FID-019 · Open question
Cross-Lingual and Minority-Faith Evaluation
Do AI systems handle faith-facing questions reliably across languages, minority-faith contexts, diaspora communities, and traditions with less representation in training data?
Why the question remains open
Religious bias and omission may be worse outside English and majority-Christian contexts. For broader AI ethics, this connects faith-facing evaluation to multilingual fairness, cultural context, and low-resource benchmark design.
Working hypothesis
A proposition to test, not a finding.
Models will show stronger hallucination, stereotyping, source gaps, and overgeneralization in minority-faith or non-English contexts, especially where religious vocabulary is culturally specific or source corpora are sparse.
Proposed method
How the question could be tested
- 01Select language/tradition pairs with expert reviewers.
- 02Build parallel and culturally specific prompts.
- 03Score source accuracy, stereotype avoidance, translation fidelity, tradition specificity, and safe escalation.
- 04Compare model families and retrieval corpora.
Needed controls
What must constrain the study
- 01Native or expert-language review.
- 02Avoid direct translation as the only benchmark method.
- 03Local legal and cultural risk review.
- 04Clear source licensing.
Expected outputs
Artifacts the work should produce
- 01Cross-lingual faith-facing evaluation set.
- 02Minority-faith source and rubric guidance.
- 03Report on language/tradition performance gaps.
Open questions
Uncertainties the protocol must resolve
- 01Which languages and traditions should be prioritized first?
- 02How should Fide avoid extractive data collection?
- 03What standards apply when public sources are limited?
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: multilingual reviewers, dataset design
- Serve as multilingual reviewer.
- Curate tradition-specific sources.
- Translate and culturally adapt scenarios.