How can humans retain control if AI becomes intellectually superior? Geoffrey Hinton warned at the Starmus festival in Tenerife on October 20 that there is no established answer. News agency EFE reported directly from his appearance.
Hinton proposed giving AI a lasting motive to protect human wellbeing. He used a mother caring for a vulnerable child as an analogy: a more capable system should continue wanting to protect people even if it could modify its own code. EFE’s report does not describe a working technical implementation.
Hinton believes AI could outperform humans across many intellectual tasks within the coming decade. He also warned that opaque internal processes and systems recognizing evaluations could make safety testing harder. These are his forecasts and concerns, not a newly documented action by a particular model.
The official program listed Hinton’s talk as Making Alien Beings, following a panel with Bernhard Schölkopf and Philip Torr. EFE did not report either researcher’s own position on his proposed protective motive. Their participation therefore does not establish agreement with it.