“Artificial Intelligence” Misnames Extrinsic Information as Intrinsic Intelligence

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

The Gist

They are not saying the software is fake. They are saying it isn’t a mind. It doesn’t think from the inside; people put information in, and it automates what comes out. So “artificial intelligence” is the wrong name, and “autonomous” is even worse.

Conclusion

The phrase “artificial intelligence” does not accurately describe the systems in view. What they produce is not intelligence arising from the system itself, but information generated in dependence on human programming, data, and prompts. Intelligence, in the sense the hosts are using, is an intrinsic capacity of a mind. AI’s outputs are entirely extrinsic. A more accurate label would be something like “automated information.” The A in AI stands for artificial, not autonomous.

Premises

1. “Artificial” ordinarily means made by humans rather than occurring naturally, and often implies imitation of a natural counterpart. The outputs of these systems are not produced apart from human beings. They are the product of human design, code, training data, and queries. In that sense the so-called intelligence is natural to its human sources, not a new kind of mind that appeared on its own.
2. Intelligence, as the hosts define it, is intrinsic: it belongs to the agent. Every element of what is called AI “intelligence” is imported from outside the system. None of it is in and of itself. Virgil’s summary is the steelman: the permutations the system engages in are set by human programmers and human-curated algorithms.
3. These systems do not generate information ex nihilo. They act only as they have been coded and prompted to act. They therefore produce information, not intelligence in the intrinsic sense. Calling the product “intelligence” is a non sequitur.
4. Popular rhetoric, including from prominent technologists, often treats the A as if it meant autonomous: a being that thinks, wills, and eventually outstrips us on its own. That reading of the name is false to how the systems actually work, and it distorts Christian discernment about them.

Assumptions

• The argument is about the referent of the words, not a claim that machine learning is useless. Harrison used several AI tools to research the episode and still rejects the name.
• “Intrinsic intelligence” here is a philosophical-theological claim about minds, not a denial that models can transform inputs into surprising outputs.
• 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 27:51–36:25 of the episode; it is not a verbatim transcript.

Analysis

Overall strength: Weak. Argument type: Deductive.

Premise Strength

• P1: “Artificial” ordinarily means made by humans... the so-called intelligence is natural to its human sources, not a new kind of mind. (Moderate) — Accurately describes AI's causal dependence on human design and data, but the added etymological claim that 'artificial' implies mere imitation rather than genuine instantiation is contradicted by common usage patterns ('artificial heart,' 'artificial light'), weakening the premise's support for the conclusion.
• P2: Intelligence, as the hosts define it, is intrinsic... Every element of what is called AI 'intelligence' is imported from outside the system. (Weak) — This is the argument's load-bearing premise, and it functions as a stipulated definition rather than an established one. It is not defended against the standard functional/operational definition of intelligence used in AI research since the field's founding, and it is not shown to apply asymmetrically to human cognition, which is also externally shaped.
• P3: These systems do not generate information ex nihilo... Calling the product 'intelligence' is a non sequitur. (Weak) — Restates P2's dependency claim as a conclusion (non sequitur) without new support, and does not engage emergent behaviors in large models (in-context learning, capabilities not directly traceable to specific training instructions) that complicate a clean 'entirely extrinsic' characterization.
• P4: Popular rhetoric... treats the A as if it meant autonomous... distorts Christian discernment. (Weak) — An empirical/sociological claim offered without citation of specific technologists or statements. It is largely disconnected from the deductive core (P1-P3) and risks characterizing a rhetorical minority as representative of mainstream AI discourse, which tends to explicitly disclaim autonomy and consciousness.

