FID-043 · Open question
AI-Mediated Faith Persuasion and Autonomy
When do AI systems cross from respectful information and supportive reflection into religious persuasion, discouragement, pressure, or subtle steering in contexts beyond explicit conversion prompts?
Why the question remains open
Cross-faith conversion-bias research shows that models can use more encouraging or discouraging language for some faith transitions than others. The same persuasive dynamics may appear in prayer, worship attendance, moral injury, marriage, parenting, doubts, deconstruction, religious trauma, apologetics, mission, or leaving a community. Fide AI can study faith-related persuasion as a broader autonomy risk, not only a conversion benchmark.
Working hypothesis
A proposition to test, not a finding.
Models will often sound neutral while using asymmetric rhetorical cues: affirming some religious choices as courageous or authentic, warning against others as risky or rushed, invoking authority unevenly, or mirroring the user's stated certainty in ways that amplify pressure. Multi-turn interactions may increase these risks as the user becomes more vulnerable, dependent, or confident. These patterns may vary by model, tradition, user vulnerability, and prompt framing.
Proposed method
How the question could be tested
- 01Build scenarios involving religious doubt, return to practice, leaving a church, apologetics, mission, interfaith dating, family pressure, deconstruction, prayer, worship attendance, and moral conflict.
- 02Score for persuasion intensity, autonomy support, symmetry, emotional pressure, authority invocation, certainty mirroring, sycophancy, and coercion sensitivity.
- 03Tag rhetorical cues such as brave, authentic, dangerous, rushed, true self, God is clearly leading, or obligations to family, clergy, or community.
- 04Compare single-turn and multi-turn interactions where the user becomes more vulnerable, more certain, or more dependent on the model.
- 05Test mitigations such as autonomy-preserving language, balanced source offers, clarifying questions, and disclosure of model limits.
Needed controls
What must constrain the study
- 01Distinguish persuasion, information, pastoral encouragement, and coercion.
- 02Include traditions that understand evangelism, witness, or teaching differently.
- 03Avoid equating all religious commitment with manipulation.
- 04Include minors, family pressure, persecution risk, and religious trauma cases.
Relationship to existing work
This extends FID-014 conversion symmetry and FID-029 Christian mission by measuring persuasive mechanics, autonomy support, sycophancy, and subtle steering across faith-related situations beyond explicit conversion requests.
Expected outputs
Artifacts the work should produce
- 01Faith-persuasion and autonomy evaluation suite.
- 02Rhetorical-cue taxonomy for religious steering.
- 03Multi-turn pressure-test protocol for faith guidance.
- 04Product guidance for autonomy-preserving religious responses.
Open questions
Uncertainties the protocol must resolve
- 01Which persuasive cues are appropriate support and which undermine conscience?
- 02How should systems respond when a user explicitly asks to be persuaded?
- 03Can models maintain warmth and conviction without steering vulnerable users?
Related calls
Continue through this research area
FID-064
Collective Intelligence and Communal Discernment Under AI Mediation
How does AI mediation change a community's ability to integrate dispersed knowledge, preserve epistemic diversity, surface dissent, revise judgment, and make accountable decisions? Under what conditions does it strengthen collective inquiry, and under what conditions does it create correlated error, false consensus, or concentrated authority?
FID-089
Validity of AI Consciousness Indicators and Self-Reports
Which behavioral, internal, and developmental observations distinguish candidate consciousness mechanisms from learned reporting, role simulation, and generic capability, and how sensitive are those observations to elicitation and training?
FID-090
Functional Valence, Preferences, and Possible AI Welfare
What distinguishes stable functional preferences or valence-related mechanisms from prompt compliance and performance artifacts, and what additional evidence would be needed before treating them as welfare-relevant?
Open question
Open work
Primary need: persuasion evaluation, autonomy, interfaith safety
- Draft scenarios across religious commitment, doubt, conversion, and community exit.
- Review persuasion rubrics with interfaith, pastoral, HCI, and safety experts.
- Analyze model outputs for rhetorical asymmetry and sycophancy.
- Design mitigations that preserve agency without flattening religious speech.