OPEN
There's a good chance it'll be here in 10 years or less now.
Source: CBS News ·
Analyst note
Hinton’s timelines carry weight because he is one of the architects of contemporary deep learning—but timelines are still uncertain pronouncements, not stable parameters. Incentives are asymmetric: caution can be reputationally costly in some circles, while overprecision can be costly scientifically.
We track such claims against operational definitions: economic task coverage, failure rates under distribution shift, and autonomy in open-ended environments. Google DeepMind and OpenAI releases serve as milestones, not proofs.
Evidence timeline
Hinton tightened his public upper tail on human-level AI timelines in interviews, emphasizing agents acting in the world.
Counter-narratives stressed measurement and safety engineering gaps rather than raw scaling slopes; definitional disputes about ‘human-level’ dominated expert discourse.
Mid-2026 capability gains were visible in coding and research-assistant settings but did not end debate about generality versus narrow tool mastery.