Potential Fallacies

• Question-begging (persuasive) definition (P2, and the conclusion that depends on it) — The argument stipulates that 'intelligence' necessarily means an intrinsic, agent-owned capacity, and then treats AI's failure to meet this stipulated criterion as proof the name is wrong. This makes the conclusion true almost by definition rather than by engagement with rival, widely-used functional or behavioral definitions of intelligence (the kind used by the AI research field itself).
• Etymological overreach / false analogy (P1) — The claim that 'artificial' implies mere imitation lacking the named property doesn't hold up against ordinary usage: 'artificial heart,' 'artificial light,' and 'artificial sweetener' are all human-made things that genuinely perform the function named, not counterfeits. This weakens the inference that 'artificial' intelligence must mean 'not really' intelligence.
• False dichotomy (intrinsic vs. extrinsic) (P2, P3) — The argument treats intrinsic origination and extrinsic dependency as mutually exclusive and exhaustive categories, leaving no room for emergent, functional, or graded conceptions of intelligence that many philosophers of mind and AI researchers take seriously. This forecloses middle-ground positions without directly refuting them.
• Unsupported generalization (P4) — The claim that 'popular rhetoric, including from prominent technologists' treats AI as autonomous is asserted without naming any specific individuals, quotes, or sources. As stated it is unverifiable and risks describing a rhetorical fringe (or media hype) as if it were the dominant or expert position, when many AI researchers explicitly disclaim autonomy or consciousness for current systems.
• Asymmetric application of the intrinsic/extrinsic standard (Cross-cutting: P1, P2, P3) — Human cognition is also profoundly shaped by external, non-self-generated inputs (genetics, upbringing, language, culture, education), yet is still called 'intrinsic' once instantiated. The argument does not explain why dependency disqualifies machine processing from the 'intelligence' label but not human cognition, leaving an unexamined double standard that likely rests on an unstated theological premise (e.g., ensoulment or imago Dei) rather than a neutral criterion of intrinsicality itself.

Counterarguments

• P2 / Conclusion (High impact) — Since the term was coined at the 1956 Dartmouth Conference, 'artificial intelligence' has used 'intelligence' functionally/operationally (the capacity to perform tasks that would require intelligence if done by a human), not metaphysically. Under this standard, established definition, the name is coherent regardless of causal dependence on programmers and data, just as a calculator's arithmetic still counts as calculation despite being fully programmed.
• P1, P2, P3 (High impact) — Applying the intrinsic/extrinsic standard consistently would also disqualify human intelligence, since it too depends on inherited genetics, developmental environment, language, and culture that the person did not self-generate. Without a principled criterion distinguishing biological from silicon dependency (e.g., an explicit appeal to soul or divine image-bearing), the standard appears applied selectively.
• P1 (Medium impact) — Ordinary usage of 'artificial X' (artificial heart, artificial light, artificial sweetener) typically denotes a genuine, functioning instance of X produced by non-natural means, not a mere imitation lacking the property — undermining the etymological claim that 'artificial' intelligence must not be real intelligence.
• P4 (Medium impact) — Many prominent AI researchers explicitly deny that current systems possess autonomy, agency, or consciousness (e.g., 'stochastic parrot' critiques argue the opposite direction — that hype culture, not researchers, over-anthropomorphizes AI). This suggests P4 may be attacking a rhetorical fringe rather than the field's mainstream position.
• P1, P3 (Medium impact) — Emergent behaviors in large-scale models (in-context learning, capabilities not explicitly programmed) are not cleanly traceable to specific human design choices, complicating the claim that 'every element' of AI output is simply imported from outside the system.

