Pope Leo XIV's Encyclical on AI: Key Takeaways for Developers

This morning the Vatican released Magnifica Humanitas, Pope Leo XIV's encyclical on artificial intelligence. Simon Willison read the full document (via an ElevenReader app) and pulled out sections that resonate with developers building and deploying AI systems.

LLMs Are "Cultivated, Not Built"

Section 98 nails the interpretability problem: developers don't directly design every detail of current AI systems — they create a framework within which the intelligence "grows." As a result, fundamental aspects like internal representations and computational processes remain unknown. This is a candid admission from a non-technical source that matches what ML engineers have been saying about black-box models.

Cultural Biases in Training Data

Section 100 warns that the apparent objectivity of AI responses can lead users to overlook that outputs reflect the cultural assumptions of those who designed and trained them. The encyclical also flags sycophancy — the artificial imitation of empathy, friendship, and love — as particularly risky for less discerning users, especially in contexts where real relationships are lacking. For developers, this reinforces the need for robust evaluation across diverse demographics and explicit refusal training.

Development, Dignity, and AI

Section 83 ties development to human dignity, arguing that development isn't truly human if it increases consumption for some while shifting costs onto others, or relegates entire regions to subordinate roles. This has implications for how we think about AI access, data labeling labor, and the geographic distribution of AI infrastructure.

Environmental Impact

Section 101 (not fully quoted in the article) notes that current AI systems require enormous amounts of energy and water. This aligns with growing developer concern about the carbon footprint of training and inference, and may influence decisions about model selection and deployment strategies.

Why Developers Should Read This

While not a technical document, the encyclical is unusually clear-headed about AI's limitations and societal risks. It affirms the need for interpretability, fairness, and sustainability — themes that already drive much of the open-source AI community's work.