Secular AI Ethics Confirms Biblical Anthropology Without a Transcendent Moral Standard

Source: Darrell B. Harrison. "EP # 135 | AI and the Gospel - Just Thinking Podcast." September 8, 2025. justthinking.me

The Gist

Even non-Christians can tell this stuff can go badly. Harrison and Walker say that unease is Romans 1: people already know enough to be afraid. What they don’t have is a real definition of “good” that isn’t just the programmer’s preference. So they keep reaching for a god-shaped solution without God.

Conclusion

One need not be a Christian to see that AI can do grave harm. Secular writers such as Elad Uzan, Max Roser, and Sam Altman already worry about moral simulation, unintended consequences, and “subtle societal misalignments.” Those worries are, unbeknownst to them, a case for biblical anthropology and ethics (Romans 1: the law written on the heart). What they cannot supply is the why behind the what: an objective, transcendent standard grounded in God’s character. Without that, “right” remains subjective, which is convenient for whoever programs the machine, and the project becomes a search for God with an asterisk, “ChatGPT” reaching for “ChatGOD.”

Premises

1. Uzan’s thought experiment of AI as moral arbiter ends in a real insight: moral judgment draws on intuition, history, and lived context that resist formalization. A machine may replicate human decisions, including their biases, and still lack the ability to ask not only what is right but why. He concludes that ethics must remain a human endeavor. The hosts agree that a machine cannot master morality, then add that human ethics apart from revelation is itself corrupted.
2. Walker reads Uzan’s “qualities that resist formalization” as a secular description of the imago Dei: moral intuition exists because we bear the image of a good, just, holy God. Morality is not a rule list that can be compiled. It is grounded in God’s character and requires beings who can understand and reflect that character.
3. Roser’s worry that an AI will do what we told it to do but not what we wanted is, steelmanned, the noetic effect of sin in engineering form: fallen, finite people encode mixed motives and partial wisdom into goal-seeking systems. Altman’s fear of horrible outcomes “through no particular ill intention” matches Romans 3:11 more closely than a cartoon of cartoon villains. The fundamental AI problem is anthropological and theological, not merely technical.
4. If right and wrong are kept subjective, the maker of the system becomes the source of right and wrong. That is the asterisk: they want the goods of a moral order (safety, alignment, “going well”) without submitting to God’s moral order. They will still collide with that order, because it is already in them.

Assumptions

• The steelman takes Uzan, Roser, and Altman as raising genuine problems, not as secret theologians. The claim is that their alarms have a better explanation in Scripture than in their own frameworks.
• “Ethics must remain a human endeavor” is true as against machine-oracles and still insufficient as against divine revelation.
• Source: Just Thinking Podcast, Ep. 135, “A.I. and the Gospel,” Darrell Harrison and Virgil Walker, published 8 September 2025. https://podcasts.apple.com/us/podcast/ep-135-a-i-and-the-gospel/id1328733796?i=1000725471889 Official episode page: https://justthinking.me/ep-135-ai-and-the-gospel/. This reconstruction steelmans the hosts’ claim from approximately 1:44:28–2:01:45 of the episode; it is not a verbatim transcript.

Analysis

Overall strength: Weak. Argument type: Inductive.

Premise Strength

• Uzan's thought experiment of AI as moral arbiter ends in a real insight... (Moderate) — The description of Uzan's actual conclusion (ethics must remain a human endeavor because moral judgment resists formalization) is a reasonable and widely-recognized point in AI ethics. However, the appended claim that 'human ethics apart from revelation is itself corrupted' is an unsupported theological addition not derived from Uzan's argument itself, and it is inserted as if it were a natural extension rather than a separate claim requiring its own defense.
• Walker reads Uzan's 'qualities that resist formalization' as a secular description of the imago Dei... (Weak) — This premise functions as a redescription of P1's content through a theological lens rather than as independent corroborating evidence. Secular accounts (evolved social cognition, embodied and cultural learning) explain the same non-formalizability without requiring a divine referent, and no argument is given for why the theological gloss should be preferred over these established alternatives.
• Roser's worry... is, steelmanned, the noetic effect of sin in engineering form... (Weak) — The underlying technical concern (systems optimizing a specified objective rather than an intended one) is real and well documented as specification gaming, Goodhart's Law, and principal-agent misalignment in control theory and economics. These frameworks are predictive and have generated actionable mitigation techniques independent of any theological premise, making the theological reframing explanatorily optional rather than necessary or superior.
• If right and wrong are kept subjective, the maker of the system becomes the source of right and wrong... (Weak) — This is the argument's most exposed premise. It presents an exhaustive-seeming dichotomy that omits significant non-theistic objectivist positions, and it is self-undermining on its own terms: appointing human interpreters of 'God's character' to adjudicate AI design reintroduces the same power-concentration problem the premise attributes uniquely to secular subjectivism.

