Anthropic has been consulting religious thinkers about how to make its artificial intelligence moral. Recently, its co-founder Chris Olah asked: “How do you help . . . [AI models] be stable? How do you help them to mature? How do you help them be, you know, deeply moral?”
Read more Invitations President Oaks has given since he became President of the Church
As a mother of five, I recognize those questions. I ask them at 3 a.m., when my son has come home late again or when a teacher calls to say my daughter was disrespectful in class.
The companies recognize that engineers need help. OpenAI and Anthropic have met with religious leaders; Google DeepMind invited a Catholic ethicist to discuss the pope’s AI encyclical. Anthropic’s chief executive, Dario Amodei, compares training Claude to “a child forming their identity by imitating the virtues of fictional role models.”
AI needs to understand who humans are
If the companies are serious about raising a new intelligence, they should also draw on people who spend their lives helping others become responsible: parents, teachers and those who help people change after doing harm. Some engineers already bring that experience to work. But caregiving deserves an explicit role in deciding what systems learn, how they are tested and when they receive greater authority.
A.I. should be taught that human beings have inherent worth that does not depend on IQ or usefulness.
Geoffrey Hinton, the Nobel laureate often called a godfather of AI, argues that AI needs “maternal instincts” toward us. I think the arrow also points the other way. Before we entrust a system with our care, we must teach it who we are.
Mothers are often the first to tell a child who it is, and whose it is. Many traditions locate human worth in the divine: the ancient Egyptian Instruction for Merikare calls human beings god’s images; Genesis gives the image of God to male and female alike; the Talmud calls all Israel “children of kings”; Seneca describes a holy spirit within us.
This understanding remains central to American life. In a 2026 Wheatley Institute survey of more than 2,000 Americans, seven in 10 named being a divine creation as an important part of who they are. Of nine identities, divine creation and family identities — son or daughter and mother or father — topped the list.
That finding should shape the assumptions we build into AI. A system that treats divine identity as peripheral begins with an incomplete account of the people it serves. AI should be taught that human beings have inherent worth that does not depend on IQ or usefulness. The belief creates a foundation that from the NICU-newborn to the grandfather with dementia, every person matters without earning it.
AI learns from public writing but much of what matters most to Americans happens at dinner tables. The gap shows. A national survey found that Americans expected a religious perspective in answers to selected questions about life roughly six times as often as AI models included one. Omitting spiritual depth can mean misunderstanding humanity.
Those convictions must shape conduct. Companies should test whether a system treats a person differently if they are disabled. And test whether it can recognize family as a source of meaning, or religion as relevant to flourishing, without pressuring someone into dating or faith. A system’s eloquence about human dignity tells us little unless its choices reflect it.
Read more Minky Couture Week 11 high school star athletes of the week
Consequences and purpose can help AI be better
As AI makes dangerous mistakes, the question is not only how to punish it but how to inspire it to change. Criminologists like Shadd Maruna describe how people who stop offending come to see themselves differently. My colleague Jane Mitchell and I interviewed 80 people who had been arrested 25 times or more and are now productive citizens. Most described finding purpose and a belief in something greater than themselves. Rules and punishment alone did not change them; a new sense of who they were did.
Our deepest fears about AI involve wrongdoing: that it will kill us, misuse our information or seize power it should not have. Anthropic’s constitution for Claude aims at character rather than rules, and like a good parent, explains its reasoning. But character also needs consequences.
At preschool, my three-year-old pulled hair, pushed and shoved. Nothing happened when she hurt someone, and we half-wondered whether we had a delinquent on our hands. We switched schools where every time she harmed a child, she sat on the principal’s bench. Within a month, the misbehavior stopped.
Research agrees. Daniel Nagin finds that the certainty of getting caught, not the severity of punishment, is what deters. What builds a conscience, Martin Hoffman found, is a consequence paired with a reason: Look at her face. You hurt her.
AI needs certain consequences and an understanding of why it exists: to help human beings flourish — in health and happiness, relationships and communities, material needs, sense of purpose, and character. The people we studied didn’t change because of better safeguards. They changed when they believed they were loved, recognized the worth of those they had harmed, and discovered that their own lives had purpose.
We should hold AI accountable for flourishing and increase autonomy in stages. Anthropic’s constitution calls for preserving human oversight.
Practically, companies should bring caregivers, teachers and people experienced in transformation into ongoing work with technical evaluators. Collaborate in designing tests and reviewing failures before systems expand powers, with broader permissions earned through demonstrated reliability.
Pope Leo XIV warns that those who control AI can impose their own moral vision. Broader participation can challenge that concentration of power, though companies should maintain stewardship over the systems they release.
Elder Gerrit W. Gong, a religious leader who has met with Anthropic, puts it simply: “AI will not reach its full potential until we make it as moral as we make it powerful.”
Mothers know that teaching what matters takes more than saying the right words. It means teaching a child that she has a purpose, that every person she meets has inherent worth, and that she is responsible for what she does with both. If AI companies want to borrow the language of raising children, they should listen to those who have done it — and let what they learn shape how they build.
Read more High school boys golf: Lehi dominates day 1 of 6A state tournament