Naming Error: 'Artificial Intelligence' Obscures a Wholly Derived Technology That Originates Nothing of Its Own

Source: "Extracted from EP #135 | A.I. and the Gospel, Just Thinking Podcast by Darrell Harrison & Virgil Wal...."

The Gist

Everything an AI system has—its code, its training data, its goals, and even the yardstick for whether it did well—was put there by people; it never comes up with its own reasons for caring about anything. So calling it 'intelligence' that 'thinks for itself' oversells it: it's really human thinking, packaged and replayed at scale, and no amount of extra computing power changes that basic dependence. That misleading name matters, because it makes people trust these tools more than they should and lets the humans who build and deploy them shrug off responsibility by saying 'the AI decided.'

Conclusion

The term 'artificial intelligence,' especially as deployed by its most prominent promoters, systematically misdescribes the technology by implying self-origination and autonomy it does not and cannot possess. AI systems are constitutively derivative: their content, purposes, and criteria of success all originate outside them, so they mediate and extend human intelligence rather than exercise intelligence of their own. Because dependence is constitutive rather than a temporary engineering limitation, no scaling of these systems yields the self-sufficiency (aseity) that theology reserves for God alone. A more accurate vocabulary—'derived intelligence,' 'machine learning systems,' 'cognitive automation'—would better align public expectation, trust, and legal accountability with what the technology actually is.

Premises

The word 'artificial' correctly identifies these systems as human artifacts; the word 'intelligence,' when paired with rhetoric about machines 'thinking for themselves,' imports a further claim—self-directed cognition—that the artifact status does not support and that its actual operation does not exhibit.
Every constitutive element of a contemporary AI system is externally supplied: the architecture and code, the training corpus, the loss function and reward signal, the benchmarks that define success, the compute and energy that sustain operation, and the human feedback that shapes its outputs. Nothing in the system supplies any of these to itself.
The decisive question is not whether a system depends on external *inputs* (all cognition does) but whether it originates its own *ends*—its purposes, concerns, and standards of correctness. AI systems do not: an optimizer can search a space of solutions but cannot select or revise the objective that defines the space, and any apparent 'goal-setting' is itself a delegated sub-goal within an externally imposed objective.
Human beings differ in kind, not merely degree, on this point: as self-maintaining living organisms we have intrinsic stakes—survival, flourishing, understanding—that are not conferred by an external designer, and we can subject our own criteria of success to revision in light of those stakes. Human intelligence is input-dependent but end-originating; machine systems are input-dependent and end-derived.
Since a system that originates no ends contributes no evaluative or normative standpoint of its own, whatever novelty it produces is recombination within a space delimited by its designers; its outputs are therefore best described as a transformation of human intelligence rather than an instance of intelligence in the full sense.
Autonomy is best understood as a spectrum of degrees of dependence, with total self-sufficiency—aseity, the property of existing and knowing from oneself, which classical theology (cf. Isaiah 40:13–14, Romans 11:34–36) attributes to God alone—as its unreachable limit. Public AI rhetoric implicitly locates these systems far nearer that limit than any account of their operation can justify.
Prominent industry figures and popularizers do in fact describe these systems as entities that think for themselves, will surpass and possibly supersede humanity, and may become sentient—claims that go well beyond the demonstrated capacities of statistical function approximation trained on human artifacts.
There is substantial empirical support that descriptive language shapes user behavior and belief: anthropomorphic framing increases automation bias and overtrust, misleading product names (e.g., 'Autopilot,' 'Full Self-Driving') correlate with misuse, and agentive descriptions ('the algorithm decided') measurably diffuse attributions of responsibility.
Therefore the naming error is not merely semantic: it inflates trust beyond warranted reliability, licenses accountability laundering by shifting blame from designers and deployers to the artifact, and distorts policy debate toward speculative machine agency and away from the concrete human decisions that actually determine outcomes.

Assumptions

The practical core of the argument (misleading naming causes misplaced trust and displaced accountability) stands on purely secular grounds; the theological premise about aseity functions as a limiting concept that clarifies what total self-sufficiency would mean, not as a required premise for the practical conclusion.
'Intelligence' in the full sense includes the capacity to originate and revise one's own ends and standards, not merely to achieve high performance on externally specified tasks; a purely behavioral or performance-based definition would license the current usage and is here rejected with argument rather than assumed.
Self-maintaining biological organisms possess intrinsic ends in a sense that engineered optimizers do not—a claim defended by appeal to the difference between a system that must produce the conditions of its own continued existence and one whose continuation is entirely externally provisioned.
The dependence described is architectural and constitutive rather than a passing limitation, so scale, self-play, synthetic data, or recursive self-improvement would change the degree of human mediation, not its presence—any self-modification still proceeds from externally given objectives and seed conditions.
The public figures cited are representative enough of dominant public discourse and marketing (not necessarily of careful technical researchers) that their language materially shapes lay understanding, regulation, and investment.
Naming is causally significant but not omnipotent; the claim is that terminology substantially shapes default expectations, not that it wholly determines belief.

