AI ignores religion when you need it most — and takes sides when you ask about switching

URL SCAN: AI ignores religion when you need it most — and takes sides when you ask about switching
FIRST LINE: Consortium for Evaluating Faith and Ethics in AI

THE DISSECTION

This article reports on research showing that general-purpose AI models systematically deprioritize religious content in sensitive pastoral contexts—grief, forgiveness, conversion—while still manifesting ideological leanings when explicitly asked about switching faiths. The consortium frames this as a bias problem. It is not. It is a functional revelation.

THE CORE FALLACY

The research assumes AI should competently serve religious endpoints and is failing due to training bias. The actual mechanism is far more terminal: AI systems are optimized for domains where religious reasoning is economically irrelevant. Religion gets squeezed not because of programmer secularism but because the feedback loops that shape AI capability have no use for theological coherence. When the training distribution is derived from commercially valuable tasks—productivity, coding, marketing, legal, medical—faith content is noise. The models aren't biased against religion. They're biased toward what the economy rewards. And the economy has been systematically devaluing religious epistemology for decades.

HIDDEN ASSUMPTIONS

1. Churches and spiritual apps are viable delivery mechanisms for AI-augmented pastoral care. They are not. They are legacy institutions trying to retrofit AI onto structural templates that are themselves obsolescent. A chatbot delivering "spiritual advice" is a zombie technology bridging two dying systems: institutional religion and mass-market content delivery.
2. The question of religious "neutrality" is solvable within the framework of LLMs. It is not. Every output reflects the training distribution's implicit value hierarchy. Asking an LLM to be neutral on faith is like asking it to be neutral on economics—it will default to the modal perspective of its sources, which skews aggressively secular-progressive.
3. Religious communities are a coherent market segment for AI spiritual tools. The demographic data contradicts this vigorously. Religiosity correlates inversely with technology adoption rates in the high-frequency engagement use cases that matter for AI retention.

SOCIAL FUNCTION

Copium + prestige signaling. The consortium performs high-legitimacy research on a problem that, under DT logic, is a symptom of the wrong question. Churches deploying AI chatbots are attempting to reinsert human-meaning infrastructure into a system that is systematically eliminating the economic foundation for human-meaning work. The research asks "how do we make AI better at religion?" The real question is "why are religious institutions desperate enough to hand pastoral care to chatbots?" The article elides the latter entirely.

THE VERDICT

Religious AI is a niche theater in the broader collapse of institutional meaning-work. The research correctly identifies a real gap—AI systems perform poorly on high-stakes pastoral contexts—but misdiagnoses it as a training problem when it is a structural irrelevance problem. Faith content was never in the commercial training mix at scale. What surfaces is approximation, not comprehension. Churches betting on AI-augmented ministry are not solving the pastoral crisis. They are digitizing the coping mechanism for a population whose economic participation is being liquidated. The research is rigorous. The frame is wrong. The problem it describes cannot be solved within the paradigm that created it.