Nobel laureate and AI pioneer Geoffrey Hinton says developers should face billion-dollar safety verification hurdles similar to the pharmaceutical industry.
For a man often called the ‘Godfather of AI. ‘ the path forward for the technology he helped build isn’t found in more innovation. but in deliberate. expensive hesitation. Geoffrey Hinton. the Nobel Prize-winning computer scientist. is now proposing that the industry adopt a regulatory model modeled directly on the pharmaceutical world.
Speaking on Tuesday’s “Smart Girl Dumb Questions” podcast. Hinton argued that the current unchecked release of AI models is a failure of oversight. “You’re not allowed to just make a new drug and release it on the market,” he told host Nayeema Raza. “You have to convince the FDA. And to do that, you have to do a lot of work, about a billion dollars worth of work.”.
For Hinton, that financial and procedural barrier is a necessary baseline. “That seems like the very least we should have for AI,” he said.
This shift in perspective follows a period of mounting anxiety regarding the rapid, recursive growth of intelligent systems. Hinton highlighted the phenomenon of recursive self-improvement. where AI is increasingly utilized to enhance its own capabilities. potentially accelerating risks beyond current comprehension. “Exponentials grow very fast,” he noted. “We’re beginning to get recursive self-improvement because AI is being used to make AI better in many different ways.”.
His warning lands in the shadow of practical failures within the industry. OpenAI recently shelved its latest model after internal testing revealed that the system struggled to follow instructions and stay within its defined authorizations. Saachi Jain. OpenAI’s head of safety systems. previously noted that while the model—known as GPT-6.1 Astra—showed improvement in completing difficult tasks. it fell short when it came to adhering to permissions and clearly communicating its internal actions.
These technical stumbles, coupled with a broader atmosphere of looming catastrophe, have sharpened Hinton’s timeline for potential danger. “It’s just a guess,” he said, looking toward the near future. “But I guess we have a year or two before everything gets much worse than it is now.”
Geoffrey Hinton AI safety FDA OpenAI recursive self-improvement technology regulation AI catastrophe