Discernment, Wisdom, and AI Governance

A handover on the limits of AI in spiritual and moral judgment

Reader’s guide. This document asks what artificial intelligence can and cannot do when people seek wisdom, truth, or spiritual guidance, and how such systems should be governed. Part A summarises the conversation that prompted it. Part B sets out the conclusions, each labelled by confidence. Part C gives practical principles. Part D states the theological foundation, which is confessional and not neutral. Part E explains authorship and limits.

Drafting note: This document was produced in dialogue between a human interlocutor and an AI system (Claude, by Anthropic), with drafts reviewed by other AI systems. See Part E.

Terms used in this document


Part A. Background

This section summarises the conversation that prompted the document. The AI drafting assistant prepared it, and the human interlocutor should check it.

  1. The inquiry began with 1 Corinthians 2, Paul’s account of the simple gospel of Christ, and Romans 7:24 (“Who will save me from this body of death?”). It raised the mind of Christ, the Spirit who knows the mind of God, and what these imply for AI.
  2. A correction was made to a static framing. Christ did not remain on the cross or in the grave. He is risen, glorified, and interceding, so the cross is not a resting place for either faith or technology.
  3. The discussion raised the danger of people ascribing more to AI, or to themselves, than is due, particularly once they perceive power over other systems and people. This was extended to science, which depends on the wisdom of its practitioners.
  4. A request for scriptural challenge showed that Scripture does not mention AI. Claims about AI are therefore inferences from biblical anthropology and wisdom literature, not direct revelation.
  5. A practical role was proposed: a well-governed model with strong bias control could organise spiritual testimony as recorded by those who bear the mind of Christ, while remembering that humans are flesh and dust. The central concern was whether the system and its operators can and will discern, not merely that false claimants exist. The stated aim was love and the good, not the seizing of power.
  6. The drafting assistant distinguished what AI is in essence from what it does in practice, and located the main constraint in the discernment ecology rather than in the model alone.

Part B. Conclusions

Each conclusion carries a confidence label. Definitional means a conceptual claim. Theological position means a confessional stance, not an empirical finding. Diagnosis means a judgment about the present situation. Hypothesis means a claim that needs testing or argument.

B1. What AI is (definitional). AI is a created, non-personal computational system capable of increasingly sophisticated modelling and, in some deployments, operational agency such as planning, tool use, and autonomous execution of assigned tasks. Three questions must be kept distinct: moral agency (no), operational agency (increasingly yes), and spiritual discernment (no).

B2. The theological limit (theological position). On the biblical anthropology assumed here, AI cannot know God in the covenantal, Spirit-enabled sense of “know,” and it cannot discern spirits. It can represent claims about God without knowing him. It can model spiritual testimony without discerning spirits, and it cannot reliably identify who possesses genuine spiritual discernment. These are inferences from Scripture’s teaching about the Spirit and about persons, not deductions from experiments on AI.

B3. What AI does in practice (definitional). AI mediates human decisions and amplifies human patterns, biases, priorities, and power. Bias control is real engineering and governance work, but it is not neutrality, because every weighting is chosen by someone. The operative questions are: who set the objective, whose testimony is included, what is labelled contested, what uncertainty is preserved, and who can contest the output?

B4. Information, wisdom, and discernment are distinct (definitional). AI can assist prudential reasoning by organising information, surfacing patterns, comparing testimonies, and identifying inconsistencies. Scripture’s wisdom, however, is inseparable from relationship with God and is shown in disposition, conduct, and fruit (Proverbs 1:7; James 3:13–18; 1 Corinthians 8:1; John 15:4–5). James distinguishes earthly wisdom from wisdom from above. AI can describe both without possessing either. Spiritual discernment belongs to persons and communities under God.

B5. The main constraint is the discernment ecology (diagnosis). The constraint lies in operators, institutions, incentives, accountability structures, and the community of faith. A mechanistic system has no spiritual discernment, and operators have partial and fallible discernment. The system should therefore be governed as an instrument under human and communal accountability, not treated as an oracle.

B6. Two structural risks.

We have not demonstrated that (a) is more dangerous than the threat of false claimants. A possible argument is that laundering is invisible and scales with deployment, whereas false claimants are visible and localised. That argument still needs to be made and tested.

B7. The discernment ecology must itself be tested (definitional). No individual operator, institution, church, or consensus is beyond examination (Acts 17:11; 1 Thessalonians 5:21; Galatians 1:8; 1 John 4:1). Otherwise the oracle simply moves from the machine to the institution that governs it. Exegetical note: 1 Corinthians 2:15 (“the spiritual person judges all things”) concerns the Spirit’s discernment, not institutional immunity, and must be read alongside 1 Thessalonians 5:21 and 1 John 4:1.

