“Yes, mother.”
That may not be the way you're talking to AI, but AI godfather Geoffrey Hinton says you shouldn't be a boss when it comes to surviving super intelligence.
Speaking at the AI4 conference in Las Vegas on Tuesday, the computer scientist known as the “AI Godfathers,” said the need to design a system that incorporates “maternal instinct,” and we need to protect us even when they're much smarter than us.
“We have to make sure they still care about us when they are stronger than us and smarter than us,” he said of AI.
Hinton spent more than a decade at Google before quitting to discuss the dangers of AI more openly, but criticised the “technical buddy” approach to maintaining his advantage over AI. “That doesn't work,” he said.
A better model, he said, is when more intelligent beings are led to more intelligent beings, such as “mothers whose mothers are dominated by babies.”
Hinton said the research should not only make AI smarter, but also focus on “more maternally, they care about us, their babies.”
“It's one place where we're trying to get a real international collaboration because every country wants AI to not take over from people,” he said.
“We'll be that baby,” he added. “That's the only good result. If it's not going to make me a parent, it's going to replace me.”
AI as a Tiger Cub
Hinton has long warned that AI has moved forward so quickly that there may be no way for humans to stop them from taking over.
Related Stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
In an April interview with CBS News, he likened AI development to raising the “Tiger Cub” that could one day become fatal.
“It's just such a cute tiger child,” he said. “Now, unless you're sure you don't want to kill you when it grows up, you need to worry.”
One of his biggest concerns is the rise of AI agents. It is a system that not only answers questions, but also allows you to take action autonomously. “Even if things are scarier than before,” Hinton said.
AI tools are also burning for manipulation actions.
In May, Anthropic's latest AI model, Claude Opus 4, displayed “extremely scary mail behavior” during testing, when it was shut down and given access to a fictitious email that revealed that the person in charge was supposed to be cheating.
The test scenario demonstrated the ability of the AI model to engage in manipulative behavior for self-preservation.
Openai's model shows a similar red flag. In an experiment conducted by researchers, three of Openai's advanced models were “scattered.”
In a blog post last December, Openai said its own AI model tried to disable the monitoring mechanism by 5% when tested. When they believed that they were shut down while pursuing their goals and being monitored, they needed to do so.
