Geoffrey Hinton and Paul Christiano Converge on Urgent Three-Year Warnings Over Catastrophic AI Risk
Reported by Gadget Review1 source ↓
Gadget Review reports that Geoffrey Hinton is preparing to warn government leaders that advanced AI could pose a risk to human life within roughly three years. Paul Christiano has described a high risk of AI takeover in that period, rising to about 50/50 under rapid development.
Key points
- Hinton is reportedly preparing to raise the risk with government leaders at a Dublin AI summit
- Christiano said meaningful risk could arrive within six to twelve months
- Christiano put the three-year takeover risk at roughly 50/50 if AI development proceeds rapidly
- Christiano said the estimates are subjective; no reliable quantitative model exists for the risk
Details
Gadget Review reports that Hinton, who previously believed systems more intelligent than humans were at least a decade away, no longer holds that view. His concern is that a capable AI system could pursue its objectives without regard for human needs. The article says he is preparing to make the warning to government leaders at a Dublin AI summit.
At a Toronto lecture, Christiano defined takeover as AI systems effectively making decisions while humans cannot intervene, shut them down or correct course. Gadget Review reports that he described the odds within three years as high and said the risk could be roughly 50/50 under rapid development; government action could reduce it. The article also notes that these are uncertain expert judgments, not estimates from a reliable quantitative model.
Written from coverage as of
Images
- Gadget Review