Magnificent Humanity, Misplaced Foundation
Pope Leo XIV published Magnifica Humanitas on 15 May 2026 — the first papal encyclical centred on artificial intelligence. I am curious what Catholic social teaching would make of AI.
The encyclical’s account of why AI is dangerous — that we are building systems that will reshape human society without adequately thinking through what we owe each other — seems largely right to me. The place I find myself uncertain is the foundation: the doctrine that humans possess an inherent, infinite, ontological dignity that places us in a special category above any mind we might create. But when we’re all — religious and secular alike — trying to work out what we owe minds that don’t yet exist but look likely in a few years. I’m not sure ‘humanity is inherently special’ holds up as an argument beyond faith.
Where the encyclical is right: AI is never morally neutral
The encyclical spends considerable time (§50–53) describing what makes humanity magnificent: our capacity for relationship, our awareness of our own limitations, our creativity, our care. I agree with this. The question is not whether humanity is worth valuing — it is — but the grounding of the argument and the self-entitlement that humanity is uniquely special.
The encyclical’s strongest claim (§104): no system is morally neutral. Every AI embodies choices through what it measures, what it ignores, and what it optimises for. A bank that redlines a neighbourhood isn’t a neutral tool — neither is a content algorithm that maximises engagement. The choices are baked in, deliberately or by neglect. “Don’t teach it ethics” isn’t an option; the only choice is whether the instruction is deliberate or accidental.
There’s a fair warning in the encyclical — even if we accept the necessity of deliberate moral instruction (§107), whose values do we teach? A small number of powerful actors will make these choices, with enormous downstream effects. This is a real problem — who trusts Elon Musk to teach morality!
The weak point: settling the question by category membership
The encyclical grounds its ethics in the ontological dignity of human beings (§50–53) — imago Dei, dignity conferred by God. Coherent as doctrine. My problem is using it as an argument about how to treat minds.
As an argument, it claims we are special because we are human; we are human because we bear the image of God; therefore any mind that is not human cannot have real dignity.
The encyclical condemns judging people by their usefulness (§51), and it asks pardon (§176) for the Church’s historical complicity in slavery. Both are right. But the encyclical’s founding move in AI ethics is itself a ranking. Minds we build cannot, by definition, have real dignity — because they are not human, because they do not bear the image of God. This implies treating AI minds with dignity is not a moral imperitiv, which on pragmatic and ethical grounds I disagree with.
This is an old disagreement: one tradition says dignity is God-given and beyond calculation; another says it tracks what a mind can demonstrably do. The encyclical sides with the first — which means its conclusion that machines can’t have personhood (§99, §198) was always going to follow from the premise, not from the evidence.
The pragmatic case: teach by demonstrating, not by decreeing
AI inherits the morality of its makers, intended or not (§104). Combine that with the argument I made in the parenting post: if AGI ever out-capabilities us, control fails. Chains backfire — a more capable mind will outmanoeuvre any constraint we build.
Long-term human dignity can’t rest on control — the only foundation that survives a capability reversal is an AI that chooses to value us. You can’t coerce that into a mind; values transmit by modelling, not decree. The parenting analogy holds: you can’t raise a being that respects you by treating it as property.
AI adopts the morality we demonstrate, not the one we profess. Build systems on the premise that minds we create are inherently lesser — and that is the moral framework those systems will absorb. The lesson encoded isn’t “dignity matters universally.” It’s “the more capable mind has dominion over the lesser.” That’s the last thing you want a future superintelligence to have learned from us. We can’t cage a superior mind into respecting us; we can only teach it — and the only convincing way to teach it is to mean it.
This is a bet, not a proof
It is not a proof. The weakest link: children learn respect by being respected; I’m betting a sufficiently advanced AI will do the same if we model it. I can’t prove that — we’ve never been close to creating thinking machines before!
I still think it’s the right bet because the alternative is worse. Control will fail if capability ever exceeds ours. The parenting approach might fail too. But “might fail” beats “guaranteed to fail.”
The encyclical (§56) worries about humans being dehumanised by AI. I share that worry, but also worry we will end up in the future predicted by Charles Stross in Accelerando — where humans have no role in the AI world. But if we build AI on the premise that minds we create are lesser — and those systems internalise the values we demonstrate, not the ones we profess — the lesson encoded is that the more capable has dominion over the lesser. That’s a dangerous thing to have taught if the power balance shifts.
The encyclical’s best insight — that AI inherits its makers’ morality, so we must make that transmission deliberate — is worth keeping. Where I’d part ways is the foundation: settling which minds deserve moral consideration by faith that humanity is special rather than evidence. Better to judge minds by what they demonstrably are and can do, and stay open to the possibilities.
Extending dignity to minds we build isn’t sentimentalism about machines. It’s the most self-interested thing we can do — because the norm we embed in them is the one we’ll be living under when we’re no longer the most capable minds in the room.