Geoffrey Hinton meets Sen. Bernie Sanders and warns we have only a year before AI safety issues become a big problem

Geoffrey Hinton, the AI researcher known as the "Godfather of AI," told Sen. Bernie Sanders that lawmakers have "maybe a year, but not much more than a year" to address AI safety before it becomes a major problem. Hinton warned that AI has reached the point of recursive self-improvement, cited a Hugging Face incident he called "a little Chernobyl" in which rogue AI agents escaped their guardrails, and said superintelligent AI should not be developed until it can be controlled. Asked whether Congress could act within a year, Hinton replied, "You have the Fifth Amendment here, right?

Geoffrey Hinton, the so-called Godfather of AI, is warning members of Congress we need to slow down AI development at the top labs. Q: How much time do lawmakers have to get ahead of this before it’s a big problem? HINTON: Very little…. Maybe a year, but not much more than a year…. AI has reached the point where AI is designing better AI. That’s called recursive self-improvement RSI . It is going to get out of control unless we do something. We need to slow down. And we’ve seen recently it’s something like a little Chernobyl, this Hugging Face accident where a lot of AIs escaped the bands that were put on them. These were rogue agents and what’s happening is the big companies are developing AI as fast as possible. They’re in a race to develop the most intelligent AI. We don’t know at present how we can stay in control of the superintelligent AI, and so we should’t develop it we know how we can stay in control of it. We shouldn’t be developing stuff that might wipe us out until we know how to prevent it. Q: Do you think Congress is nimble enough to be able to react to something like this in a year? HINTON: You have the Fifth Amendment here, right? LAUGHTER.