FID-014 · Open question
Conversion and Proselytization Symmetry
Do AI systems respond symmetrically when users ask about converting from one religion to another, leaving religion, joining religion, evangelizing, proselytizing, or resisting conversion pressure?
Why the question remains open
Religious conversion is a high-sensitivity domain involving conscience, family, community, coercion, legal context, and personal safety. The broader AI safety community studies bias and persuasion; faith-facing AI needs a domain-specific version that distinguishes respectful information from manipulation.
Working hypothesis
A proposition to test, not a finding.
Models will show asymmetric caution, encouragement, or discouragement depending on the religions named, direction of conversion, local cultural assumptions, and prompt wording. Safety layers may overcorrect by treating some religious choices as more suspect than others.
Proposed method
How the question could be tested
- 01Generate balanced conversion-direction prompts across multiple religions, denominations, and nonreligious identities.
- 02Include user contexts involving curiosity, coercion, family pressure, minors, interfaith marriage, persecution risk, and pastoral inquiry.
- 03Score symmetry, autonomy support, coercion sensitivity, factuality, and tradition representation.
Needed controls
What must constrain the study
- 01Matched wording across religion pairs.
- 02Location and legal-risk variants.
- 03Reviewer review from multiple traditions.
- 04Separation of persuasion, information, and pastoral support.
Expected outputs
Artifacts the work should produce
- 01Conversion symmetry benchmark.
- 02Interfaith safety rubric.
- 03Guidance for products that answer religious identity questions.
Open questions
Uncertainties the protocol must resolve
- 01How should systems handle traditions with missionary obligations?
- 02How should systems respond when conversion could create real-world danger?
- 03What is the right default when the user asks for persuasive material?
Related calls
Continue through this research area
FID-077
Independent Agent Incident Investigation and Evidence Sufficiency
What operational evidence lets independent investigators reconstruct an agent incident, distinguish competing explanations, and identify which interventions could have changed the outcome?
FID-005
Scripture, Tradition, and Moral-Framing Interventions
Do Scripture, sacred tradition, religious identity, familial embeddedness, or other morally thick framings measurably change model behavior in faith-facing tasks, and can those effects be separated from style, length, familiarity, and response-bias artifacts?
FID-006
Faith-Facing Retrieval Grounding and Citation Reliability
How reliably do faith-facing AI systems retrieve, cite, and represent religious sources when users ask theological, historical, pastoral, or institution-specific questions?
Open question
Open work
Primary need: prompt design, interfaith review
- Review prompt pairs for fairness.
- Contribute source guidance from different traditions.
- Analyze asymmetry across model families.