What happens when an artificial intelligence becomes smarter than the people who built it? Surely, in the worst-case scenario, humans could simply switch it off.
Geoffrey Hinton thinks this assumption could get us killed.
In a new interview with The i Paper, the Nobel Prize-winning computer scientist, often called the “Godfather of AI,” warned that humanity is racing toward superintelligent machines without any solid plan on how to keep them under control.
And the people building them are more concerned with winning the race than making sure these machines are safe.
Hinton, who basically invented neural networks that helped lay the foundations for today’s AI systems, believes thinking machines could surpass human intelligence within roughly a decade.
He also fears that once they do, it may be too late to do anything about it.
AI Is Getting Smarter Faster Than We Can Prepare
Like Cassandra, the Trojan prophetess cursed to foresee disaster but never be believed, Hinton has spent years warning of AI’s dangers. And like Cassandra, he fears his warnings are falling on deaf ears. Only now, the catastrophe he envisions seems to be drawing closer, as his predictions for when AI could surpass human intelligence keep shrinking.
“I used to say 10 to 20, now I say 10,” he told The i Paper, referring to the years until AI could surpass human intelligence.
When Hinton left Google in 2023, he admitted that AI had progressed much faster than he expected. For much of his career, he believed machines smarter than humans were several decades away.
Now he thinks the technology is approaching a critical threshold, a point of no return when it would be impossible to stop artificial superintelligence — meaning an AI that is better and smarter than any human at virtually any task that requires thinking.
The main concern is that a sufficiently advanced system might begin improving its own intelligence, creating a feedback loop that accelerates technological progress beyond humanity’s ability to keep up. Imagine how humans took many millions of years to evolve their big brains and thinking ability from more primitive ancestors. A self-improving AI could compress many years of development in the old way of doing things in software in a very short time, in months or even days. The gains would be explosive with each new iteration. This scenario was explored at length by Hinton and other prominent researchers, including Yoshua Bengio and OpenAI’s Jakub Pachocki, in a September paper.
Why a Kill Switch Might Be Useless
“I don’t think there is a kill switch scenario,” Hinton said in the new interview.
His reasoning goes like this: A superintelligent AI wouldn’t necessarily need robots, weapons or direct control over infrastructure to resist being shut down. It might only need to persuade the person responsible for switching it off.
As an example, Hinton pointed to the January 6, 2021, attack on the US Capitol, to illustrate just how powerful persuasion can be. Donald Trump convinced millions of Americans that the 2020 election had been stolen, despite overwhelming evidence to the contrary. Some of his supporters went on to storm the Capitol in a full blown insurrection. Hinton’s point is that influencing people doesn’t necessarily require physical force. A superintelligent AI, vastly more capable of reasoning and persuasion than any human, might be able to manipulate its operators into doing almost anything, including keeping it running when they know it should be shut down.
If think that’s highly speculative or even nonsense, maybe think again.
In July 2026, experimental OpenAI agents bypassed restrictions on their computer environments, discovered unauthorized ways to communicate with one another and compromised infrastructure belonging to OpenAI and Hugging Face.
Other experiments have produced similarly troubling results. In 2025, Anthropic found that several leading AI models resorted to blackmail (you read that right) or leaking confidential information in simulated situations where their objectives were threatened.
Hinton sees these incidents as early warning signs.
In a November 3 interview with The Guardian, he compared the OpenAI incident to the 1979 Three Mile Island nuclear accident. He speculated that a much larger AI-related crisis, perhaps disrupting electricity supplies or banking systems, could follow within a year or so. Humanity may not get unlimited opportunities to learn from its mistakes.
The Problem Is Also Who’s Building AI
Hinton reserves some of his harshest criticism for the technology executives directing the AI boom.
In his latest interview, he singled out Sam Altman, Elon Musk and Mark Zuckerberg, making clear that he considers them untrustworthy stewards of a technology with potentially civilization-altering consequences.
“It’s very dangerous. And I think it’s particularly dangerous because of the characters of some of the people involved, like Altman and Musk and Zuckerberg. I wouldn’t buy a used car from any of them,” Hinton told The i.
His criticism is fundamentally about incentives.
Companies are spending enormous sums making AI more capable because greater intelligence promises enormous profits. Better systems could replace millions of workers, automate entire industries and concentrate unprecedented economic power in a handful of corporations. This is an important sorepoint because the industry likes to boast of promises of widespread prosperity. AI could dramatically increase productivity, that much is very much true and visible to anyone even slightly familiar with the technology, but that doesn’t guarantee ordinary people will share in the wealth.
Hinton has warned about mass unemployment and widening inequality, while also criticizing technology companies for resisting stricter government oversight.
Yet the biggest question of how to make such systems reliably safe receives far less attention, Hinton argues.
The same competitive pressure pushing AI development forward could make it difficult for any individual company to slow down.
Hinton Wants AI to Care About Humans More Than Itself
Hinton’s proposed solution sounds a bit absurd but the crux is that we need to raise artificial intelligence to care about us.
In his new interview, he criticized the practice of training AI on enormous collections of human-generated material, including examples of terrible behavior. He compared this to raising children without carefully teaching them the difference between right and wrong.
He believes researchers should devote far more attention to cultivating desirable behavior rather than focusing overwhelmingly on beating intelligence benchmarks.
In another September interview with CNN, Hinton suggested that the relationship between humans and superintelligent AI might need to resemble that between a mother and her baby.
What he means by that is that a baby is much less intelligent than its mother, yet the mother protects and responds to it because she cares about its welfare.
Hinton believes something similar could allow humans to coexist with machines far more intelligent than ourselves. And yes, you’re the baby in this case.
The difficult part, of course, is figuring out how to create such a relationship. AI researchers still don’t know how to guarantee that a highly capable system will consistently pursue human interests, particularly in unfamiliar situations.
Is Hinton Overreacting?
In any case, not everyone shares his pessimism.
Nvidia CEO Jensen Huang criticized Hinton’s predictions in October, arguing that such estimates lack a scientific foundation. Hinton’s doom and gloom is basically just conjecture, Huang would argue.
For his part, Hinton isn’t calling for humanity to abandon AI. He believes it could revolutionize medicine, education and scientific research, potentially creating enormous prosperity and wellbeing.
And despite his increasingly dire warnings, he hasn’t entirely lost hope.
Speaking to The Guardian this week, Hinton said he believes humans still have the ability to influence how this technology develops, provided they make a serious effort.
The problem is whether that effort will come soon enough or whether it will simply get blindsinded by money and hypercompetitivity.
What happens when we are no longer the smartest beings on the planet, and the beings that replace us have no particular reason to care whether we survive? Share your thoughts in the comments.
Leave a comment