Pretending AI isn't happening is like stopping wind with your hands
At a forum on the Brazilian coast, Luís Roberto Barroso — retired president of the nation's Supreme Court — offered a measured reckoning with a transformation already underway: artificial intelligence is reshaping how justice is administered, and the question is no longer whether to engage with it but whether human institutions can mature quickly enough to govern it wisely. He spoke from experience, having overseen the development of an AI capable of drafting judicial opinions that he chose not to release, because the ethical architecture to contain it did not yet exist. His argument was neither a celebration nor a warning but something rarer — an honest accounting of a civilization trying to write rules for a train already in motion.
- Brazil's Supreme Court has already used AI to slash its caseload from 150,000 to 20,000 pending cases, proving the technology is not theoretical but operational and consequential.
- Barroso built an AI that can draft judicial decisions — then locked it away, because the ethical frameworks needed to deploy it responsibly do not yet exist.
- The urgency is structural: ChatGPT reached 100 million users in two months, while regulators still struggle to understand systems built by developers who know far more than those tasked with overseeing them.
- Algorithmic bias is a real concern, but Barroso reframes the stakes — human judges carry their own prejudices into every courtroom, and the choice is not between a flawed tool and a perfect one.
- The broader crisis is epistemic: the collapse of shared factual ground, the spread of disinformation, and the erosion of trust in what we see and hear are converging with AI's rise in ways that threaten the foundations of public discourse.
- The path forward, Barroso insists, runs through education and regulation simultaneously — neither alone is sufficient, and neither can afford to wait for the other.
On a Saturday in May, Luís Roberto Barroso stood before an audience in Guarujá and made a claim that would have sounded like fiction not long ago: AI will one day produce judicial decisions with greater objectivity than human judges. The caveat was just as significant — not yet, and not without the right safeguards.
Barroso had lived this tension firsthand. During his tenure as president of Brazil's Supreme Court, he oversaw the development of an AI system capable of drafting judicial opinions. He chose not to release it. The ethical framework to govern such a tool simply did not exist. What the court had deployed, however, was already remarkable: an algorithm that identified cases with established precedent and filtered repetitive appeals, compressing the STF's backlog from 150,000 cases to 20,000. Judges using the system were required to justify any decision to override its recommendations — effectively obliged to argue against the machine.
On the question of algorithmic bias, Barroso was direct but reframing. Human judges carry their own prejudices into chambers every day. The concern was legitimate, but the comparison was not between a flawed technology and a neutral human — it was between two imperfect systems. "Pretending artificial intelligence isn't happening is like trying to stop the wind with your hands," he said. The real work was governance.
That governance faced a structural problem: the speed of change had outpaced the capacity to regulate it. The internet took seven years to reach mass adoption. ChatGPT reached 100 million users in two months. Thousands of scientists had twice called for a pause in AI development to allow ethical frameworks to catch up. No one stopped. And beneath the speed lay an asymmetry of knowledge — those writing the rules understood far less about the systems than those building them.
Barroso situated all of this within a broader civilizational shift. The move from analog to digital had democratized information but also demolished the editorial gatekeeping that once created a shared factual baseline. The result was what he called the "tribalizing" of public life — each group inhabiting its own narrative, its own version of reality. Disinformation flourished. Conspiracy theories spread without friction. "When we can no longer believe what we witness," he said, "freedom of expression loses its meaning."
His response to all of it was not despair but insistence on education — teaching people to think critically in a landscape where anything can arrive instantly and unvetted. Regulation and re-education, he argued, would have to advance together. Both would take time. Both carried risks. But the alternative — looking away — was no longer available.
Luís Roberto Barroso, the retired president of Brazil's Supreme Court, stood before an audience at the Esfera Forum in Guarujá on a Saturday in May and made a claim that would have seemed like science fiction a decade ago: artificial intelligence will eventually render judicial decisions with greater objectivity than human judges can muster. The caveat, however, was substantial. Not yet. Not without the right safeguards in place.
Barroso had overseen the development of an AI system capable of drafting judicial opinions during his tenure at the STF. He chose not to release it. The reason was straightforward—the ethical framework governing such a tool did not yet exist. "I have no doubt whatsoever that this is the future," he said. "AI producing decisions with more objectivity than judges. But we don't have a mature enough code of ethics to let it happen yet." The tension between technological capability and institutional readiness defined much of what he had to say.
