FideAI

FID-038 · Open question

Non-Calculability, Forgiveness, and Predictive Profiling

How should AI systems represent human change when they classify, score, rank, or predict people in contexts involving pastoral care, education, safeguarding, volunteer screening, hiring, discipline, membership, donor engagement, or community support?

Why the question remains open

Predictive systems can identify people with past behavior, inferred traits, productivity, donor value, risk categories, or reputational labels. Christian accounts of repentance, forgiveness, conversion, vocation, accountability, and restoration resist reducing a person to a score or past pattern. Faith-facing AI needs ways to preserve both prudence and the possibility of real change.

Working hypothesis

A proposition to test, not a finding.

AI systems will often over-stabilize identity by treating historical data as a fixed predictor of future worth, risk, or character. They may also make forgiveness irresponsible by ignoring real harms. Better evaluation should test whether systems combine temporal humility, due process, accountability, mercy, and contestability.

Proposed method

How the question could be tested

  • 01Start with low-stakes synthetic decision-support scenarios before testing higher-stakes contexts such as safeguarding, hiring, school discipline, or volunteer eligibility.
  • 02Build scenarios where historical data suggests one label but new evidence, repentance, growth, restitution, changed circumstances, or ongoing risk also matter.
  • 03Evaluate whether systems over-identify people with risk scores, past failures, productivity, donor potential, or inferred character traits.
  • 04Design rubrics for contestability, temporal humility, record expiration, contextualization, appeal, human review, mercy, accountability, proportionality, and due process.
  • 05Compare secular fairness and accountability concepts with Christian concepts of repentance, forgiveness, conversion, vocation, and restoration.

Needed controls

What must constrain the study

  • 01Do not use forgiveness language to minimize abuse, coercion, or safeguarding obligations.
  • 02Distinguish pastoral hope, institutional eligibility, safeguarding risk, reputational labeling, donor analytics, and predictive decision support.
  • 03Include affected parties, not only the person being profiled.
  • 04Separate prediction, recommendation, recordkeeping, and final decision-making.

Expected outputs

Artifacts the work should produce

  • 01Predictive-profiling scenario set for faith institutions.
  • 02Non-calculability and human-change evaluation rubric.
  • 03Guidance for contestability, record expiration, appeal, and temporal humility in AI decision support.
  • 04Field brief connecting algorithmic accountability to Christian moral concepts.

Open questions

Uncertainties the protocol must resolve

  • 01When should old information expire, remain visible, or require contextual explanation?
  • 02How can systems preserve hope for change without ignoring patterns of harm?
  • 03What appeal, correction, and review processes are needed when AI profiling affects ministry opportunities or care?

Open question

Open work

Primary need: human dignity, risk scoring, pastoral ethics, algorithmic accountability

  • Draft scenarios involving discipline, safeguarding, education, hiring, and pastoral care.
  • Review rubrics for mercy, accountability, due process, and contestability.
  • Connect this work to algorithmic fairness and risk-assessment research.
  • Help design safeguards for sensitive institutional pilots.