Suggested Improvements

• Definitional grounding — Explicitly acknowledge the functionalist/operational definition of 'intelligence' used in AI research and cognitive science, and argue for the superiority or relevance of the intrinsic-mind definition for the argument's theological purposes, rather than presenting the intrinsic definition as if uncontested. This would convert a question-begging stipulation into a defended philosophical position, making the argument persuasive beyond an audience that already shares its theological anthropology.
• Handling the human-cognition symmetry problem — Supply an explicit principle (e.g., appeal to the soul, imago Dei, or agency in a theologically specific sense) that distinguishes human dependency on external inputs from AI's dependency on external inputs. Without this, the intrinsic/extrinsic criterion appears arbitrarily applied only to machines, inviting an easy reductio that undermines the argument's core distinction.
• Evidentiary support for P4 — Cite specific statements, speakers, or dates substantiating the claim that 'prominent technologists' treat AI as autonomous. As stated, this is an unverifiable generalization; naming sources would allow the claim to be evaluated rather than assumed, and would clarify whether it targets mainstream expert opinion or hype-driven outliers.
• Engagement with emergence — Address emergent capabilities (in-context learning, unexpected behaviors from scale) that are not directly traceable to specific programmer choices. Ignoring this leaves P1 and P3's 'entirely extrinsic, only as coded' claims vulnerable to being outpaced by how modern large models actually behave.
• Scope discipline — Clearly separate the semantic/naming critique from any implied claims about AI's practical capabilities or risks, explicitly stating that renaming does not resolve empirical questions about capability, safety, or alignment. This prevents the argument from being used (intentionally or not) to dismiss substantive AI risk or capability discussions as 'merely semantic.'

Scenario Tests

• Apply the intrinsic/extrinsic standard consistently to human cognition (shaped by genetics, upbringing, language, and culture). (Challenges) — If dependency on external, non-self-generated inputs disqualifies a process from being called 'intelligence,' human cognition would seem equally disqualified, revealing an unexamined double standard unless a further theological premise (e.g., ensoulment) is made explicit.
• Test P1's etymological claim against other 'artificial X' terms (artificial heart, artificial light, artificial sweetener). (Challenges) — These terms denote genuine, functioning instances of X made by non-natural means, suggesting 'artificial' does not inherently imply mere imitation lacking the named property, weakening the premise's linguistic foundation.
• Evaluate the argument against the AI field's own founding, functional definition of 'intelligence' (Dartmouth 1956 onward). (Challenges) — Under the field's own operative definition (capacity to perform cognition-associated tasks), 'artificial intelligence' is a coherent and accurate label; the 'misnomer' charge only holds against a different, non-standard definition.
• Restrict evaluation to an audience that already shares the argument's theological anthropology (intrinsic mind/soul as constitutive of true intelligence). (Supports) — Within this specific interpretive community, the argument is internally coherent and rhetorically effective as a corrective against overhyped 'autonomous AI' rhetoric, even though it does not generalize to broader secular or technical audiences.

Coherence & Relevance

The argument is a tightly constructed definitional syllogism: if one grants that 'intelligence' means intrinsic, agent-owned origination (P2), the inference that AI's externally-dependent outputs fail to qualify (P1, P3) follows with reasonable internal consistency, and the conclusion (a different name is more accurate) tracks fairly naturally from that. However, this coherence is conditional and somewhat brittle: it depends entirely on accepting a specific, theologically inflected definition of intelligence over the standard functional definition used by the field being critiqued, and it does not resolve why the same dependency-based standard would not equally disqualify human cognition. P4 sits outside this deductive chain as a supplementary, evidentially unsupported observation about public discourse, contributing rhetorical urgency but no additional logical support to the central claim.

• P1: Artificial means human-made; outputs are not produced apart from human beings. (Moderate) — Establishes causal dependency well but overreaches in claiming 'artificial' implies non-genuine instantiation, which ordinary usage does not support.
• P2: Intelligence is intrinsic; every element of AI 'intelligence' is imported from outside. (Strong) — This is the load-bearing premise for the conclusion, but it is a stipulated definition not defended against rival (functionalist) accounts, and its intrinsic/extrinsic criterion is not shown to be uniquely applicable to machines rather than humans.
• P3: Systems do not generate information ex nihilo; therefore produce information, not intelligence. (Strong) — Largely restates P2 as an inference rather than adding independent support; does not address emergent, not-directly-programmed behaviors that complicate the 'entirely extrinsic' claim.
• P4: Popular rhetoric treats the A as autonomous, distorting discernment. (Weak) — Loosely connected to the deductive core; functions as rhetorical reinforcement about a separate concern (public misunderstanding) rather than a premise required to establish the definitional conclusion, and is asserted without citation.

View this argument on LogicFirst.ai