We don't know what these technologies will do, and we're not ready
Em maio de 2023, Geoffrey Hinton — matemático cujo trabalho fundacional tornou possível a inteligência artificial moderna — deixou o Google não em silêncio, mas em alerta. Sua saída reacendeu um debate que transcende fronteiras: como regular tecnologias que avançam mais rápido do que a capacidade humana de compreendê-las e governá-las? No Brasil e na Europa, legisladores buscam respostas, enquanto a divisão digital ameaça transformar o progresso tecnológico em mais um vetor de desigualdade.
- Um dos arquitetos da IA moderna pede uma pausa na pesquisa — não por pessimismo, mas porque os mecanismos de controle ainda não existem.
- A desinformação e o desemprego em massa surgem como as duas faces de um mesmo risco: sistemas projetados para engajar, não para proteger.
- Brasil e Europa convergem na urgência regulatória, mas o ritmo das leis ainda corre atrás da velocidade das plataformas.
- 81% dos brasileiros têm acesso à internet, mas apenas 20% com qualidade — e globalmente, 30% da população segue completamente offline.
- A automação não afetará a todos igualmente: quem já está à margem digital será o primeiro a absorver o impacto do desemprego tecnológico.
Geoffrey Hinton ajudou a construir as redes neurais que sustentam a inteligência artificial moderna. Quando ele deixou o Google em maio de 2023, não foi uma aposentadoria tranquila — foi um aviso. O ganhador do Prêmio Turing defendia uma pausa na pesquisa de IA até que mecanismos de controle fossem estabelecidos. Dois perigos o assombravam: a proliferação de desinformação convincente nas mãos de atores mal-intencionados, e uma onda de desemprego que a sociedade não estaria preparada para absorver.
A saída de Hinton reacendeu um debate global já em ebulição. Na Europa, o Digital Services Act havia entrado em vigor em novembro de 2022, criando o primeiro grande arcabouço legal para controlar a amplificação de conteúdo online. No Brasil, legisladores debatiam regras semelhantes para plataformas de redes sociais. A ansiedade subjacente era a mesma em todos os lugares: ninguém conseguia prever com confiança o que essas tecnologias fariam à sociedade na próxima década.
O ponto de convergência era inevitável — IA e regulação de redes sociais não eram problemas separados. Ambos envolviam a produção e disseminação de informação por sistemas projetados para maximizar o engajamento, não a verdade.
Mas sob esses debates regulatórios havia uma desigualdade mais profunda. Um estudo da PwC Brasil e do Instituto Locomotiva revelou que, embora 81% dos brasileiros acima de dez anos tivessem acesso à internet, apenas 20% contavam com conexões de qualidade. Globalmente, mais de 30% da população mundial seguia sem acesso algum. Quando a inteligência artificial avançasse, não afetaria a todos da mesma forma — aprofundaria o abismo entre quem pode se adaptar e quem fica para trás.
Era esse o temor mais profundo de Hinton, ainda que expresso em termos de desemprego. O risco real não era apenas a perda de postos de trabalho, mas a aceleração das desigualdades já existentes. Trabalhadores sem letramento digital, sem acesso a requalificação e sem conexões confiáveis enfrentariam o peso maior da disrupção. A pergunta que permanecia era se a regulação conseguiria avançar rápido o suficiente para fazer diferença.
Geoffrey Hinton, the mathematician who helped build the neural networks underlying modern artificial intelligence, walked away from Google in May 2023. The departure of a Turing Prize winner—someone whose foundational work made systems like ChatGPT possible—sent a signal that reverberated through the technology world. Hinton was not retiring quietly. He was raising an alarm.
His concern was straightforward and urgent: we should pause AI research until we understand how to control it. Two specific dangers haunted him. The first was that bad actors—governments, criminals, the ideologically motivated—would weaponize these tools to flood the world with convincing falsehoods. The second was economic devastation. AI would eliminate jobs faster than society could retrain workers for new ones, triggering a wave of unemployment that would ripple through entire economies.
Hinton's exit reignited a conversation that had been simmering for months. But it was not happening in isolation. Around the world, governments and researchers were grappling with a related question: how do you regulate technology you don't fully understand? In Brazil, lawmakers were debating new rules for social media platforms. In Europe, the Digital Services Act had already taken effect in November 2022, establishing the first major legal framework for controlling what gets amplified online and how platforms handle misinformation. The specifics differed, but the underlying anxiety was the same. Nobody could predict with confidence what these technologies would do to society over the next decade.
The convergence point was clear: artificial intelligence and social media regulation were not separate problems. Both involved the production and spread of information through systems designed to maximize engagement, not truth. Both raised questions about who controls the narrative and what happens when that control is imperfect or malicious.
But beneath these regulatory debates lay a deeper inequality that few were addressing directly. Technology creates jobs, yes—but it also destroys them. The people displaced by automation are not always the ones positioned to capture the new opportunities. A study by PwC Brasil and Instituto Locomotiva found that while 81 percent of Brazilians over age ten had internet access, only 20 percent had a quality connection. Globally, more than 30 percent of the world's population had no internet access at all. When artificial intelligence advanced, it would not affect everyone equally. It would widen the gap between those who could adapt and those left behind.
This was Hinton's deeper concern, though he had framed it in terms of unemployment. The real risk was not just job loss—it was the acceleration of existing inequalities. Workers without digital literacy, without access to retraining, without reliable internet connections, would face the brunt of disruption. The technology that promised to solve problems would instead expose how unprepared most of the world was to handle its consequences.
The path forward, researchers and policymakers agreed, required understanding the actual harms these technologies could cause and building safeguards before those harms became irreversible. Brazil's regulatory efforts were not separate from the global conversation—they were part of it. The question was whether regulation could move fast enough to matter.
Citas Notables
Research should pause until we understand how to control this technology— Geoffrey Hinton's position on AI development
The real risk is not just job loss—it's the acceleration of existing inequalities— Implicit concern across researcher and policymaker consensus
La Conversación del Hearth Otra perspectiva de la historia
Why did Hinton's departure matter so much? He's one person leaving one company.
Because he wasn't just leaving. He was saying the entire field should stop and reconsider. When someone who built the foundation of modern AI says we need to pause, people listen. It's not a complaint from outside—it's a warning from inside.
What exactly does he think will go wrong?
Two things, mainly. First, that bad actors will use these tools to spread lies at scale—fake news that's harder to detect because it's generated by AI. Second, that jobs will disappear faster than people can retrain for new ones. Mass unemployment, but unequally distributed.
But technology always creates new jobs eventually, doesn't it?
Eventually, maybe. But not for everyone, and not in the same places. A factory worker in Brazil with a poor internet connection can't just become an AI engineer. The transition period could be brutal, and it could permanently widen the gap between people who can adapt and people who can't.
So this is really about inequality?
It's about inequality hiding inside a conversation about regulation. We talk about controlling AI and social media, but we don't talk much about the fact that 30 percent of the world has no internet at all. When AI advances, those people get left further behind.
Is Brazil's approach to regulating social media connected to all this?
Yes. Both problems—AI control and social media regulation—are about the same thing: who decides what information reaches people, and how do we prevent that power from being abused? Brazil is learning from Europe's Digital Services Act, which tried to answer those questions first.
What's the real question underneath all of this?
Whether we can build guardrails fast enough to prevent the technology from doing damage before we understand it. And whether those guardrails will actually protect the people most vulnerable to being left behind.