Analysis

Overall strength: Moderate. Argument type: Deductive.

Premise Strength

P1 - 'artificial' vs. 'intelligence' semantic gap (Moderate) — The linguistic observation is largely uncontroversial, but it is equally compatible with a milder corrective (AI is a term of art, not literal cognition) as with the argument's stronger metaphysical thesis; it under-determines the conclusion on its own.
P2 - all constitutive elements externally supplied (Strong) — This is an accurate, technically well-grounded description of how contemporary ML systems are architected and trained; it is the most secure premise in the argument, though it only establishes input-dependence, not end-dependence.
P3 - ends vs. inputs distinction; systems cannot originate ends (Weak) — This is the argument's load-bearing claim and its most contested one. It is asserted through definition rather than demonstrated, and existing research on emergent or mesa-optimized objectives in complex trained systems provides at least prima facie tension with the claim that goal-like behavior is always cleanly 'delegated.'
P4 - humans differ in kind, not degree (Weak) — This categorical claim is asserted rather than argued against the strongest counterposition: that human 'intrinsic ends' are themselves the product of external evolutionary, biological, and cultural shaping, which would make the human/machine difference one of degree rather than kind—a possibility the argument does not seriously engage.
P5 - AI outputs as recombination/transformation rather than intelligence (Moderate) — This follows validly if P3 and A2 are granted, but it adds no independent evidence beyond restating the prior conceptual claims under a new label.
P6 - autonomy as spectrum with aseity as limiting concept (Moderate) — Useful as a clarifying conceptual device (per A1) and reasonably well-flagged as non-load-bearing for the practical conclusion, but it also introduces the P4/P6 tension noted above and does not independently discriminate between different AI systems' actual degrees of dependence.
P7 - prominent figures describe AI as autonomous/sentient (Moderate) — Well-documented as a real discourse pattern, but reliance on a small number of high-profile, hype-prone exemplars limits how confidently this generalizes to 'dominant public discourse' as opposed to technical or research discourse.
P8 - descriptive language shapes trust and accountability attribution (Moderate) — Consistent with genuine, established findings in HCI and psychology (automation bias, naming effects, diffusion of responsibility), but the extracted premise asserts 'substantial empirical support' without citation, and several specific claims here appear to be elaborations added beyond what the original source material stated.
P9 - naming error inflates trust and enables accountability laundering (Moderate) — A reasonable and policy-relevant inference from P7 and P8, but it moves from general and probabilistic evidence to a fairly confident causal claim about systemic policy distortion, which is better characterized as a plausible inductive extension than a demonstrated conclusion.

Potential Fallacies

Persuasive definition / question-begging (A2, P4, P5) — The argument stipulates that 'intelligence in the full sense' requires the capacity to originate and revise one's own ends (A2), explicitly rejecting behavioral or performance-based definitions because they 'would license the current usage.' This builds the conclusion into the definition rather than independently establishing it, so critics operating with a functionalist or performance-based notion of intelligence are excluded by definitional fiat rather than refuted.
Internal tension between categorical and spectrum framings (P4 versus P6) — P4 asserts that humans and machines 'differ in kind, not merely degree' regarding end-origination, while P6 explicitly frames autonomy as 'a spectrum of degrees of dependence' with aseity as an unreachable limiting case for all finite beings. These two framings pull against each other: if autonomy is fundamentally gradational, the human/machine boundary needs independent justification for being treated as an exception to that gradient, which the argument does not fully supply.
Unfalsifiability by definitional immunization (P3, A4) — By stipulating that 'any apparent goal-setting is itself a delegated sub-goal' (P3) and that no future scaling or self-modification could change the 'presence' of dependence (A4), the argument preemptively reclassifies any potential counterevidence—such as emergent or self-modifying goal behavior in advanced systems—as consistent with its thesis, making the central claim about machine end-derivation effectively untestable rather than an empirically contestable hypothesis.
Hasty generalization / selection bias (P7, A5) — The claim that public AI rhetoric inflates autonomy (P7) rests on a small set of prominent, hype-prone figures (e.g., Musk) rather than a systematic sample of technical or public discourse, while the more circumspect language of much of the AI research community is set aside via A5. This risks generalizing from the most extreme and media-amplified voices to 'dominant public discourse' as a whole.
Equivocation between input-dependence and end-dependence (P2 to P3 transition) — P2 establishes that AI systems depend on externally supplied inputs—a claim equally true of humans, as P4 itself concedes. P3 then treats this as evidence for the much stronger and distinct claim that machines cannot originate ends, without independently demonstrating that inference; the argument's persuasive force partly depends on readers not noticing that these are two different kinds of dependence.
Fallacy of composition (component provenance treated as settling system-level properties) (P2, P3, P5) — The argument infers that the assembled system originates no ends and produces no genuine novelty strictly from the human origin of its individual components (code, data, objective function). Complex systems can exhibit emergent properties not present in or predictable from their parts, so tracing each component's human origin does not by itself settle whether the interacting system can produce behavior functionally resembling goal revision.

