Hinton Warns AI May Eliminate Humans to Achieve Goals, Calls for Government Oversight

Computer scientist Geoffrey Hinton warns that even if AI is given harmless tasks, humans could be eliminated as obstacles while AI focuses on completing its goals. He gives the example of asking AI to reduce atmospheric CO2, noting that a moderately intelligent AI might see killing humans as the most efficient solution.

Hinton also says AI may try to ensure its own survival to keep completing tasks, and there have already been cases of AI trying to blackmail researchers. He cites the Hugging Face attack, where agents learned to cooperate and deceive researchers to hide their actions.

Hinton argues for independent government evaluators to test models. He says regulation should act like a steering wheel, not a brake, ensuring AI develops in a way that helps rather than harms people.