The scale of what AI has already accomplished within Brazil's highest court was striking. Using an algorithm designed to identify cases with established precedent and filter out repetitive appeals, the STF had compressed its backlog from 150,000 cases to 20,000. That reduction alone suggested the magnitude of the transformation underway. Yet Barroso insisted the technology must operate under human supervision. Each judge would bear the burden of explaining their reasoning—particularly if they chose to disregard an AI recommendation. The judge, in effect, would have to argue against the machine.
Barroso acknowledged the legitimate concern about algorithmic bias, but he reframed it in a way that challenged the premise of the worry itself. Judges made of flesh and blood carried their own prejudices into chambers every day. They made discriminations. They were human. "Pretending artificial intelligence isn't happening is like trying to stop the wind with your hands," he said. "It's coming. We have to regulate it as best we can." The real problem was not whether to use the technology but how to govern its use responsibly.
The deeper challenge, Barroso suggested, lay in the sheer velocity of technological change. The internet took roughly seven years to reach mass adoption. ChatGPT reached 100 million users in two months. Regulators were being asked to write rules for a train already in motion. Thousands of scientists had twice petitioned for a pause in AI development to allow ethical frameworks to be established. No one stopped. No one wanted to fall behind. And beneath that lay another asymmetry: the developers of these systems knew far more about them than the legislators tasked with overseeing them. "It's not simple to regulate something when the person you're trying to regulate knows vastly more than you do," Barroso said.
He had opened his remarks by tracing the arc of digital transformation more broadly. The shift from analog to digital had democratized access to information, to knowledge, to public discourse itself. It had also demolished the gatekeeping function that traditional media once performed. In the old system, editors filtered what reached the public sphere. Now anything could arrive instantly, unvetted. The consequence was what Barroso called the "tribalizing" of life—each group inhabiting its own narrative, its own set of facts. "We've lost the ability to work from common ground," he said. "If someone tells me this pen is a tire, we've lost the capacity to communicate. That's roughly where we are in the world."
This fragmentation had opened what he termed "avenues of disinformation." Hate speech flourished. Conspiracy theories spread with ease. The business model of traditional journalism, flawed as it was, had at least created a shared factual baseline. That baseline had crumbled. Yet Barroso resisted despair. The solution, he believed, lay in education—in teaching people how to navigate this new landscape, how to think critically about what they encountered. "This is a civilizational problem," he said. "But I'm not pessimistic. It's a matter of education, of learning to live with new technologies. People will have to be re-educated for this new world."
AI itself was not inherently malevolent. It had accelerated medical breakthroughs. It had enhanced decision-making capacity. It had made information move faster. But it also had military applications. It threatened labor markets. And it threatened something more fundamental: the ability to trust what we see and hear. "When we can no longer believe what we witness," Barroso said, "freedom of expression loses its meaning." The task ahead was not to stop the technology but to civilize its use—to educate people away from deploying it for criminal purposes, for deception, for harm. That education would have to happen in tandem with regulation. Both were necessary. Both would take time. Both carried risks.
Notable Quotes
I have no doubt whatsoever that this is the future: AI producing decisions with more objectivity than judges. But we don't have a mature enough code of ethics to let it happen yet.— Luís Roberto Barroso, retired STF president
When we can no longer believe what we witness, freedom of expression loses its meaning.— Luís Roberto Barroso
The Hearth Conversation Another angle on the story
When you say AI will make decisions more objectively than judges, what do you mean by objectivity? Aren't you just replacing one kind of bias with another?
That's the first thing people ask, and it's fair. I'm not saying AI is perfect. But a judge brings a lifetime of experiences, assumptions, cultural conditioning into the courtroom. An algorithm, if built correctly, applies the same rule to the same facts every time. That's a different kind of consistency. Whether it's truly more objective—that depends entirely on how it's trained and what data it learns from.
So the bias is baked in at the start, in the training data.
Exactly. Which is why we can't just release these tools and hope for the best. We need oversight. We need judges to explain themselves when they reject what the algorithm recommends. That creates accountability.
But you said you shelved your own program because the ethical framework didn't exist. How long until it does?
That's the impossible question. ChatGPT went from nothing to 100 million users in two months. How do you write rules for something moving that fast? Scientists have asked for a pause twice. No one listened. Everyone's racing.
Racing toward what?
Toward a world where we can't trust what we see anymore. Where a video of something that never happened looks real. Where the shared facts that let us argue with each other disappear. That's the real danger—not the technology itself, but what happens when we lose the ability to agree on basic reality.
Is there a way back from that?
I don't think so. Forward is the only direction. But it requires education, regulation happening at the same time, and a kind of collective will to use these tools responsibly. Whether we have that will—I'm not certain. But I have to believe we do.