Using AI Under Common Grace Is Legitimate; Using It to Play God Is Not

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

The Gist

They are not anti-AI. They used it to make the episode. The line is: use the tool, check the work, thank God for common grace, and do not let the tool become a way of pretending you are God.

Conclusion

Christians need not reject AI as such. Under God’s common grace, even unbelievers can make tools that save time and serve dominion. The distinctive Christian concern is not the spreadsheet or the search; it is the almost unbounded potential of these systems to tempt people to play God, to celebrate the removal of God from the story, and to treat the making of “autonomous intelligent beings” as a triumph. Use is permitted. Uncritical worship, and the refusal to verify, is not.

Premises

1. Harrison used multiple AI applications to prepare the episode and calls the time-saving “unarguable.” Walker agrees: it shrinks research time. Genesis 1:28’s dominion mandate includes developing creation’s resources. Matthew 5:45’s common grace means rain falls on the just and the unjust; useful inventions from unbelievers are not thereby unclean.
2. The communicable attributes of intelligence, thought, and creativity are gifts. Availing ourselves lawfully of means that make work more productive can be done to God’s glory. Nobody in the episode is arguing against practical benefit as such.
3. What does concern them is the seduction: proponents lauding the ability to create intelligent, supposedly autonomous beings and to take God out of the equation. The live risk is not Excel. It is self-deification.
4. Even as a research aid, the tool is fallible. Walker’s practice is the implied rule of use: treat the output as a start, then check original sources. Efficiency without verification is not wisdom.

Assumptions

• Common grace does not baptize every use. It explains why a fallen inventor can still make a helpful instrument.
• “Playing God” here means claiming creator-status, moral supremacy, or soul-satisfying presence for the machine or its makers, not using a calculator.
• 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 1:00:13–1:19:58 of the episode; it is not a verbatim transcript.

Analysis

Overall strength: Moderate. Argument type: Deductive.

Premise Strength

• P1 (time-saving + dominion mandate + common grace) (Moderate) — The theological warrant (Gen 1:28, Matt 5:45) is internally coherent within a Reformed framework and well-established as doctrine, but its application to AI specifically is an unargued analogical extension, and the empirical time-saving claim rests on n=2 anecdotal testimony treated as 'unarguable' when it is in fact actively contested in productivity research.
• P2 (communicable attributes as gifts, productivity to God's glory) (Weak) — Largely restates P1's permissibility claim from the same theological source rather than adding independent support; it applies equally to any productivity tool and does not differentiate AI's distinctive risk profile.
• P3 (seduction toward self-deification) (Moderate) — This is the argument's most distinctive and diagnostically important premise, but it generalizes from unnamed, unsourced 'proponents,' risking a strawman or an overweighting of the most sensational AI-hype rhetoric relative to more common, mundane industry framing.
• P4 (fallibility and verification) (Strong) — Well-supported by documented AI failure modes (hallucination, factual error) and is the argument's most empirically solid claim, though the 'verify against sources' rule is underspecified (no threshold for sufficiency) and does not extend to generative, relational, or agentic use-cases lacking a checkable 'original source.'

Potential Fallacies

• Category shift between epistemic and moral registers (Inference from P4 to the conclusion) — The argument moves from an epistemic/practical claim ('unverified efficiency is not wisdom,' P4) to a moral-theological verdict ('is not permitted') without a bridging premise showing that imprudence rises to the level of impermissibility. This doesn't break the argument's validity given its stipulated definitions, but it papers over a shift in the kind of 'ought' being invoked.
• Unattributed generalization about 'proponents' (P3) — The central diagnosed danger — that AI proponents laud the creation of autonomous, God-displacing beings — is asserted without naming individuals, sources, or a representative sample. This resembles hearsay treated as settled fact and leaves it unclear whether the argument targets a fringe rhetorical minority or a mainstream industry stance.
• False dichotomy / category narrowing (Overall framing, especially P3 and the conclusion) — By framing the field of concern as either mundane tool-use (like a spreadsheet) or theologically catastrophic self-deification, the argument sidelines a wide middle ground of ethically significant AI risks — bias, surveillance, labor displacement, epistemic dependency — that fit neither pole.
• Hasty generalization from testimonial evidence (P1) — A personal, uncontrolled anecdote (two hosts' experience preparing one episode) is elevated to an 'unarguable' general claim about AI's time-saving benefits, without comparative data or controls.

