Photo: EPA
Nobel Prize-winning physicist and computer scientist Geoffrey Hinton, widely known as the “godfather of AI,” estimates the probability that artificial intelligence could cause the destruction of humanity at around 10%. He acknowledged that such a risk cannot be calculated precisely.
Speaking to BBC Newsnight, Hinton urged world leaders to strengthen regulation of AI development, warning that the time available to prepare for potentially dangerous systems is shrinking.
“I think 10% is a reasonable estimate, although nobody really knows how to calculate it accurately,” Hinton said.
Why Hinton sees the risk as real
Hinton said humanity is creating systems that could eventually surpass humans in intelligence for the first time. This makes it difficult to rely on past experience to predict the consequences.
He argued that a superintelligent AI would not necessarily need a physical body to cause catastrophic damage. It could create chaos through interactions with people, organize large-scale cyberattacks or help develop dangerous biological and computer viruses.
“We need to figure out how to build it so that it doesn't want to do that,” Hinton said.
His estimates of how much time remains before potentially dangerous superintelligence emerges have also become shorter. Hinton previously suggested 30–50 years, then 10–20 years, but now believes the timeframe could be around 10 years or less.
The AI race
Hinton believes AI developers should work simultaneously on increasing the capabilities of their systems and reducing known risks.
However, he argues that major technology companies are primarily competing to build increasingly powerful models because of the short-term financial incentives.
Hinton does not believe AI development can simply be halted. Unlike nuclear weapons, he said, AI has enormous practical benefits in areas such as healthcare, education and drug development, while chatbots are already useful in everyday life.
Hinton calls for faster regulation
Hinton sees regulation and international cooperation as the main ways to manage AI risks, while acknowledging that there is no universal solution.
Different risks require different forms of oversight, he said, but international cooperation should be possible where countries share common interests.
He compared the challenge to the Cold War, when the United States and Soviet Union cooperated to reduce the risk of nuclear catastrophe. In his view, a similar approach is needed to address the possibility of humanity losing control over advanced AI.
Hinton’s comments came shortly after Anthropic researcher Jacob Coxon publicly announced his departure amid concerns that an uncontrolled race to develop self-improving AI systems could pose an existential threat to humanity. Another Anthropic researcher, Evan Hübinger, has estimated the probability of AI causing human extinction within the next decade at more than 10%.
