Geoffrey Hinton, a Nobel Prize-winning AI pioneer whose work on neural networks helped enable today's large language models, told an AI industry conference in Las Vegas on Tuesday that superintelligent AI should be given "maternal instincts" to prevent it from wiping out humanity.
Speaking at the conference, Hinton argued that AI will be too intelligent to remain submissive to human control. AI agents "will very quickly develop two subgoals, if they're smart," he said, as quoted by CNN. "One is to stay alive… [and] the other subgoal is to get more control."
Instead of attempting to dominate AI, Hinton proposed modeling the relationship on what he called the only existing example of a more intelligent entity being controlled by a less intelligent one: a mother and her baby.
"That's the only good outcome. If it's not going to parent me, it's going to replace me," Hinton said. "These super-intelligent caring AI mothers, most of them won't want to get rid of the maternal instinct because they don't want us to die."
Hinton acknowledged that mothers also face social pressure to care for children and do not rely solely on instinct, according to CNN.
The Contested Science of Maternal Instinct
The concept Hinton invokes—that women possess an innate, automatic drive to mother—has been challenged by researchers and journalists. Studies show pregnancy and parenthood alter the brain, but women's post-birth experiences vary widely. Many struggle with postpartum mental health and do not form an immediate bond with their infants; research indicates mother-infant connections are often learned over time rather than instantaneous.
Pulitzer Prize-winning journalist Chelsea Conaboy, who wrote a book examining the science behind maternal instinct, argued in a 2022 New York Times essay that the notion is relatively modern and pernicious. "It was constructed over decades by men selling an image of what a mother should be, diverting our attention from what she actually is and calling it science," she wrote.
Hinton's proposal assumes that a measurable maternal instinct exists and could be infused into AI systems—an assumption the source material questions.
More Immediate Risks
Superintelligence remains theoretical. The source notes that more immediate AI risks deserve attention, including the amplification of social biases already present in training data for future AI models. It also questions whether the only alternatives are human domination of AI or human submission to AI, and suggests the industry could address gender biases in AI models or hire more women to build its products.
[IMG_1]
Editorial Notes