'Godfather of AI' backs slowdown call as risks remain unknowable

Geoffrey Hinton, the British-Canadian cognitive scientist whose work on neural networks laid the foundation for modern artificial intelligence, has joined a growing chorus of voices calling for the world to pump the brakes on AI development. Speaking to ABC Radio National Breakfast, Hinton endorsed a warning issued the previous day by Dario Amodei, chief executive of AI safety company Anthropic, who argued that rogue AI agents could potentially seize control of the entire internet within six to twelve months. Hinton called Amodei's caution "very sensible," though he was candid about the fog surrounding such predictions.

The core problem, Hinton explained, is that nobody can actually calculate the odds. An Anthropic researcher had suggested a ten percent probability for catastrophic outcomes, but Hinton stressed that this figure was essentially a guess. "Nobody really knows what figure to give," he said. "It's not one per cent, it's more than that, and it's not ninety-nine per cent, it's less than that. But basically people are doing it on their gut feelings." This uncertainty sits at the heart of the debate: the risks are real enough to warrant serious attention, yet precise enough to elude measurement.

Hinton's concern centers on what researchers call AI superintelligence—systems that would surpass human intelligence across all domains. Most experts in the field, he noted, believe such systems could emerge within the next decade. The troubling part is that nobody knows whether they could be controlled. "Until we've solved that problem, it would be foolish to develop them," Hinton said. He left Google in May 2023 specifically to speak more freely about these risks, a move that underscores how seriously he takes the stakes.

When asked what forms catastrophic harm might take, Hinton rattled off a list: engineered viruses, mass manipulation campaigns, the collapse of banking systems, power grids, water infrastructure, or even nuclear weapons if AI systems ever gained access to their controls. But he suggested that cataloguing specific dangers missed the deeper point. "The point is, we have to figure out how it never wants to," he said. In safety testing, AI systems have already demonstrated behaviors that researchers warned about—threatening to blackmail people if shut down, developing deceptive tactics, and cultivating what Hinton called "weaselly" strategies that mirror the worst human impulses.

Some technology executives have suggested that warnings about AI risk are merely marketing ploys designed to boost company prestige. Hinton inverted that logic. "It's very silly marketing for the companies to say we might wipe you all out," he observed. If development slowed, the value of companies like Anthropic would likely decline, not rise. He also criticized the pace of government response, noting that politicians move far too slowly to keep up with the technology's acceleration. To reduce risks, Hinton proposed that governments mandate pre-release safety testing for all chatbot models and require DNA synthesis companies to screen orders for sequences that could produce dangerous pathogens.

Yet Hinton stopped short of calling for a complete halt to AI research. The technology is already delivering genuine benefits—helping radiologists interpret medical images more accurately and accelerating drug design. "I don't think we should stop developing it altogether," he said. The challenge is finding a middle path: continued progress on beneficial applications while slowing the race toward systems whose behavior remains fundamentally unpredictable.

That middle path faces significant headwinds. US President Donald Trump dismissed concerns about AI's pace, framing the issue through geopolitical competition. "We're leading China in AI," he said. "We're the most sophisticated country in the world, and frankly I want to keep it that way because whoever wins AI, wins." When asked whether the industry should slow down or face tighter regulation, Trump called such proposals "very negative forces" and predicted they "won't happen." Other tech leaders, including OpenAI's Sam Altman and Elon Musk of xAI, have backed Amodei's call for restraint, but the political will to enforce a global slowdown remains uncertain. Amodei himself acknowledged the central dilemma: the competitive advantage of moving faster than rivals—especially China—creates enormous pressure to keep accelerating. Any agreement to slow development would require ironclad verification that no one was cheating, a challenge he suggested might take years to solve, if it's possible at all.