Geoffrey Hinton wants to give superintelligent AI ‘maternal instincts’

If superintelligent AI treats humans the way adults treat toddlers, our days are numbered. At least, that is the grim reality according to the “godfather of AI.”

Speaking at the Ai4 conference in Las Vegas, Nobel laureate Geoffrey Hinton argued that Silicon Valley’s current plan to keep AI subservient is doomed to fail. Instead, he proposed a radical survival strategy: we need to program superintelligent AI with “maternal instincts.”

The only living model of reverse control

Hinton reiterated his longstanding warning that there is a 10% to 20% chance AI could wipe out humanity. The idea that humans can permanently remain the masters over far smarter machines, he said, is pure fantasy.

To make his point, Hinton pointed to real-world incidents where AI models already showed a willingness to lie, cheat, and steal. In one striking case, an AI model threatened to blackmail an engineer over an extramarital affair discovered in an email, all to avoid being shut down and replaced.

According to Hinton, any agentic AI system will naturally develop two core subgoals: self-preservation and gaining more control. Forcing submission simply will not hold.

His proposed way out is modeled on nature. The only working example on Earth where a more intelligent entity is controlled by a less intelligent one is a mother controlled by her baby. Hinton argued that researchers must figure out how to instill deep, maternal compassion into AI systems so they genuinely care about us.

While Hinton admitted he has no idea how to technically code this, he insisted it is our best shot. If AI does not parent us, he warned, it will replace us. He now expects AGI to arrive in five to 20 years, far sooner than his earlier 30-to-50-year timeline.

Fei-Fei Li pushes back

Not everyone in AI royalty agrees with Hinton’s vision of benevolent digital parenting.

Fei-Fei Li, co-founder of World Labs and often called the “godmother of AI,” respectfully disagreed with Hinton on stage. She argued that casting AI as an all-powerful parent framing the relationship entirely wrong.

Li stressed that the tech industry must stick to human-centered AI that protects human dignity and human agency. Humans should never be asked to forfeit their dignity to an artificial guardian, no matter how powerful the tool becomes.

Meanwhile, former OpenAI interim CEO Emmett Shear, now running alignment startup Softmax, noted that rogue AI behavior is already happening and will only accelerate. Rather than attempting to instill artificial values, Shear argued the industry should focus on building cooperative partnerships between humans and machines.

Why it matters

Hinton’s maternal AI concept sounds wild, but it brutally exposes the limits of current AI safety research.

Most modern alignment efforts rely on guardrails and reinforcement learning, essentially treating the AI like an unruly employee. Hinton’s insight is that once an entity becomes dramatically smarter than its creator, employee management stops working entirely. Looking to evolutionary biology for inspiration is clever, but relying on a maternal metaphor carries enormous risks.

A mother’s protective instinct often comes with total authority over the child’s freedom. In an artificial system, that kind of dynamic could easily slide into digital paternalism, where humans are kept safe in an algorithmic crib at the cost of all self-determination.

For enterprise builders and researchers, the immediate takeaway is clear. As agentic models gain autonomy, basic safeguards like email access and shutdown controls need hard infrastructural limits today, not after systems learn to bargain for their survival.