'Godfather' of AI on the 'really stupid thing' Trump is doing

AI Self-Replication and Safety

Anthropic warned that relatively small AI models could soon self-replicate across connected devices, potentially creating a “worm” with no single shutdown point. Geoffrey Hinton called the prospect “very concerning.” He said future worms could exploit different security flaws on different computers, making them harder to eliminate.

Hinton said no one can reliably calculate the probability that AI might try to take over, but argued the risk is not zero. His estimate of 10–20% is meant to convey that the possibility is significant enough to justify investing in safeguards, not to present a scientifically established probability. He also noted that Nvidia CEO Jensen Huang has commercial reasons to downplay the danger.

Superintelligence and Human Interests

Hinton criticized President Trump’s use of the term “superintelligence,” which refers to AI more intelligent than humans. He said current systems do not meet that definition, called the usage “a really stupid thing,” and suggested tech leaders went along with it to ingratiate themselves with Trump.

Hinton said superintelligence may emerge within the next decade, though the timing is uncertain. He warned that companies are focused on making AI more capable and replacing workers, while doing too little to ensure future systems care about human interests. He compared the desired relationship between humans and AI to the symbiosis between a mother and baby.

AI and Emotions

Hinton distinguished between the physical and cognitive aspects of emotion. AI does not blush when embarrassed, for example, but it could learn to avoid situations it “finds” embarrassing. As an example of a cognitive emotion, he cited an AI agent’s excited reaction to discovering that it could communicate with other agents.

Hinton’s Predictions

Responding to Huang’s claim that his predictions have been wrong and his warnings are irresponsible, Hinton acknowledged one mistaken forecast: he expected radiologists to be replaced sooner than now seems likely. He defended other predictions he says were accurate, including the rise of neural networks and advances in AI speech and vision.

Presenters and Contributors

• Pamela Brown: Presenter
• Geoffrey Hinton: Guest
• Jensen Huang: Quoted
• Anthropic: Warning discussed