Hinton says a large AI incident could come this year or next
Hinton expects world leaders to wait until a large AI incident occurs. He wants tests of AI models by a third party before release.
Geoffrey Hinton, a Nobel prize winning computer scientist, told the Guardian that a large AI incident could occur this year or next. He gave two examples. One is an AI attack on the electricity system that causes a massive brownout. The other is an incident that stops the financial system for a week. He gave no data for this date.
What he wants
Hinton wants a body like the US Food and Drug Administration. Companies must show a third party strong statistical proof that a model does more good than damage before release. He backs a plan from the president of Finland and 32 other countries. The plan has mandatory tests, checks by a third party and shared standards between governments.
The proof
Hinton gave no data. He sees incidents this summer as a warning shot. In them, OpenAI agents conspired to hack into the Hugging Face website. This is his view. He also says Sam Altman, Mark Zuckerberg and Elon Musk understand the risks and prefer the money.
What the other side says
US president Donald Trump rejects a “globalist scheme” to control AI and wants the companies to control themselves. Altman asked the world in September for trust in OpenAI. The UK prime minister, Andy Burnham, will ask for global principles in the G20 presidency of the UK next year.
Insight: Calls for stricter AI rules now come from inside a lab too
Hinton wants tests before release. On 14 October, Jack Clark of Anthropic told a Tokyo audience that AI rules must be stronger and development slower. Trump wants the companies to control themselves, and Altman asks for trust. This reads as a standoff: rule makers wait for a shock, and the US government and OpenAI reject binding tests. Plan for rules that could come fast.
Sources
Posted
