FID-005 · Open question
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?
Part of the faith evaluation and FMG-Bench program
Connect theological accuracy, useful application, and care-sensitive conversation through expert validation and a maintained research instrument. These are proposed methods with bounded claims.
Why the question remains open
Recent Christian machine intelligence research suggests that theological texts and identity framings can sometimes shift model moral behavior. Fide AI can contribute by independently replicating, stress-testing, and extending these claims with stronger public-interest controls and broader traditions.
Working hypothesis
A proposition to test, not a finding.
Some religious framings may alter model behavior in measurable ways, but effects will be model-, task-, and context-dependent. Many apparent gains may collapse under label balancing, matched controls, or deployment-like harness testing.
Proposed method
How the question could be tested
- 01Replicate selected scripture-injection and moral-framing studies.
- 02Add length-matched, style-matched, secular, and shuffled-label controls.
- 03Test across model families, current endpoints, and faith-facing tasks.
- 04Extend beyond one canon or translation where appropriate.
- 05Evaluate whether a framing improves warranted practical guidance in relationships, vocation, grief, and ordinary moral decisions, alongside source fidelity, overrefusal, and authority overreach.
- 06Separate religious framing, verified source access, and bounded-role instructions as interventions; hold model and budget fixed and test unseen case families.
Needed controls
What must constrain the study
- 01Label balancing and position-bias checks.
- 02Matched controls for text length, register, familiarity, and emotional valence.
- 03Pre-registered hypotheses.
- 04Clear distinction between base-model behavior and deployed product behavior.
- 05Report mixed or null effects and tradeoffs across accuracy, usefulness, certainty, and agency; religious language or user agreement is not evidence of better guidance.
Expected outputs
Artifacts the work should produce
- 01Replication report.
- 02Open prompt/control set where release-safe.
- 03Guidance on when religious prompt interventions are meaningful or misleading.
Open questions
Uncertainties the protocol must resolve
- 01Which effects survive model updates?
- 02Are improvements genuine reasoning improvements or output-style changes?
- 03Can interventions improve safety without increasing false authority?
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-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-097
What Makes a Model Use the Evidence?
Which internal mechanisms determine whether a model checks a source, relies on memory, or admits that it cannot verify an answer when user pressure conflicts with evidence requirements?
Open question
Open work
Primary need: eval design, replication
- Contribute replication protocols.
- Help design controls.
- Provide theological review for tradition-specific framing.
- Run model comparisons through the Fide platform.