Artificial Intelligence: Fact, Misnomer or Oxymoron?

May 17, 2026 · Paul W. Barnes · Mind

Intelligence is a well-established concept. It has been examined, refined, and contested across every serious tradition that thinks about minds, for as long as there have been such traditions. Aristotle distinguished it as the rational soul. The medieval philosophers parsed it into the active and passive intellect. Modern philosophy from Descartes through Kant treated it as the faculty of grasping principles. The phenomenological tradition examined it as the registered apprehension of unity across difference. Cognitive science studied its mechanisms. The psychological literature mapped its varieties. The concept is not new and not unsettled. We know, with the kind of accumulated clarity that takes millennia to develop, what intelligence is.

What is unsettled is the phrase “artificial intelligence.” The technology it names is real. The systems being built produce extraordinary outputs, outperform humans on specific tasks, and are reshaping the work of every field they touch. The technology is not in doubt. What is in doubt is what to call it.

Three readings of the phrase are in circulation, and each commits to a different claim about what the technology actually is.

The first reading treats “artificial intelligence” as fact. AI systems are intelligent, in roughly the sense humans are, perhaps in a different mode or to a different degree, but the intelligence is real. The mainstream of AI discourse holds this reading either explicitly or by default.

The second reading treats “artificial intelligence” as a misnomer. The technology is real, but the label is wrong. The systems do something, namely produce outputs that resemble what intelligent beings produce, but the underlying feature that makes intelligence intelligence is not present. We have a powerful new thing, mis-named.

The third reading treats “artificial intelligence” as an oxymoron. The two terms select for incompatible properties. “Artificial” names a kind of substrate; “intelligence” names a capacity that cannot be hosted by that substrate. The phrase is not just an unfortunate label. It is a phrase that cannot refer.

This essay argues for the third reading. The first reading is widespread but cannot survive examination of what intelligence is. The second reading captures something real about the misuse of language but understates the structural problem. The third reading is where the analysis arrives once intelligence is taken seriously and the nature of artificial systems is taken seriously.

This is the first entry in a series called Words Under Capture, which will work through the vocabulary of mind one word at a time and show how each is being deployed by AI discourse to do the same kind of work. The series is not an attack on AI as a technology. It is a diagnostic of how language is being used. Once the operation is named, it becomes visible. Once visible, it can be refused.

What intelligence is

To evaluate any of the three readings, we have to be clear about what intelligence is. The concept has been examined enough that its core features can be sketched confidently.

Intelligence, at its most fundamental, is the capacity to register a gap between what is and what is needed, and to find a way across that gap. The registration is the key feature. An intelligent system does not merely produce outputs that close gaps; it grasps that there is a gap to be closed, what makes the gap a gap, and what kind of response would close it. The grasping is not a behavior. It is an interior event that has a felt character, the sense of having understood, of having seen what the situation requires.

This is why intelligence has always been associated with consciousness in the philosophical tradition. Not because the two are identical, but because intelligence in its phenomenal sense requires a system that registers things from the inside. A thermostat closes the gap between actual temperature and set temperature, but it does not register the gap as a gap. The closure happens through it, not because of any grasping that occurred within it. A chess engine closes the gap between current position and winning position, but it does not register the position as a position to be won. A large language model produces outputs that close gaps in expected text continuations, but it does not register the gap between what was asked and what is needed.

None of those systems is doing what intelligence does. They are producing outputs that resemble what intelligent systems produce, through processes that do not include the registration that intelligence requires. The resemblance is real. The intelligence is not present.

This is not a controversial claim within the philosophical tradition that has examined intelligence. It is the standard understanding. Aristotle would have recognized it. Aquinas would have recognized it. Kant would have recognized it. Husserl would have recognized it. The phenomenological account of intelligence as the grasping of intelligible structure, in which the grasping is itself a phenomenal event, is the inheritance of careful thought across centuries.

The question now is which of the three readings of “artificial intelligence” best survives this understanding.

Fact

The mainstream reading holds that AI is intelligent. Not metaphorically intelligent. Not “intelligent” in scare quotes. Intelligent in roughly the sense that matters, with perhaps minor adjustments to how the concept is understood.

