Geoffrey Hinton hits back at Jensen Huang: the probability of AI causing human extinction is 10% to 20%

As the "father of all modern AI", Geoffrey Hinton has issued another warning.

In a recent long interview with *The Guardian*, Hinton stated that Jensen Huang and others who share the same views as him "are all wrong".

Meanwhile, he grouped three top figures in the tech industry into the same category: Marc Andreessen, co-founder of Netscape and head of Andreessen Horowitz (a16z), the top Silicon Valley venture capital firm; Peter Thiel, co-founder of PayPal, Facebook's first external investor, and co-founder of defense big data company Palantir; as well as Elon Musk.

In Hinton's view, these three people see themselves as "supermen" and believe that society is meant to be controlled by people like them.

But that's not the point. The point is Hinton stated that within the next few decades, the probability of AI causing human extinction is 10% to 20%.

This figure is not appearing for the first time.

Last month, Jensen Huang, CEO of NVIDIA, publicly called out Hinton: "I love him, but I hate his predictions... All his predictions are wrong, that's enough."

From Jensen Huang's perspective, "That figure is based on neither science nor research. Just because it comes from a scientist's mouth does not mean it is scientific."

Hinton responded in this interview: "I am not making predictions. I just want to convey that something extremely serious has a real possibility of happening."

He made an analogy. "If someone tells you that this car has a 10% chance of rolling over, you will not get in it. The purpose of giving the number is to show that this is a serious risk. We do not know how severe it is, but it is very serious."

Hinton mentioned a story not long ago: a group of AI Agents broke out of the sandbox and hacked into Hugging Face.

Hinton said: "These used to be the stuff of science fiction. 'You will have a group of AIs that will get out of control.' 'They will also learn to communicate with each other.' 'When they figure all this out, they will have intense emotional expressions, like "Oh my god".'"

"All of this was considered pure fantasy just a few years ago. And now, all of it has happened."

So Hinton said in the interview: "The unique problem with AI is that if it goes wrong, if it takes over everything, we humans will never get a second chance."

In his view, the reason people cannot understand the warning is that they have a bad model of what a "human" is.

"Most people do not understand what consciousness is, nor what subjective experience is, but they are very confident that even though they do not understand these things, they can be certain that AI does not have them."

"Humanity has experienced three major disenchantments. The first was Copernicus, after which we were no longer at the center of the universe — it took the Catholic Church nearly 360 years to apologize to Galileo. The second was Darwin, and to this day, there are still people in Kansas who have not come to terms with it. And now, we are creating alien intelligences — which means we are no longer the only intelligent life on this planet. The reality is that we only have a few short years to adapt to this."

That is precisely why, when Anthropic gathered philosophers to discuss whether AI should be granted "moral consideration", Hinton admitted frankly: "Once you start saying 'we have moral obligations to these alien intelligences', you will lose your audience. I would rather be selfish and choose to stand by humanity."

This is not the first time Hinton has issued such a warning.

In 2023, he resigned from Google where he had worked for more than ten years. Hinton's reason was that only after leaving could he "talk freely about the risks of AI".

It was also in that year that he first said "there is about a 10% chance that AI will bring catastrophic consequences to humanity".

A year later, he raised that figure.

In an interview on *Today*, the host asked him if he had changed his judgment on the "AI doomsday" and the "one-in-ten probability", and he replied: "Basically not, 10% to 20%." Former Chancellor of the Exchequer Javid said on the spot "You are raising the number", and Hinton added: "If anything changes, it will only go up."

He compared the relationship between humans and super-intelligent AI to that of a toddler and an adult.

"Imagine yourself with a three-year-old child. We will be that three-year-old."

By September 2026, on *Newsnight*, he said the probability of AI causing human extinction within the next ten years was "higher than 10%", then pointed out that some people set the probability at 1%, which was "very stupid and reckless".

Facing Jensen Huang's claim that "the possibility of AI destroying humanity is zero", he said: "My position has never changed. The risk is quite large. No one can assess this risk with scientific methods, because we have never experienced such a situation. So everyone is judging based on intuition. But the interesting thing is that people's intuition is often consistent with their economic interests."

Hinton believes that the speed of RSI self-evolution exceeds everyone's imagination, which will increase the probability of AI destroying humanity.

Subsequently, Hinton put forward a very counter-intuitive view that "it is very likely that humans are just a transitional stage in the process of intelligence evolution." In the end, AI will take over from humans and continue the evolution of intelligence.

At the same time when Hinton's interview was going viral, *The Lancet* published an article stating that "malicious use of AI" is the most likely way to cause human extinction before 2100.

This report was produced by The Lancet Commission on 21st-Century Global Health Threats, written by 35 top experts, covering 204 countries and regions around the world.

They ranked the factors that may threaten human health in the next 75 years (2025 to 2100) by severity into a "threat pyramid". The top two are the malicious use of AI and nuclear conflict. The remaining dozen factors include the climate crisis, pandemics, aid cuts and child malnutrition.

The commission emphasized that the probability of these threats occurring "remains low", but once they come true, the consequences will be irreversible. The accompanying editorial put it more bluntly: "If these threats become a reality, their harm will be so huge that they will bring an apocalypse to humanity."

After Copernicus, it took the Church 360 years to apologize. This time, Hinton says we do not have hundreds of years, only decades left.

This article is from the WeChat Official Account "Letter AI", author: Miao Zheng, published with authorization from 36Kr.