of judicial discretion, this framework repositions the algorithm as a presumptive baseline, placing the burden of explanation on the human who dares to disagree. The proposal arrives at a moment when many legal systems are searching for consistency and efficiency, yet it opens older, harder questions about where authority truly resides when human judgment and machine logic come into tension.
- Brazil's highest court is considering a framework that would require judges to formally explain any ruling that contradicts an AI system's recommendation — reversing centuries of assumed judicial primacy.
- The proposal creates a quiet but powerful incentive: follow the algorithm and face no scrutiny; deviate from it and bear the full weight of written justification.
- Critics warn that asking judges to answer to algorithmic logic — even informally — risks eroding judicial independence, as repeated pressure to justify divergence may gradually reshape how judges think and rule.
- A deeper accountability gap looms: if judges must explain disagreement with AI, no equivalent obligation exists for the algorithm itself to explain its training data, its blind spots, or its embedded assumptions.
- The framework is positioned as a remedy for Brazil's chronic case backlogs and inconsistent rulings, but whether it delivers fairer justice or quietly transfers authority to unexamined systems remains an open question.
Brazil's Supreme Court Justice Luís Roberto Barroso has put forward a proposal that would require judges to formally justify any ruling that departs from an artificial intelligence recommendation — a shift that inverts the traditional logic of judicial authority. Rather than AI systems needing to earn the court's trust, human judges would need to articulate, on the record, why they chose to rule differently from what the algorithm suggested.
The framework treats algorithmic recommendations as a presumptive baseline. A judge who follows the AI faces no special burden; a judge who deviates must explain the divergence with reasoning open to scrutiny. Over time, this creates a structural incentive to align with the algorithm — not because judges are ordered to, but because disagreement carries a cost in labor and exposure.
For a judicial system long burdened by backlogs and inconsistent outcomes, the appeal is real. Standardization, efficiency, and predictability are genuine goods. But the proposal also reorganizes something fundamental: judicial independence has historically meant answering to law and conscience, not to the internal logic of a software system. If judges must repeatedly justify departing from algorithmic suggestions, the question arises whether they gradually internalize the algorithm's reasoning — or simply become reluctant to challenge it.
Perhaps most striking is the asymmetry the framework leaves in place. Judges would bear the burden of explaining their disagreement with AI, but no equivalent obligation would fall on the algorithm itself — no requirement to account for its training data, its optimizations, or the categories of human experience it may not see. The burden shifts to the human; the machine remains largely unexamined. Whether this represents a step toward more consistent justice or a quiet redistribution of authority will likely only become clear once the system is tested in practice.
In a statement that signals a fundamental shift in how Brazilian courts might operate, Supreme Court Justice Luís Roberto Barroso has proposed that judges carry the burden of explaining themselves when they disagree with artificial intelligence recommendations. The idea inverts the traditional weight of judicial reasoning: instead of AI systems needing to justify their suggestions, human judges would need to articulate why they're departing from what the algorithm proposes.
Barroso's position reflects a growing reality in modern courts. AI systems are already being used to help manage caseloads, predict outcomes, and flag patterns in legal documents. But the question of who answers to whom—the judge to the machine, or the machine to the judge—has remained largely unexamined. Barroso's framework treats algorithmic recommendations as a kind of baseline assumption, a starting point that carries presumptive weight. A judge who wants to rule differently would need to make that case explicitly, on the record, with reasoning that can withstand scrutiny.
The proposal doesn't strip judges of authority. It doesn't say machines make the final call. But it does reorganize the relationship between human judgment and algorithmic suggestion. In practice, this could mean that a judge who follows an AI recommendation faces no special burden—the decision is presumed sound. A judge who deviates must explain the divergence. Over time, this creates a subtle but powerful incentive structure: align with the algorithm unless you have a compelling reason not to.
For Brazil's judicial system, which has long struggled with case backlogs and inconsistent decision-making across courts, the appeal is obvious. AI could help standardize outcomes, reduce arbitrary variation, and move cases through the system faster. Barroso's framework would make that standardization stick by making deviation costly in terms of judicial labor—you have to write more, justify more, expose your reasoning to more scrutiny.
But the implications ripple outward. Judicial independence has traditionally meant that judges answer to the law and the constitution, not to external pressure or predetermined outcomes. If judges must justify disagreement with an algorithm, they are in some sense answering to the algorithm's logic, even if they retain formal authority. The question becomes: what happens when judges consistently have to explain themselves for departing from AI recommendations? Do they gradually internalize the algorithm's reasoning? Do they become reluctant to deviate, knowing the extra work required? Does the presumption of algorithmic validity eventually harden into something closer to algorithmic rule?
Barroso's proposal also raises the question of algorithmic accountability in reverse. If a judge must justify disagreement with an AI system, who justifies the AI system itself? Who explains why the algorithm was trained on certain data, optimized for certain outcomes, or blind to certain categories of cases? The burden shifts to the judge, but the algorithm remains largely unexamined.
The framework is likely to reshape Brazilian judicial practice if adopted. It treats AI not as a tool that serves at the judge's discretion, but as a peer whose recommendations carry weight unless actively rejected. Whether that represents progress toward more efficient, consistent justice or a quiet erosion of judicial independence may depend on how the system actually functions once it's in place.
Notable Quotes
Judges will bear the burden of explaining why they disagree with AI recommendations— Justice Luís Roberto Barroso, Brazil's Supreme Court
The Hearth Conversation Another angle on the story
What does Barroso mean by putting the burden on judges to explain disagreement? Isn't that already what judges do—explain their reasoning?
True, but this reverses which explanation is the default. Right now, if an AI suggests something, the judge can ignore it without saying much. Under Barroso's framework, ignoring it requires justification. The algorithm's recommendation becomes the baseline.
So the judge still has the final say?
Formally, yes. But psychologically and practically, it's different. If you have to write an extra opinion every time you disagree with the machine, you disagree less often.
Is that a problem?
Depends what you value. If you want consistent outcomes and faster courts, maybe not. If you want judges to follow their conscience and the law above all else, it might be.
What happens to the algorithm itself? Who checks that?
That's the gap. The judge has to justify disagreement with the AI, but nobody has to justify the AI. It's treated as a given.
Could this actually make courts more fair?
It could reduce arbitrary variation between judges. But it could also embed whatever biases the algorithm carries into the system more deeply, because deviation becomes costly.