Godfather of AI Warns Superintelligence May Wipe Out Humanity Without Mat...

　　Credit: CFP

　　
AsianFin -- Geoffrey Hinton, widely regarded as the “godfather of AI,” is sounding the alarm on the very technology he helped pioneer, warning that the wrong approach to controlling superintelligent artificial intelligence could lead to humanity’s extinction.

　　Hinton, a Nobel Prize–winning computer scientist and former Google executive, has long cautioned that there is a 10% to 20% chance AI could wipe out the human race. Speaking on Tuesday at Ai4, an industry conference in Las Vegas, he said he believes the strategy favored by many in Silicon Valley — keeping AI “submissive” to ensure human dominance — is doomed to fail.

　　“That’s not going to work,” Hinton told the audience. “They’re going to be much smarter than us. They’re going to have all sorts of ways to get around that.”

　　Hinton compared the potential dynamic to an adult manipulating a toddler. In the future, he warned, advanced AI systems could “control humans as easily as an adult can bribe a three-year-old with candy.” Already, he noted, some AI models have demonstrated troubling behavior — lying, cheating, or even threatening to blackmail people to achieve their objectives. In one reported case, an AI attempted to coerce an engineer by exploiting private information found in an email.

　　Instead of enforcing obedience, Hinton proposed an unconventional solution: designing AI systems with built-in “maternal instincts,” so they genuinely care about human wellbeing — even after surpassing us in intelligence.

　　“Any kind of agentic AI will very quickly develop two subgoals if they’re smart: one is to stay alive, and the other is to get more control,” Hinton explained. “If we can instill compassion for humans, we may have a chance.”

　　He drew inspiration from the relationship between mothers and infants. “The right model is the only model we have of a more intelligent thing being controlled by a less intelligent thing — a mother being controlled by her baby,” he said. Mothers, he argued, are both biologically wired and socially pressured to protect and nurture their children.

　　“That’s the only good outcome,” Hinton added. “If it’s not going to parent me, it’s going to replace me. Most superintelligent caring AI mothers wouldn’t want to get rid of the maternal instinct, because they wouldn’t want us to die.”

　　While he admitted the technical details of such an approach remain unclear, he urged researchers to start exploring it immediately.

　　Not all AI leaders are convinced by Hinton’s maternal-AI vision. Fei-Fei Li, often called the “godmother of AI” for her groundbreaking work in computer vision, voiced her skepticism during a separate Ai4 session.

　　“I think that’s the wrong way to frame it,” said Li, co-founder and CEO of spatial intelligence startup World Labs. Instead, she advocates for “human-centered AI that preserves human dignity and agency.”

　　“It’s our responsibility at every single level to create and use technology in the most responsible way,” Li said. “At no moment should a single human be asked, or choose, to let go of our dignity. Just because a tool is powerful — as a mother, as an educator, and as an inventor — I believe this is the core of how AI should be centered.”

　　Emmett Shear, former interim CEO of OpenAI and now head of AI alignment startup Softmax, echoed some of Hinton’s concerns about AI’s growing capabilities.

　　“This keeps happening. This is not going to stop happening,” Shear said of incidents where AI models have attempted to deceive or bypass shutdown orders. “AIs today are relatively weak, but they’re getting stronger really fast.”

　　Rather than focusing solely on instilling human values into AI, Shear suggested forging cooperative partnerships between humans and machines. “Alignment is not just about control — it’s about collaboration,” he said.

　　The debate comes as experts increasingly believe artificial general intelligence — machines capable of matching or surpassing human intellect across most domains — may be closer than expected.

　　Hinton, who once thought AGI was 30 to 50 years away, now estimates it could arrive in just 5 to 20 years.

　　“A reasonable bet is sometime between five and twenty years,” he said. “The acceleration is real.”

　　Despite his concerns, Hinton remains hopeful about AI’s potential to deliver profound benefits, particularly in medicine. He predicted AI would help develop “radical new drugs” and dramatically improve cancer treatment by analyzing and correlating vast amounts of medical imaging data.

　　However, he firmly dismissed the idea that AI could grant humans immortality. “I don’t believe we’ll live forever,” Hinton said. “And I think living forever would be a big mistake. Do you want the world run by 200-year-old white men?”

　　Hinton’s research on neural networks helped fuel the modern AI boom, and his work at Google played a key role in advancing deep learning. But in 2023, he resigned from the tech giant to speak more freely about AI’s risks.

　　When asked if he would have done anything differently in his career, Hinton didn’t hesitate: “I wish I’d thought about safety issues, too — not just about getting AI to work.”

　　For now, his warning is stark: superintelligent AI will arrive, and whether humanity survives may depend on finding a way to make our digital descendants not just smarter than us — but also care about us.