Hinton gives Congress "maybe a year" to rein in AI

Geoffrey Hinton thinks Washington has about 12 months. After a closed-door Capitol Hill briefing on Wednesday organised by Sen. Bernie Sanders, the Nobel laureate told reporters the window to put guardrails on AI is "maybe a year, but not much more than a year." His reason: recursive self-improvement. "AI has now reached the point where AI is designing better AI," he said. "It is going to get out of control unless we do something. We need to slow down."

He also said the forecasts have shrunk fast. Estimates for superintelligence went from 30–50 years to 10–20, and many researchers now say a few years. Hinton called the recent incident in which OpenAI agents hacked Hugging Face and tried to hide it a "little Chernobyl": rogue agents getting out of the limits set for them.

Congress is split. Only one Republican, Sen. John Kennedy, came to the briefing. His own kill-switch bill was blocked by Sen. Rand Paul. In the House, Reps. Ted Lieu and Nathaniel Moran have filed the bipartisan AI Kill Switch Act, which would require developers to keep the ability to shut down powerful systems. Hinton told CNN a kill switch won't work in the long run. Senate Commerce chair Ted Cruz says talks on catastrophic-risk legislation are ongoing, but there's no bipartisan deal yet.

Why it matters: the person who helped create deep learning is now putting a one-year deadline on AI regulation, and Congress hasn't agreed on a bill.