Hinton warns AI could soon outpace human control

Geoffrey Hinton has warned that artificial intelligence could soon become difficult for humans to control, as fellow researcher Paul Christiano forecasts a high risk of an AI takeover within three years. Christiano also said there is a meaningful risk of such an outcome within six to 12 months. These are forecasts by the researchers, not established outcomes.

Hinton, a Nobel laureate known for his work on neural networks, had previously said he thought AI might surpass human intelligence within about a decade. He now believes that prospect is much closer, and said Christiano’s shorter timeline had alarmed him. Hinton has described the central challenge as ensuring that increasingly capable systems remain under human control when their goals may differ from people’s.

The timing of Hinton’s remarks to European leaders in Dublin is reported differently. Some accounts published November 13 described him as set to deliver his warning at a summit on Friday, November 13. Another account said he had spoken at the European Council summit and urged European governments to exercise caution. His policy recommendations, reported in that account, included requiring evidence of safety before...