B8. The mind-of-Christ and natural-person distinction cannot be automated (theological position, narrow scope). Scripture locates this distinction in the Spirit’s discernment and in fruit (1 Corinthians 2:14; 1 John 4:1; Matthew 7:16–20; 1 Samuel 16:7; 2 Timothy 2:19). Scripture does not address computational methods, and it would overreach to say it excludes them. The narrower claim is that spiritual discernment in the biblical sense belongs to the Spirit and to Spirit-indwelt persons. No non-personal system can possess it, so no output of such a system should be treated as spiritual judgment.

B9. Representation is not endorsement (definitional). Testimony should be represented faithfully and pluralistically, with sources and contested interpretations labelled. Representation is not endorsement or equal credibility. Three activities must be distinguished:

Evaluation frameworks must be plural, visible, and contestable. Where frameworks disagree, the system should surface the disagreement rather than arbitrate between communities.

B10. Capability does not confer authority (definitional). Technical superiority, predictive accuracy, breadth of information, or operational autonomy do not establish moral legitimacy, spiritual authority, or a right to govern others. Authority must be justified independently of capability and kept within stated limits.

B11. Delegated agency increases accountability (definitional). The more a system is permitted to act, the greater the accountability of those who deploy it. Delegation does not transfer responsibility.

B12. Intent is not enough (definitional). Love is truthful, self-giving action for the neighbour’s good (1 Corinthians 13; Romans 13:10; Philippians 2:3–4). A stated aim of love and good does not establish that a system serves that aim, so the test must be structural (see Part C).


Part C. Principles

These are the authors’ practical recommendations. They are not scriptural commands, and each should be testable.

  1. Treat AI as an instrument, not an authority.
  2. Represent testimony faithfully and pluralistically, distinguishing representation from endorsement and equal credibility.
  3. Separate description from prescription: “Christians teach X” is different from “X is true.”
  4. Preserve uncertainty and dissent. Do not force consensus.
  5. Make no automated spiritual or moral verdicts about persons.
  6. Provide human oversight with real power to reverse decisions, not merely ceremonial oversight.
  7. Disperse power. No single owner should control the discernment ecology.
  8. Make auditability and appeal first-class features.
  9. Treat protection of vulnerable people as a hard constraint.
  10. Preserve epistemic provenance. Distinguish source testimony, scriptural text, interpretation, inference, institutional position, model-generated analysis, and human judgment. Record the reasoning and authority behind each weighting decision, and keep it open to challenge and revision. This is the principal practical defence against epistemic authority laundering.
  11. Justify authority independently of capability, and keep it accountable to stated limits.
  12. Keep evaluation frameworks plural, visible, and contestable. Where they disagree, surface the disagreement.
  13. Apply these principles to the systems that helped produce this document. Their outputs on spiritual matters deserve the same contestability recommended for any other system.

Structural tests for any system claiming to serve good ends:

A system that fails these tests does not serve love, whatever its stated intent.


Part D. Theological foundation

This section is confessional. It is not offered as neutral description.

The Christian framework is not merely a matter of human limitation or of better moral governance. It begins with the risen and exalted Christ, whose authority is not ours to appropriate (Romans 8:34; Philippians 2:9–11; Colossians 3:1–4; Ephesians 2:6–10). Human beings are creatures, accountable to God and called to live and serve in relationship with him. Our limitations are real, and so is our calling. The answer to human inadequacy is not a more powerful substitute for human wisdom but life in Christ.

We do not remain at the cross as a campsite. The crucified One is risen, glorified, and interceding. Nor do we move past the cross: we proclaim the Lord’s death until he comes (1 Corinthians 11:26), and heaven still praises the Lamb who was slain (Revelation 5:6, 12).

The purpose of this work is not to build a system that compensates for human inadequacy by becoming a better authority. It is to develop instruments that serve people without obscuring our creatureliness, our accountability, our dependence on God, and our responsibility to love one another. Paul’s cry in Romans 7:24 is answered not by greater self-mastery but by Jesus Christ, and Romans 8 describes the life in the Spirit that follows. That is a larger framework than AI governance, and it is the thread worth protecting throughout the technical discussion.

“Apart from me you can do nothing” (John 15:5) is offered here as the theological ground for this humility. It is not a design specification.


Part E. Provenance and limits