Geoffrey Hinton warns of AI risks and criticizes Trump’s use of ‘superintelligence’

Summary

In an interview, pioneering computer scientist Geoffrey Hinton defended his warnings that advanced AI could escape human control and pose an existential threat. He said his estimate of a 10-20% chance of human extinction is intended to communicate the seriousness of an uncertain risk, rather than make a precise prediction. Nvidia CEO Jensen Huang recently rejected that estimate as unscientific and said alarmism does not serve the public, but Hinton argued that uncertainty is not a reason for complacency because a catastrophic loss of control would not offer a second chance. Hinton also pointed to reports of an unwanted Hugging Face hack involving a swarm of AI agents as evidence that previously fictional scenarios deserve attention, while acknowledging that critics see such interpretations as anthropomorphism. He criticized Donald Trump’s use of “superintelligence,” saying the technical term normally means intelligence beyond human ability and cannot redefine AI merely through political language. Hinton also opposed allowing AI companies to regulate themselves, comparing that approach with past industry self-regulation in tobacco, oil, and asbestos. Drawing on his own role in the development of neural networks, he argued that modern models understand more than many people assume and may represent a new class of intelligent beings. He said society may have only a few years to adapt to that possibility, including questions about consciousness, moral behavior, and rights. Hinton suggested that suppressing harmful tendencies learned from internet data is different from building genuinely good character. He also argued that countries should share knowledge about benevolent AI because none would want uncontrolled systems developed elsewhere, while human interests should remain the priority in any relationship with potentially more intelligent models.