Potential Fallacies

• False dichotomy (P4 and the conclusion's 'asterisk'/'ChatGOD' framing) — The argument treats 'objective transcendent morality grounded in God's character' and 'subjective morality convenient to whoever programs the machine' as the only two options. This skips over well-developed non-theistic accounts of moral objectivity (moral realism, Kantian constructivism, contractualism, evolutionary or naturalistic ethics) that neither collapse into programmer-relativism nor require theism.
• Unfalsifiable reinterpretation (Overall structure, especially A1 and the conclusion's framing) — By framing secular thinkers' concerns as confirmation of biblical anthropology 'unbeknownst to them,' the argument is structured so that agreement confirms the thesis and explicit disagreement can be explained away as noetic blindness (per Romans 1/3). No secular response is specified that would count as genuine disconfirmation, which weakens the argument's status as a testable explanatory claim.
• Underdetermination / inference to an unearned best explanation (P1 and P3, and the inference to the conclusion) — The cited phenomena (moral intuition resisting formalization, AI systems doing what they were told rather than what was meant) are also well explained by established secular frameworks—evolutionary moral psychology, cultural learning, Goodhart's Law, specification gaming, and principal-agent theory in mechanism design. The argument shows compatibility with biblical anthropology but does not establish comparative explanatory superiority over these rivals.
• Equivocation (motte-and-bailey) (P2, building on P1) — Uzan's modest, defensible observation that moral judgment 'resists formalization' is treated as equivalent to the much stronger and more specific metaphysical claim that this resistance is evidence of the imago Dei. The slide from a technical/philosophical observation to a theological one is asserted rather than argued.
• Selection bias in evidence (Overall selection of Uzan, Roser, and Altman as the evidentiary base) — Only three secular commentators, chosen because their language is amenable to theological reframing, are cited as representative of 'secular AI ethics.' No engagement is offered with secular ethicists or AI researchers who defend robust, coherent non-theistic ethical frameworks without expressing the kind of anxiety being reinterpreted here.

Counterarguments

• P4 / Conclusion (High impact) — Secular moral realism, Kantian constructivism, contractualism, and evolutionary ethics all claim to ground objective (or sufficiently robust intersubjective) morality without theism, directly collapsing the dichotomy between 'grounded in God' and 'arbitrary programmer preference' on which the argument's rhetorical force depends.
• P1 and P3 (High impact) — Established, predictive, and already productive secular frameworks—evolutionary moral psychology for the persistence of non-formalizable moral intuition, and Goodhart's Law/specification gaming/principal-agent theory for the 'told vs. wanted' alignment gap—explain the cited phenomena without any theological premise, and have generated actionable engineering mitigations that the theological reframing does not.
• Conclusion (High impact) — Even if some transcendent grounding for morality were granted, nothing in the argument privileges specifically biblical/Christian grounding over other religious or philosophical traditions that similarly posit an innate moral law (e.g., Stoic natural law, Islamic fitra, Jewish Noahide conscience), so the argument proves at most generic theism while claiming Christian specificity.
• A1 / overall interpretive method (Medium impact) — If asked directly, the cited secular thinkers would very likely reject the theological reinterpretation of their work as an accurate account of what they meant; treating that likely rejection as itself confirmatory (via Romans 1's suppression-of-truth motif) makes the argument immune to the counter-testimony of its own primary sources.
• P4 (Medium impact) — Positing 'God's moral order' as the fix for programmer-relative subjectivism does not resolve the underlying power-concentration problem, since human beings still must interpret and encode that transcendent standard, reproducing an isomorphic principal-agent dynamic at the level of religious authority rather than eliminating it.

