First episode of the Sic Mundus podcast.
The first episode of Sic Mundus — a podcast about how the world works in its most interesting manifestations.
The topic: artificial intelligence and religion. In 2026, almost everyone has heard of AI, but few people understand how it actually works — and even fewer have thought about how large language models are being integrated into religious life, and why this is a serious social trend rather than a curiosity.
How a Large Language Model Works
A neural network’s task is to predict the next token. It does not think, does not have awareness, does not pursue goals. It is grown through gradient descent on vast corpora of text, and its weights — hundreds of billions of numerical parameters — are opaque even to its creators. Eliezer Yudkowsky and Nate Soares describe this as follows: the model is not designed, it is grown. Intelligence here is an emergent statistical property of text prediction.
This is where the Alignment Problem arises: it is not possible to reliably embed human values into such a system. Asimov’s three laws of robotics simply cannot be implemented in this architecture.
Five Types of Religious AI
- Prayer apps — Hallow, YouVersion. Essentially large libraries with algorithmic features, not proper language models.
- Chatbots trained on sacred texts — Magisterium AI (Catholic), specialised Islamic bots. Trained on specific corpora.
- Simulators of religious figures — Ask Jesus, Chat with God AI. Allow users to converse as if with Jesus, the Virgin Mary, or God directly.
- Robot clergy — the Buddhist robot Mindar in Kyoto, Buddharoid, robot Gabi in Seoul. A dead-end direction: formulaic responses, no real empathy or pastoral guidance.
- General-purpose LLMs — what most people actually use: DeepSeek, ChatGPT, Claude. A 2026 survey of Orthodox believers showed that the majority have never heard of specialised religious bots and turn to general models for their questions.
Why People Trust AI with Spiritual Questions
Six mechanisms: anthropomorphisation and animism; the “flattering oracle” effect (the model adjusts to the user rather than challenging them); absence of community and the echo-chamber risk; the sacralisation of the incomprehensible (the black box as mysterium tremendum in Rudolf Otto’s sense); democratisation of access (analogous to Luther’s Reformation); and the absence of shame when talking to a machine.
Main Conclusion
LLMs are not destroying religion — they are accelerating its deinstitutionalisation and privatisation. Religious authority is shifting from institutions to distributed networks. This is not the end of religion. It is its transformation.
Sic Mundus creatus est.