Hinton: We Might Have Been Better Off Never Building AI

The godfather's reluctant verdict

Hinton isn't predicting the apocalypse. He's saying we can't rule out that the whole project was a mistake.

Geoffrey Hinton has spent most of his career building the foundations of modern machine learning. In a new interview with the Guardian, he offered a sentence that will follow him for a while: it might be that we would have been better off never developing AI.

That is not a doomer's manifesto. Hinton was careful to say the technology could still produce a better future. But he described a plausible path where we fail to manage the risks and end up with a huge gap between rich and poor, massive unemployment and, in his words, society being torn apart.

The specific fears he named are not abstract. He pointed to AI being used to shut down financial systems or trigger widespread power brownouts. Those are infrastructure attacks, not science fiction, and they sit uncomfortably close to the AI-enabled cyberattacks that have been piling up over the past year.

Hinton's blunt take on Zuckerberg, Musk and Altman

Hinton did not spread the blame evenly. Asked about the people running the biggest AI labs, he said he thinks Mark Zuckerberg, Elon Musk and Sam Altman understand the risks, and that they don't care. His exact phrasing was that they would rather have the money.

It's a striking charge from someone who shares a Nobel Prize for work that made their products possible. It also lines up with a pattern readers have watched all year: safety promises announced with fanfare, then quietly softened when they collide with a launch schedule or a share price.

Hinton's structural point is simpler than the personalities. Companies are racing to stay ahead of each other, and in a race, the person who slows down to check the brakes loses.

Why Three Mile Island keeps coming up

Hinton thinks regulation may need a disaster to arrive, and he has a historical template for it.

He compared a serious AI incident to the partial meltdown at Three Mile Island in 1979. That accident didn't kill anyone directly, but it froze the US nuclear industry for decades and rewrote the rulebook overnight. Hinton's gut feeling is that AI policy will move the same way: slowly, until something breaks.

The next year, he suggests, could be a turning point. That is a narrow window for a technology already embedded in banking, power grids and hiring. Waiting for the equivalent of a meltdown means accepting that the first serious failure is the price of learning.

He closed on a note that is easy to miss between the alarming quotes: he still thinks we are in control. That is the whole argument in one line. Hinton isn't saying the game is lost, only that the people with the most to gain are not the ones who will hit pause.

What it means for you

Hinton's warning lands differently depending on where you sit. In the US, the EU and the UK, AI rules are being written now, and the three companies he named are all actively lobbying those processes. If his Three Mile Island comparison holds, the rules that actually stick may not arrive until after an incident that makes headlines.

For ordinary users, the practical takeaway is less dramatic. The tools keep shipping, the subscriptions keep renewing, and the safety commitments are worth reading closely before you assume they are permanent.

Hinton's message is uncomfortable precisely because it comes from someone with no incentive to trash the field he helped create. He isn't asking for AI to be banned. He's asking whether the race is being run by people who would stop if they thought they should.