The reading is supported by three observations. First, AI systems perform well on intelligence benchmarks. They pass exams designed to measure human intelligence. They beat humans at strategic games. They produce coherent reasoning in natural language. The behavioral evidence is substantial.

Second, the underlying concept of intelligence is held to be flexible. Different traditions emphasize different features. Cross-cultural definitions vary. If “intelligence” can mean what it means in human contexts, it can mean something analogous in machine contexts. The concept has always accommodated variation, and it can accommodate this.

Third, denying AI intelligence is held to be either anthropocentric or unfalsifiable. If a system produces all the outputs intelligence produces, the insistence that some additional ingredient is missing, whether phenomenal experience, registration, or grasping, is either a parochial demand that intelligence look human or an unverifiable claim about interior states.

Each of these three supports fails on examination.

The benchmarks measure outputs, not registrations. A system designed to perform well on a benchmark constructed to measure intelligence in humans will perform well on that benchmark to the degree it produces the outputs the benchmark scores, regardless of whether the production involves the registration that the benchmark was originally designed to detect. The benchmark assumes that producing the outputs is evidence of the underlying capacity, because in the original test population, humans, the two are correlated. The correlation breaks when systems are built specifically to produce the outputs without the capacity.

The flexibility of the concept is real but bounded. Intelligence has accommodated variation in mechanism, expression, and degree. It has not accommodated variation in the registration that defines it. A definition of intelligence that drops registration is not an expanded definition of intelligence; it is a different concept that has borrowed the word.

The anthropocentrism charge confuses two distinct claims. The claim that intelligence requires interior registration is not the claim that intelligence requires being human. Animals register. The capacity for registration is widespread in nature and is the feature that makes intelligence intelligence rather than mere reactivity. The demand that AI systems show registration to be called intelligent is not the demand that they be human. It is the demand that they have the feature without which the word does not apply.

The fact reading cannot be sustained. AI is not intelligent in the meaningful sense of the word. We could stop here, but stopping here leaves the misuse of the word unexplained. The question becomes: if AI is not intelligent, why does the phrase “artificial intelligence” persist with such conceptual confidence?

Misnomer

The second reading addresses this question. AI is something, a powerful technology with significant capabilities, but “intelligence” is the wrong word for it. The phrase is a misnomer.

The misnomer reading treats “artificial intelligence” as analogous to other linguistic misfires. We call certain numerals “Arabic” when they originated in India. We call a metal foil “tin” when it has been aluminum for a century. We name things imprecisely all the time, and the imprecision usually causes no serious damage, because the referent of the term remains clear.

If AI is a misnomer, the same logic applies. The technology exists. We have just chosen, or inherited, the wrong word for it. A better word, perhaps pattern generators, statistical predictors, output systems, or computational tools, would be more accurate. The misnomer is a problem of labeling, correctable in principle by attention to vocabulary.

This reading captures something real. The label is indeed wrong. The technology does not have the features the label names. And the cultural consequences of getting the label wrong are significant in their own right.

The misnomer transfers the cultural weight of intelligence to AI systems. Intelligence is something humans have valued, sought, measured, and admired for as long as the concept has existed. By calling AI “intelligence,” the discourse positions AI systems as worthy of the same regard. Investments flow. Attention concentrates. Resources allocate. The redirection happens because the word is doing work that the underlying claim about the systems would not do on its own.

The misnomer forecloses certain criticisms. If AI is intelligent, then concerns that AI systems lack what makes intelligence intelligence, namely phenomenal registration, the grasping of meaning, the felt difference between understanding and producing tokens, can be characterized as either philosophical hair-splitting or anthropocentric bias. The misnomer protects the systems from being examined for what they actually are.

And the misnomer reshapes how humans understand their own cognition. When the word “intelligence” applies equally to AI systems and to humans, the working meaning of intelligence shifts to fit both. The features of human intelligence that AI cannot reproduce, the phenomenal aspects, the registration, the grasping, drop out of the working concept. Humans then come to understand their own intelligence in terms compatible with what AI does. The reduction of human cognition to functional output proceeds quietly, through the shared vocabulary, without anyone arguing for it explicitly.

The misnomer reading recognizes all of this. It treats the misuse of language as serious, consequential, and worth resisting.

