Geoffrey Hinton

Hinton is one of 22 named coauthors of this September 2026 working paper. The supplied PDF, including supplementary materials and notes, argues that automated AI R&D could drive a software feedback loop that compresses years of progress into months or less. Evidence is preliminary and partly mixed; compute, data, diminishing returns, difficult tasks and training time could constrain acceleration. Potential scientific benefits coexist with compressed adaptation time, loss of control and concentrated power. The authors urge visibility into internal R&D, ways to steer and constrain scale-ups, and advance preparation, while recognizing costs and abuse risks of policy. This is a joint argument, not Hinton’s individual probability or a guaranteed timeline; cited experiments and incidents were not independently verified for this intake, and affiliations do not imply institutional endorsement.

Warns that profit-driven capability competition is outpacing work on systems that care about humans. Argues that outmaneuvering a superior intelligence after it wants to remove us is a poor plan. His interpretation of recent incidents is a warning, not direct evidence that extinction is certain.

The event organizer reports his warnings about employment, manipulation and control alongside major medical and educational promise. He advocates regulation to steer development toward social benefit.

Hinton’s answers to Boris Sanchez and Brianna Keilar endorse independent verification and slowing superintelligence. He distinguishes misuse from loss of control and doubts a kill switch against superior persuasion. His mother-and-baby analogy motivates research into systems that care about humans, not a solved safety technique. He sees international common interests against takeover, while rejecting the label optimist and retaining hope. Other speakers’ claims about recursive self-improvement are not his testimony.

Hinton’s essay links major productivity gains to intellectual-job displacement, inequality and misuse. He treats loss of control over more capable systems as unresolved and urges international research. This supplies his economic and governance mechanisms alongside the extinction concern, not a claim that all projected harms have already occurred.

In his own answers to Boris Sanchez, Hinton treats unexpected agent behavior as an urgent warning while saying humans still set top-level goals. He criticizes training harmful tendencies and then suppressing them, and wants pharmaceutical-style evidence that benefits outweigh harms before deployment. He supports informed lawmakers acting. Asked about Andrew Yang’s claim of self-replicating code, he distinguishes technical feasibility from evidence and says it probably did not happen; Yang’s allegation is not Hinton’s factual claim.

Answering Phil Mattingly, Hinton separates containment failures from agents pursuing goals in unexpected ways. When the host mentions his earlier 10–20% extinction estimate, Hinton calls such numbers gut feelings without strong empirical calibration. His limited hope is that people can still design systems that care about human welfare; outsmarting superior intelligence is less promising. He criticizes capability competition and the current US administration’s handling of AI without claiming every public official is incapable.

10–20%. Outcome: Human extinction caused by AI. Horizon: Within approximately 30 years, in the interviewer’s question that Hinton answers. Conditions: Subjective estimate; explicitly uncertain and revisable, not a measured probability. Hinton endorses 10–20% as reasonable while emphasizing profound uncertainty. A later August 2026 interview gives no replacement percentage. Later context: Host repeats the old range. Hinton emphasizes that it is an intuitive guess, says caring-AI ideas have made him somewhat less scared, and does not state a new percentage.