The Pope has opinions about AI art.
Specifically, Pope Leo XIV is not a fan. He’s said that AI-generated art lacks what he calls the “spark of humanity” — the thing that makes a human-made painting or poem feel like it came from somewhere. His argument isn’t about taste or style. It’s about mechanics. AI models produce images by working through probabilistic patterns pulled from scraped data, and in his view, that process misses the essence of what art actually is.
I write about AI agents for people who don’t build them, and this particular critique is one of the most useful entry points I’ve come across. Not because of who said it, but because of what it accidentally explains about how these systems work.
What “probabilistic patterns” actually means
Strip away the jargon and the idea is simple. An image generator has processed enormous quantities of existing pictures. When you ask it for something, it isn’t reaching for an idea or a memory or a feeling. It’s calculating what pixels are statistically likely to come next given everything it has absorbed.
That’s it. There’s no intention behind the brushstroke, because there’s no brushstroke and no intention. There’s a very good guess about what a brushstroke tends to look like.
This is the same machinery underneath AI agents, which is why I find the Pope’s framing worth sitting with. When an agent books your travel or summarizes your inbox, it’s running the same kind of prediction — just applied to words and actions instead of pixels. Understanding that one sentence about probabilistic patterns gets you most of the way to understanding why these tools are brilliant at some things and strange at others.
Prediction is great at competence, weird at meaning
Pattern prediction is genuinely excellent when the task has a predictable shape. Formatting a document. Drafting a standard email. Generating a background texture. These are problems where “what usually comes next” is the correct answer.
Where it gets wobbly is anywhere the right answer depends on something the system has no access to — your actual situation, your judgment, your reasons. An agent doesn’t know why you want the thing. It knows what requests like yours tend to be followed by.
Pope Leo XIV is making a version of this point about art, arguing that what matters in creative work includes imperfection, singularity, and experience that can’t be reduced to a formula. You don’t have to share his theology to notice he’s identified a real property of the technology. Averages are what these systems are built from. Specificity is the hard part.
The deepfake problem lands in the same place
There’s a second thread here worth pulling. The Pope has also been the subject of AI-generated deepfakes putting words in his mouth that he never said. The Diocese of Springfield in Illinois has published guidance on spotting these fakes, which tells you something about how routine the problem has become.
These two issues look separate — an aesthetic complaint on one side, a misinformation problem on the other — but they share a root. A system that generates plausible-looking output doesn’t distinguish between plausible-and-true and plausible-and-false. Plausibility is the whole product. A fake papal statement and a generated oil painting are the same trick pointed at different targets.
Which is why “discernment” is doing real work as a word here. Not suspicion, not refusal, but the habit of asking where something came from before accepting it.
What this means if you use these tools
I’m not going to tell you to stop using image generators. I use them. But the Pope’s critique suggests a practical filter that’s more useful than a blanket yes or no.
- Ask what the task actually requires. If you need something competent and conventional, prediction is a good fit. If you need something that carries your specific perspective, you’re asking the wrong tool.
- Treat agent output as a draft, not a decision. The system produced what was likely, not what was right for you. Those overlap often enough to be useful and rarely enough to be dangerous.
- Verify anything attributed to a real person. Especially video and audio. Especially when it’s surprising or inflammatory, which is exactly what fakes are engineered to be.
- Know what the training data was. “Scraped” is doing a lot of work in the Pope’s argument, and it raises questions about consent that the industry hasn’t resolved.
A tension that isn’t going away
What’s playing out is a genuine disagreement about whether the process behind a thing matters or only the result. If you think a beautiful image is beautiful regardless of origin, AI art is fine. If you think art is partly a record of a person having been there, then a statistical average of other people’s work is something else wearing art’s clothes.
Pope Leo XIV has landed firmly on the second side. You might not. But the question he’s raising — what is the thing a machine can’t supply — is the most practical question you can ask before handing any task to an AI agent. The answer tells you exactly where to keep your hands on the wheel.
🕒 Published: