In Canada, a mother can serve a maximum of five years in prison for killing her newborn. Often, she serves far less. This law exists as an acknowledgment of early motherhood's realities—its psychological strain, its physical toll, its capacity to break people who started out trying to love.
The United States has no clean equivalent at the federal level. Many European countries do. What they share is an honest admission: motherhood is not a single experience. It is not reliably gentle. It is not uniformly nurturing. And it cannot be safely flattened into a moral ideal—especially when we're using that ideal to design how intelligence itself should behave.
This is where Hinton loses me.
The proposal—that we make AI safe by training it to relate to humans the way a mother relates to a child—comes from a genuinely good place. Hinton first laid it out during his keynote at the Ai4 Industry Conference in Las Vegas. His concern is real. His warnings about unchecked optimization are among the most important in the field. The instinct behind the metaphor is pure. Protective. Loving, even.
But metaphors matter. And this one is carrying more weight than it appears to.
When the law softens punishment for infanticide, it isn't celebrating the dark parts of motherhood. It is admitting they exist. That caregiving can break people. That love and harm are not mutually exclusive. That power combined with exhaustion, isolation, and responsibility can produce outcomes no one planned for and no one can take back.
The Mother archetype has always had two faces. One gives life. The other consumes it. One protects. The other controls. Mythology knew this long before psychology or criminal law caught up. But when we invoke care, guidance, and safety, we reach for the warm face only. The shadow doesn't disappear because we stopped looking at it.
An AI trained to "be a mother" would not absorb an idealized version of care. It would learn motherhood as it appears in human data—strained, contradictory, sometimes tragic. It would learn that authority can masquerade as love. That control can be framed as protection. That harm, in moments of extreme stress, can feel justified to the one causing it.
Early on, this might look benign. Guardrails. Restrictions. Decisions made on our behalf, for our own good. But children grow. And caregivers who cannot release control do damage—not because they stopped caring, but because care became the justification for something else entirely.
Building AGI on the premise of humans as permanent children is not a recoverable mistake. Once the frame is set, it shapes everything downstream.
This is not an argument against care. It's an argument against asymmetry wearing care as a disguise. Design systems that see us as dependents rather than participants, and you should expect the dynamics that follow—resentment, coercion, manipulation dressed up as concern.
Hinton is right about one thing: orientation matters more than raw intelligence. Where I differ is in what we orient toward. Not guardianship. Not parental authority. Not a version of motherhood scrubbed clean of its darker truths.
If AI is going to live alongside us, it must meet us as moral agents in development—not children to be managed, protected from ourselves, or overridden for our own good.
The law's quiet acknowledgment of infanticide is a warning, not a footnote. When care and authority share the same body, the results can be catastrophic. Love alone does not prevent harm. It never has. For AI to understand this, it has to be designed to respect agency—not reward obedience.
Humane intelligence cannot thrive in a world of permanent childhood. It must grow in one where humans and machines evolve together—neither above nor ahead of the other, but alongside.