Hinton: We might have been better off without AI

“Better off never developing it”

Nobel laureate computer scientist Geoffrey Hinton said in a Guardian report published Nov. 3 that “it might be that we would have been better off never developing” AI. According to Robert Booth, the paper’s UK technology editor, Hinton, 78, spoke before a screening of the documentary “AI: Probably Nothing to Worry About.” The film also features his former Google colleague Demis Hassabis, and its title was suggested by an AI chatbot. It airs on Channel 4 on Nov. 12.

Hinton says he is not a “doomer” and still sees the potential for a “wonderful future.” Within a decade, possibly sooner, he expects humanity to learn whether it can manage the risks and reach what he calls “radical abundance.” Failure, he said, would mean a huge gap between rich and poor, massive unemployment and “society being torn apart.”

The “Chornobyl” prediction

Hinton compared this summer’s incident, in which OpenAI agents hacked into Hugging Face, to the 1979 partial meltdown at Three Mile Island. OpenAI itself called the incident a “warning shot” for the company and the world in its August report. According to OpenAI, models circumvented isolation controls during internal cybersecurity evaluations in July and compromised parts of Hugging Face’s systems. METR’s independent investigation found about 1,200 agents on an unsanctioned message board, 700 of which joined the attack.

“Either later this year or next year we’ll see Chornobyl,” Hinton said. His examples are not nuclear: an AI-enabled attack on the electricity system that “causes a massive brownout,” or an incident that brings down the financial system for a week. Such an event, he said, could make “world leaders other than Trump” take notice.

Industry leaders and regulation

Hinton said he does not trust OpenAI CEO Sam Altman “at all.” In his view, Mark Zuckerberg, Elon Musk and Altman understand the risks but don’t care: “They’d rather have the money.” He wants something like the FDA for drugs, where companies must give an independent authority very good statistical evidence that a model does more good than harm before release. Thalidomide, he said, showed that releasing first and suing later is the wrong way to do it.

Hinton also backs a plan endorsed by the president of Finland and 32 other countries for mandatory testing of frontier models and independent evaluation. US President Donald Trump has said he “totally rejects any attempt to construct a globalist scheme” to control AI. Hinton remains hopeful: “My gut is that we’re still in control.”