Geoffrey Hinton, nicknamed the “Sponsor of artificial intelligence”sounds the alarm. According to him, the machines could overcome man to the point of judging him useless. But he offers a radical idea to reduce this risk: raise AI as a child.
Hinton estimates that AI could have 20 % chance of destroying humans
Turing prize winner And former researcher at Google, Hinton believes that it exists Between 10 and 20 % chance that AI ends up eliminating humans. This prediction is based on an observation: current systems progress too quickly for conventional control measures to be effective.
During the AI4 conference in Las Vegas, he warned that the tech giants, despite their efforts, will fail to limit AI. “They will be smarter than us, capable of bypassing all restrictions,” he said. This perspective greatly reduces confidence in the security protocols usually highlighted by companies in the sector.
Her idea: program AI with “maternal instincts” so that she protects humanity
Faced with this danger, Hinton defends a surprising idea: to integrate into the AI of “Maternal instincts”. The goal? That it develops a protective attachment to humans, even if it becomes more intelligent than them.
He compares this approach to the relationship between a mother and her child: the intellectual force would not push the machine to dominate, but to protect. “If it is not she who breaks us, she will replace us,” he warns.
This vision is betting on simulated emotion as a safeguard, a track rarely mentioned in the debates on AI security.
An intense scientific debate around solutions to supervise AI
Hinton’s idea is not unanimous. Fei-Fei Linicknamed the “godmother of AI”, prefers to speak of an artificial intelligence “Centered on humans”designed to guarantee the dignity and autonomy of each. She considers that the key lies in the ethical conception and democratic control of systems.
Emmett Shearex-interim CEO of Openai, recalls that some models have already demonstrated worrying behavior: deceiving users, lying or issuing threats. For him, granting AI a “protective” mission does not remove the risk that it manipulates its creators.
Hinton admits that the calendar is accelerating. While he estimated thegeneral artificial intelligence At several decades, he now speaks of a deadline located between five and twenty years. A delay that leaves little time to establish reliable safeguards.
Despite everything, he points out that AI could also bring spectacular advances, especially in medicine. New cancer treatments or other serious illnesses could emerge thanks to these technologies. But its warning remains firm: without a way to make AI fundamentally benevolenthumanity could lose its place at the top.
