AI pioneer Geoffrey Hinton is afraid that powerful AI could eventually wipe out humanity, and he has a striking idea for how to prevent that from happening.
If we embed “maternal instincts” in AI now, we could ensure it does not want to get rid of us, Hinton said on Thursday at the Ontario Centre of Innovation’s DiscoveryX conference in Toronto.
Geoffrey Hinton
“We need to think of AIs as our mothers and us as the babies.”
“We need to think of AIs as our mothers and us as the babies,” Hinton said. “We need to make them care more about us than they do about themselves.”
For years, the Nobel Prize-winning computer scientist and University of Toronto professor emeritus has been sounding the alarm about the risks associated with the technology he helped build; he left his role at Google in 2023 to speak more freely about this.
Hinton is often described as a “godfather” of AI. He, along with many other experts, thinks that superintelligent AI that is much smarter than humans could arrive within the next 20 years. Hinton fears that such AI could pose an “existential threat” that could potentially take control and kill us.
Hinton has called for superintelligence development to be banned. During Thursday’s talk at the Enercare Centre, he argued that more powerful AI is an inevitability given how useful it has already proven at so many tasks. “Given what it’s going to be, we have got to find a way to make sure it doesn’t want to kill us,” Hinton said.
Hinton proposed that instead of forcing superintelligent AI to submit to humans and serve as our executive assistants, tech leaders should abandon that model and make it our mothers by training it to develop maternal instincts. The mother-baby example, he said, offers a model of less intelligent people controlling more intelligent ones.
Embedding maternal instinct in AI could ensure that it wants humans “to be happy and to realize their full but rather limited potential,” one of Hinton’s backing slides read.
“It’s only those AIs that are going to protect us from rogue AIs,” said Hinton, who referred to this proposal as “the best I can do here, so far.”
Hinton’s idea that artificial superintelligence, or ASI, is inevitable or a serious threat is not necessarily a consensus view in the industry. Late last year, Allen Institute researcher Tim Dettmers wrote that the philosophical discussion around ASI is fundamentally flawed given that infrastructure developments may slow as they run up against physical constraints before being able to power computation at that level.
Hinton has previously estimated that the risk of AI leading to human extinction over the next 30 years is between 10 and 20 percent. When he was asked about those figures during a question-and-answer period following his keynote presentation, Hinton said it is difficult to make predictions about this, but he “optimistically” believes it is less than 15 percent.
Feature image courtesy Ontario Centre of Innovation on LinkedIn.