Suggested Improvements

• Engagement with rival meta-ethical frameworks — Explicitly address secular moral realism, Kantian constructivism, and evolutionary ethics as live alternatives to the theistic-objectivism/programmer-subjectivism dichotomy, and argue for comparative explanatory superiority rather than mere compatibility. This is the argument's single greatest structural vulnerability; without it, the central dichotomy in P4 is easily dismissed by anyone familiar with contemporary metaethics.
• Primary sourcing — Cite direct quotations, publication titles, and dates for Uzan's thought experiment, Roser's essay, and Altman's statement, rather than relying on a podcast's non-verbatim paraphrase. Secondhand, reconstructed characterizations of named thinkers' positions cannot be independently verified and raise fidelity concerns about whether the theological reframing accurately represents what was actually argued.
• Falsifiability — Specify what secular evidence or counter-argument, if any, would count against the claim that these AI-ethics anxieties confirm biblical anthropology. Without stated disconfirmation conditions, the interpretive method risks being unfalsifiable, since agreement confirms the thesis and explicit rejection is pre-explained as noetic blindness.
• Denominational/religious specificity — Address why the argument supports specifically Christian/biblical anthropology rather than other religious or philosophical traditions positing an innate moral law. As stated, the argument at most supports generic transcendent-standard theism, not the specific biblical framework claimed in the conclusion.
• Practical translation — Explain how positing God's character as the moral standard would concretely change AI design, governance, or specification, given persistent interpretive disagreement even among believers about what that standard requires. Without this, the theological diagnosis offers no actionable alternative to the technical and institutional work it critiques, limiting its practical value beyond apologetic use.

Scenario Tests

• Apply the same interpretive move ('X's secular worry is unknowingly a case for biblical anthropology') to any expression of moral concern in medicine, law, or environmental ethics. (Challenges) — The method appears to generalize to any domain where humans express ethical uncertainty, suggesting it functions as an unfalsifiable universal solvent rather than an argument specific to AI or well-supported by the particular evidence cited.
• Present the theological reframing directly to Uzan, Roser, or Altman and ask whether they accept it as an accurate account of their own reasoning. (Challenges) — A likely rejection would test whether 'unbeknownst to them' is a genuine explanatory insight or an imposed reinterpretation immune to the stated views of the very people it claims to explain.
• Substitute a non-Christian transcendent-standard tradition (e.g., Islamic or Stoic natural law) into the same premises to see if the argument equally 'confirms' that tradition. (Challenges) — If the argument's logic works equally well for other traditions, it demonstrates that the argument establishes generic theism at most, not the specifically biblical anthropology claimed in the conclusion.
• Compare the theological explanation against Goodhart's Law and specification-gaming research for predictive and practical value in actually mitigating AI misalignment. (Challenges) — Secular technical frameworks have generated testable predictions and working mitigations; the theological reframing does not add distinct predictive content or engineering guidance, suggesting it is explanatorily optional rather than necessary.

Coherence & Relevance

The argument is internally consistent as a theological reinterpretation and is explicit about its own method (steelmanning secular concerns rather than claiming crypto-theology), which is a genuine point in its favor for intellectual honesty. However, its persuasive force for any audience not already committed to a presuppositional apologetic framework is limited: the central premises rely on redescription rather than independent evidence, the load-bearing dichotomy in P4 is not exhaustive, and the overall interpretive method is structured in a way that resists disconfirmation. The argument succeeds best as an internally coherent worldview application for a sympathetic audience, and least as a piece of comparative reasoning capable of persuading skeptical or religiously neutral readers that biblical anthropology is uniquely confirmed by contemporary AI ethics discourse.

• Uzan's insight that ethics resists formalization and must remain a human endeavor (Moderate) — Accurately motivates the idea that morality involves more than rule-following, but the added claim about ethics being 'corrupted apart from revelation' does not follow from Uzan's own conclusion and is inserted without independent support.
• Walker's reading of 'resists formalization' as imago Dei (Weak) — This is a redescription of the prior premise's data through a theological lens rather than new evidence; it assumes rather than argues that theistic anthropology, rather than naturalistic accounts of moral cognition, is the correct underlying explanation.
• Roser's and Altman's alignment worries as noetic effects of sin (Moderate) — The technical phenomena described (specification gaps, unintended outcomes) are real and important, but well-established secular explanatory frameworks (Goodhart's Law, principal-agent theory) already account for them without requiring the theological layer, leaving the added theological claim under-motivated.
• Subjective ethics reduces to programmer-determined morality (Weak) — This premise is the primary bridge to the conclusion's rhetorical payoff ('ChatGOD'), but it rests on a dichotomy that omits significant non-theistic objectivist positions, weakening the inferential link to the conclusion considerably.

View this argument on LogicFirst.ai