Counterarguments

• Conclusion / overall framework (High impact) — The use-versus-worship dichotomy locates all moral risk in the user's internal disposition, but this cannot distinguish a Christian using AI to write efficient sermons from one using it to run an exploitative surveillance or gig-labor platform, since both can equally disclaim 'worship' and perform token verification. A framework silent on structural harms (power concentration, labor displacement, bias, environmental cost) is incomplete as an applied AI ethic.
• P4 (High impact) — The 'verify against original sources' safeguard only works for factual/research use-cases; it is a category error for generative, creative, relational, or autonomous-decision applications (companionship apps, AI counseling, autonomous agents) where there is no discrete 'source' to check.
• P1 / A1 (Medium impact) — The common-grace defense ('useful inventions from unbelievers are not thereby unclean') proves too much if left unqualified — the same logic would license weapons, propaganda technology, or exploitative systems, since it says nothing about the nature or scale of the harm a given invention can cause.
• A2 (definition of 'playing God') (High impact) — The definition is asserted ostensively rather than derived, and provides no operational test for the vast middle ground of agentic or anthropomorphic AI. It also fails to catch hubris dressed in pious language (e.g., 'AI reflects the imago Dei's creativity' used to justify ethically dubious deployments), since it only flags explicitly secular self-deification rhetoric.
• P3 (Medium impact) — If systematic discourse analysis showed that most AI industry communication is cautious, tool-oriented framing rather than 'autonomous being' triumphalism, the premise's claim that self-deification is 'the live risk' would be undercut as targeting a rhetorical minority rather than a pervasive hazard.

Suggested Improvements

• Sourcing of the 'proponents' claim — Cite specific individuals, companies, or statements exhibiting the 'autonomous being' / God-displacing rhetoric, and indicate how representative this is of the industry as a whole. Without this, the argument's central diagnostic claim is unfalsifiable and vulnerable to strawman charges, weakening its persuasive and evidentiary force.
• Operationalizing 'playing God' and 'verification' — Supply concrete case studies or decision criteria for where legitimate dominion-work ends and self-deification begins, and specify what counts as sufficient verification for non-factual AI use-cases. As stated, both boundaries are asserted rather than derived, making the framework rhetorically flexible enough to bless or condemn almost any application depending on the arguer's prior commitments.
• Incorporating outcome/neighbor-love criteria — Supplement the attitude-based test (worship vs. use) with consequence-oriented criteria — does this deployment harm third parties, concentrate power unjustly, or deceive — drawing on Christian teaching about justice and neighbor-love, not just idolatry. This would close the argument's most significant blind spot: its inability to distinguish pious-but-harmful AI use from genuinely responsible use, since its current test is entirely internal/dispositional.
• Empirical grounding of the time-savings claim — Replace the anecdotal 'unarguable' characterization with reference to broader, controlled findings on AI productivity effects, including verification costs. Strengthens the argument's empirical credibility and avoids overclaiming from a sample of two personal testimonies.

Scenario Tests

• A Christian-run company uses AI for predictive hiring or surveillance, never claims the system is autonomous or god-like, and performs nominal source verification on outputs. (Challenges) — The framework offers no grounds to condemn this use since its entire test is attitudinal (worship vs. use); this exposes the argument's blind spot regarding structural/third-party harms.
• An AI companionship or pastoral-counseling application is used, where there is no discrete 'original source' to verify against. (Challenges) — P4's safeguard collapses entirely, leaving only the vague 'don't worship it' rule, which cannot adjudicate this case.
• A user treats AI output as a preliminary draft and diligently checks factual claims against primary sources before publishing research. (Supports) — This is the paradigm case the argument was built for, and its guidance (efficient but verified use) functions well here.
• AI marketing rhetoric is examined across the industry and found to be predominantly cautious and tool-oriented, with 'autonomous being' triumphalism confined to a small, non-representative subset of high-profile statements. (Challenges) — Would undercut P3's claim that self-deification is 'the' distinctive live risk, suggesting the argument targets a rhetorical minority rather than a pervasive threat.

Coherence & Relevance

The argument is internally consistent and validly structured as a bipartite case — a permission-establishing sub-argument (P1, P2, A1) paired with a restriction-establishing sub-argument (P3, P4, A2) — provided its Reformed theological premises and stipulated definitions are granted. Its chief weakness is not logical invalidity but narrow scope and thin evidentiary support: it rests on anecdotal testimony from two speakers, an unsourced characterization of AI industry rhetoric, and a use/worship distinction that, while conceptually clear, offers no operational test for the vast middle ground of AI applications (agentic, relational, generative) where neither 'it's just a calculator' nor 'claiming creator-status' cleanly applies. The framework also concentrates moral risk entirely in individual disposition, leaving structural and societal harms (power concentration, bias, labor displacement) outside its evaluative reach.

• P1 (Strong) — Establishes the permission side of the conclusion well within the stated theological framework, but the analogy from ancient agrarian dominion language to synthetic cognition, and the 'unarguable' empirical claim, are asserted rather than defended.
• P2 (Moderate) — Reinforces rather than independently supports P1; adds rhetorical weight but little new evidential content toward the conclusion.
• P3 (Strong) — Directly grounds the conclusion's distinctive prohibition, but relies on an unsourced characterization of 'proponents' that weakens its evidentiary standing.
• P4 (Strong) — Directly grounds the conclusion's verification requirement; well-supported for factual research use but not generalized to other AI applications the conclusion implicitly covers.

View this argument on LogicFirst.ai