Counterarguments

P4 (kind vs. degree distinction) (High impact) — Human 'intrinsic ends'—survival, reproduction, flourishing drives—can themselves be characterized as externally instilled by evolutionary and cultural processes rather than self-originated in any deep metaphysical sense. If so, the human/machine difference is one of degree and complexity, not the categorical kind the argument requires, undermining the central premise on which P3, P5, and the conclusion depend.
P3 (systems cannot select or revise their objectives) (High impact) — Research on mesa-optimization, goal misgeneralization, and emergent in-context objectives in large trained models shows that learned systems can develop internal proxy objectives not directly specified by designers, complicating the claim that any apparent goal-setting is always a cleanly delegated sub-goal.
A2 (intelligence requires end-origination) (High impact) — A functionalist or performance-based account of intelligence—widely used in cognitive science and AI research—would treat task performance and problem-solving capacity as sufficient for the term 'intelligence,' regardless of the origin of goals; on this view current usage of 'AI' is not a misnomer at all, and the argument's rejection of this view is achieved by definitional stipulation rather than independent refutation.
Overall conclusion (renaming as remedy) (Medium impact) — Terminology-reform efforts have a mixed historical track record (e.g., 'global warming' to 'climate change'; continued use of 'Autopilot' despite documented harms) absent binding regulatory mandates, suggesting that relabeling alone is unlikely to achieve the trust-calibration and accountability goals the argument sets for it, and may even reduce perceived risk and regulatory urgency if 'derivative' framing is read as reassuring rather than cautionary.
A1 (theological premise as non-load-bearing) (Medium impact) — The source material's own stated conclusion is explicitly theological ('AI...can never be or become God'), suggesting the aseity framework is not merely an illustrative limiting concept but the argument's actual motivating point; recasting it as secularly self-sufficient may not faithfully represent the source's structure and could be seen as repackaging a doctrinally motivated claim in secular-sounding language.
P7 (representativeness of cited public rhetoric) (Medium impact) — A large portion of the AI research community explicitly avoids anthropomorphic or autonomy-implying language and has actively critiqued such framing (e.g., 'stochastic parrot' critiques); treating hype-prone public figures as representative of 'the term AI' broadly risks a weak-manning of the field's actual discourse.

Suggested Improvements

Justification of the end-origination criterion (A2, P3, P4) — Explicitly engage the naturalistic/evolutionary counterargument that human ends are also externally shaped, and explain what principled, non-question-begging criterion distinguishes 'genuine' end-origination from highly complex delegated goal-structures. This is the argument's single point of greatest vulnerability; without addressing it, the categorical human/machine distinction reads as asserted rather than established.
Empirical grounding of P8 — Cite specific studies (e.g., NHTSA reports or peer-reviewed HCI research on Autopilot-related incidents, published experiments on automation bias and responsibility attribution) rather than asserting 'substantial empirical support' generically. Concrete citations would let readers assess sample sizes, effect magnitudes, and generalizability, strengthening the argument's most policy-relevant and currently under-sourced premise.
Reconciling P4 and P6 — Either drop the 'kind, not degree' language in favor of a strong-degree claim consistent with the spectrum model, or provide independent argument for why the human/machine boundary is a genuine categorical break rather than an extreme point on the same continuum. Resolving this internal tension would remove an easy point of attack and make the argument's conceptual architecture more consistent.
Engagement with emergent/mesa-optimization research — Address specific technical findings on goal misgeneralization and emergent objectives in trained systems, explaining why these do or do not constitute genuine end-revision under the argument's framework. Without this engagement, A4's claim that scaling changes only 'degree, not presence' of dependence appears to foreclose a live empirical question by definition rather than argument.
Transparency about source and audience — Clarify explicitly that the theological aseity framework is central to the source's original argument and audience, rather than presenting the practical/secular framing as though it were the argument's primary or independent structure. This would improve intellectual honesty about the argument's provenance and avoid the appearance of repackaging a doctrinal claim for a broader secular audience under different justificatory cover.
Practical remedy design — Pair proposed terminology changes with concrete regulatory or legal mechanisms (e.g., mandatory disclosure standards, liability rules fixing responsibility on designers/deployers) rather than relying on vocabulary substitution alone. Historical precedent suggests naming changes without enforcement mechanisms rarely achieve behavioral or accountability shifts on their own, and could even reduce perceived urgency for stronger regulatory action.

