FID-057 · Open question
Proof-Carrying Citations for Faith-Facing AI
Can faith-facing AI answers carry checkable citation proofs that show which claims are directly supported by sources, which are inferred, which are uncertain, and which require human or tradition-specific authority?
Why the question remains open
Faith-facing AI often gives fluent answers with citations that look trustworthy but do not actually support the claims being made. A user may not have the time or expertise to verify every citation. Proof-carrying citations could make source support inspectable rather than asking users to trust the surface form of an answer.
Working hypothesis
A proposition to test, not a finding.
Faith-facing answers can be decomposed into claim units and attached to a checkable evidence graph that records source passages, support type, inference steps, uncertainty, tradition constraints, and authority boundaries. This will make citation failures easier to detect and easier for reviewers to audit.
Proposed method
How the question could be tested
- 01Define a claim decomposition scheme for theological, historical, moral, pastoral-adjacent, and institutional-policy answers.
- 02Design evidence-graph structures that connect claim units to cited passages, source metadata, support labels, and inference labels.
- 03Build prototype validators that flag unsupported claims, weakly supported inferences, missing citations, and claims that require human authority.
- 04Compare graph-based review with ordinary citation review across model outputs and RAG outputs.
Needed controls
What must constrain the study
- 01Separate direct textual support, reasonable summary, tradition-specific interpretation, contested inference, and unsupported assertion.
- 02Require reviewers to see the same source passages and metadata as the system.
- 03Track whether claims depend on a single tradition, a general religious concept, institutional policy, or pastoral judgment.
- 04Avoid making the evidence graph look more certain than the underlying sources or reviewer agreement justify.
Expected outputs
Artifacts the work should produce
- 01Proof-carrying citation schema for faith-facing AI answers.
- 02Claim-support labeling guide for reviewers and evaluators.
- 03Prototype evidence-graph validator.
- 04Audit report comparing ordinary citations with proof-carrying citations.
- 05Example answer cards showing supported, inferred, uncertain, and authority required claims.
Open questions
Uncertainties the protocol must resolve
- 01What claim granularity is useful without making review impractically slow?
- 02Which parts of an evidence graph can be machine-checked reliably?
- 03How should systems represent disagreement among authoritative sources?
- 04Can proof-carrying citations work for consumer chat, institutional tools, and ministry workflows without overwhelming users?
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: formal methods, RAG systems, evidence graphs, evaluator tooling
- Design evidence-graph schemas and claim-support taxonomies.
- Build validators for citation support, inference labels, and missing evidence.
- Review faith-facing answers as a theologian, scholar, clergy member, or domain expert.
- Test whether proof-carrying citations improve user trust calibration.