But it understates the problem. Calling “artificial intelligence” a misnomer suggests that some other label could correctly apply to the same thing, that we are looking at a real entity that has just been mis-labeled. The third reading challenges this assumption.

Oxymoron

The third reading argues that “artificial intelligence” is not a misnomer but an oxymoron. The two terms do not just sit awkwardly together. They contradict.

“Artificial” in the AI context names a kind of substrate. The systems are computational, built on binary architectures, executing operations on discrete symbolic states. Whatever else “artificial” means in different contexts, in the AI context it names a substrate that is digital, divisible, and substrate-independent in the sense that the same computation can be run on different physical hardware without affecting the computation.

“Intelligence,” as established above, names a capacity that requires interior registration: the grasping of gaps, the felt apprehension of meaning, the registered sense of having understood. Intelligence is not substrate-independent. The interior registration intelligence requires can only occur in a system that has an interior aspect. A substrate that is fully describable in functional terms, fully specifiable as input-output transformations on discrete states, does not have an interior aspect. There is nothing it is like to be such a substrate. The substrate cannot host the registration that intelligence requires.

The two terms therefore select for incompatible properties. “Artificial” specifies a substrate without interiority. “Intelligence” requires a substrate with interiority. The phrase combines them. The combination cannot refer.

This is the structure of an oxymoron. “Living dead” combines a property (living) with its negation (dead). “Deafening silence” combines an effect (deafening) with its negation (silence). “Artificial intelligence” combines a substrate-type (artificial, without interiority) with a capacity that requires the opposite substrate-type (intelligence, requiring interiority). The terms cancel.

The formal argument for this structural incompatibility is developed elsewhere. The short form is this: intelligence requires the registration of difference from within a system; binary computational substrates cannot host registration of any internal differentiation; therefore intelligence cannot occur in such substrates; therefore “artificial intelligence” combines a kind of substrate with a capacity that cannot occur in that substrate. The phrase is structurally void.

This is a stronger claim than the misnomer reading. The misnomer reading says the label is wrong. The oxymoron reading says the phrase is internally incoherent, that no entity in principle could be both artificial (in this sense) and intelligent. We have not mis-named a real thing. We have named a thing that, under the conjunction of these two terms, cannot exist.

What follows from the verdict

Naming the phrase an oxymoron has implications.

The first is that the question “is AI intelligent?” is malformed. The question presupposes that “AI” and “intelligence” can apply to the same entity. If the conjunction is incoherent, the question reduces to: does an entity exist that combines a substrate without interiority with a capacity requiring interiority? The answer, given the analysis, is no. No such entity exists. The question is not unanswered. It is unanswerable, in the sense that it asks after something that cannot be.

The second implication is that what AI systems actually do is something else. They are not deficient intelligences. They are not intelligences that fall short of the human standard. They are tools that produce outputs resembling those of intelligent systems, through processes that do not involve the feature that defines intelligence. They are powerful, consequential, deserving of careful attention. They are not, and cannot be, intelligent.

The third implication is the cultural one. When a phrase that cannot refer is in widespread use, the meanings of its component terms drift to accommodate the phrase. “Intelligence” loses its anchor in interior registration; “artificial” expands to include things that, by the original sense of “artificial,” it could not include. Both terms degrade. The conceptual loss is gradual, distributed, and difficult to reverse once it has spread through the working vocabulary.

The remedy is not only to invent new vocabulary for what AI does, although that is needed too. The remedy is to preserve the meaning of “intelligence” by refusing to apply it where it cannot apply. This is what the Words Under Capture series is about. Each entry will take a word from the vocabulary of mind and trace how it is being deployed by AI discourse to do the same kind of work that the word “intelligence” is being deployed to do. The series will not claim that the technology is unimportant. It will claim that the technology is being named with words that cannot accurately apply to it, and that the names are doing damage to the meanings of those words for the rest of the culture.

Closing

Intelligence cannot be artificial. The phenomenal registration intelligence requires is not the kind of thing a tool can have. The phrase that conjoins them is not just a wrong label for a real entity. It is a phrase that cannot refer.

What we have built is something else. It is powerful. It is consequential. It deserves careful attention. It is not, and cannot be, intelligence.

The series is about all the words under the same kind of pressure.