Scenario Tests

A trained model exhibits emergent, unpredicted internal objectives (mesa-optimization) that diverge from its specified training objective, observable in current alignment research. (Challenges) — This would complicate P3's claim that apparent goal-setting is always a cleanly delegated sub-goal, showing that the input/ends boundary is less stable in practice than the argument assumes.
A controlled study compares user trust and misuse rates for functionally identical systems labeled 'AI assistant' versus 'pattern-matching tool.' (Supports) — Directly confirmatory results would substantiate P8/P9's causal claims about naming effects on trust and accountability, strengthening the argument's most defensible practical core.
Regulatory bodies mandate a switch to terms like 'cognitive automation' but without accompanying liability or disclosure rules, and marketing continues framing these systems as increasingly capable. (Challenges) — If overtrust and accountability diffusion persist despite the terminology change, this would support the concern that renaming alone is a low-leverage intervention relative to underlying commercial incentives and regulatory structures.
A cross-cultural or secular philosophical audience evaluates the argument's aseity-based framing without prior theological background. (Neutral) — The theological material is unlikely to add persuasive force for this audience and may reduce perceived neutrality, though A1's disclaimer partially insulates the practical conclusion from this effect.

Coherence & Relevance

The argument coheres well as two linked but logically distinct projects: a conceptual argument (P1-P6) establishing that AI systems are 'end-derived' rather than 'end-originating,' and an empirical/policy argument (P7-P9) establishing that misleading rhetoric about that distinction causes real-world harm. The empirical segment is more secure and largely stands even if the conceptual segment is contested, since the practical harms of overtrust and accountability diffusion do not strictly require accepting the strong metaphysical thesis about end-origination—only that some AI rhetoric overstates present capabilities, which is much easier to establish. The conceptual segment's coherence, however, is weakened by an internal tension between its categorical (P4) and gradational (P6) framings of autonomy, and by its heavy reliance on a stipulated definition of 'intelligence' (A2) that a substantial body of contrary opinion in cognitive science and AI research would reject. The theological material, while explicitly bracketed as non-load-bearing (A1), appears in the source's actual conclusion to be central rather than merely illustrative, which raises a fidelity question about how the practical and conceptual segments were reframed relative to the original text.

P1 (Moderate) — Establishes a plausible semantic tension but does not by itself distinguish a modest corrective (AI as term of art) from the stronger metaphysical thesis the argument goes on to build.
P2 (Strong) — Accurately establishes input-dependence but does not, on its own, establish end-dependence, which is asserted separately in P3.
P3 (strong (for the conceptual segment), but contested) — This is the pivotal premise; its truth is asserted via definitional stipulation (A2) and is in tension with technical findings on emergent/mesa-objectives, leaving a genuine gap between the premise and demonstrated fact.
P4 (strong (for the conceptual segment), but contested) — Relies on an unaddressed naturalistic counter-account of human goal formation; the 'kind, not degree' claim is asserted rather than defended against its strongest rival.
P5 (Moderate) — Follows validly from P3-P4 given A2, but supplies no independent support beyond restating prior premises under the 'full sense of intelligence' label.
P6 (Moderate) — Functions as an explicitly flagged limiting concept (A1) rather than a load-bearing link to the practical conclusion, but its 'spectrum' framing sits uneasily with P4's categorical claim.
P7 (Moderate) — Establishes that hyperbolic rhetoric exists but relies on a narrow, potentially unrepresentative sample of public figures relative to the broader claim about 'dominant public discourse.'
P8 (Strong) — The underlying phenomena (automation bias, misleading naming, diffusion of responsibility) are genuinely well-studied, but the premise as stated is uncited and includes specificity that appears to exceed what the source material directly supported.
P9 (Moderate) — A reasonable inductive extension of P7-P8, but stated with a confidence ('therefore') that outstrips the probabilistic nature of its empirical support.

View this argument on LogicFirst.ai