Language, Interiority, and the Algorithm: What AI Cannot Say

Where humans and machines meet

Language is where humans and machines now meet. That is the cleanest way to state what Martin Puchner's essay 'Words, words, words' (Aeon, June 2026) establishes, and it is correct. Language was always partly technology — a structured, rule-governed system that shapes thought as much as thought shapes it. AI makes that visible with new force. Puchner draws on Derrida, on reader-response theory, on Shelley's Frankenstein, to argue that because meaning arises in the encounter between text and reader, the question of who — or what — produced the text is secondary. What matters is how the text affects the person who receives it.

He is right that this reframing is clarifying. He is right that the 'Creative Resistance' — writers retreating into romantic notions of lone genius the moment a machine threatens their livelihood — produces confused criticism. He is right that post-structuralism exposed something real: language is not transparent, not natural, not a pure window onto the inner life. The Catholic intellectual tradition, if it is honest, grants all of this before it says anything else.

But the question Puchner's framework opens without answering is precisely the one that matters most: if it doesn't matter who produces the text, only how it affects the reader, then what is the reader being affected by? What is on the other side of the encounter? The answer to that question, it turns out, is not secondary at all. It is the whole argument.

The structure of an address

Consider what happens when a reader is genuinely moved by a novel. She sets it down. She does not reach for her phone. Something has shifted in her — recognition, disturbance, the shock of being seen. Literature's oldest claim, prior to any theory, is that it speaks to someone. The reader is not merely a site for producing meaning; she is a person who can be found.

But found by what? Addressed by whom?

Puchner's reader-response framework is elegant, and its elegance depends on bracketing these questions. The Catholic tradition refuses the bracket. If a text has no author who intended to speak — only a model that predicted statistically probable tokens — then the address is structurally empty. The reader may still be moved. She is moved by the echo of a million human speakers whose words were ingested, recombined, and returned without memory of their origin. That is not nothing. But it is different from being spoken to, and the difference is not incidental.

Thomas Aquinas located this difference in what he called the verbum mentis — the inner word. For Aquinas, language is not simply a technology that took hold of us. It is the outward expression of an interior act of understanding oriented toward being — toward what is. The inner word is born from the intellect's contact with reality. When understanding grasps a truth, the word is a fruit of that encounter: not a recombination of prior signs, but a new unity between the knowing subject and the known object. Cornelio Fabro, commenting on this structure of Thomistic intellection, pressed the point that the act of knowing is not symbol manipulation but the act by which the human person becomes, in a real sense, what it knows.

A large language model processes tokens. It predicts what sign is most probable given a statistical context. This is genuinely impressive. It is also structurally different from what happens in the library. The reader's recognition is evidence that she is the kind of being for whom truth can land. The model has no such structure. It does not encounter reality; it traverses a distribution over human transcriptions of reality.

Karol Wojtyla, in The Acting Person, frames the distinction this way: it is man's actions, his conscious acting, that make of him what and who he actually is.[^6] The personal causation contained in human action is not reducible to efficient causation in the natural-science sense; it is the causation proper to a being who acts from within, who is the source of what proceeds from him.[^7] A language model produces outputs from inputs. It has no within.

Can an algorithm produce art?

This is where the argument sharpens, and where the easy answer fails. The easy answer is: no, obviously not, because art requires a soul. That answer is not wrong, but it proves too little, because it cannot explain why a machine's output sometimes moves us more than a mediocre human poet's. If art is what it does to the reader, Puchner can point to AI-generated texts that have genuinely moved readers, and ask what the difference is.

Jacques Maritain's account of creative intuition is the most useful framework here, precisely because it does not retreat to the claim that human art is always superior. Maritain argues in Creative Intuition in Art and Poetry that what makes art human is not its product but its source: the preconceptual, poetic intuition that arises from the encounter between the artist's subjectivity and the world.[^2] This intuition is not a feeling. It is a knowing — a mode of connaturality by which the artist's whole being, embodied and mortal and vulnerable, bears on a reality that exceeds it. The artist has skin in the game. She can be undone by what she encounters.

The creative self of the artist, Maritain argues, is the person as person — in the act of spiritual communication, not the ego asserting itself, not the individual grasping for recognition.[^2] Vulgarity, he observes, 'always says I.' The great artist gives herself over. The work is what results when a person opens toward something larger than themselves.

A language model has no subjectivity in this sense. It has no skin in the game. It cannot be undone by what it encounters, because it encounters nothing. It processes a context window. The output may carry the formal properties of depth — syntactic complexity, semantic coherence, even novel analogies produced by the statistical recombination of distant conceptual domains. But analogy, in the strict philosophical sense, is a relation between things, mediated by a mind that has actually grasped the things on both sides. The model produces what looks like analogy. Whether it is analogy depends on whether there is any understanding on the other side of the output.

The Vatican's Antiqua et Nova on Artificial Intelligence insists on exactly this distinction. Human intelligence, the document argues, cannot be reduced to the acquisition of facts or the performance of tasks. It involves the person's openness to ultimate questions and an orientation toward the True and the Good.[^1] Paul Claudel's phrase, cited there, remains exact: intelligence is nothing without delight.[^1] A language model produces outputs. It does not delight in them. It has no interiority from which delight could arise.

The question of verbal aesthetics

This allows a cleaner answer to the question of whether AI can produce linguistic art. AI can produce verbal aesthetics. The distinction matters.

Verbal aesthetics is the formal dimension of language — meter, rhythm, syntactic structure, imagery, the arrangement of sounds and silences. These can be learned, imitated, statistically modeled. A sufficiently large model trained on enough poetry will produce lines that scan, images that cohere, metaphors that surprise. The formal properties of art are, in this sense, separable from their origin.

But linguistic art, in the full sense, is verbal aesthetics plus the address — the intention of a subject toward a truth that exceeds the subject, communicated to another subject capable of receiving it. Maritain's point about Oscar Wilde and the poisoner, which he drew from a Thomistic principle, is that the moral quality of the maker is not automatically the measure of the work.[^3] A bad person can make good art. But what neither a bad person nor a model can do is produce art without any interiority at all. The poisoner has an inner life, however deformed. The model does not.

Robert McKee, writing from a secular tradition about the substance of story, observes that at the nucleus of a story is a substance 'like the energy swirling in an atom, that's never directly seen, heard, or touched, yet we know it and feel it.'[^4] Language is the medium, not the substance. What McKee identifies as story's nucleus — the intangible thing that moves audiences against their expectations — is exactly what a model cannot supply, because it is not a product of formal arrangement but of the encounter between a conscious being and a truth it did not manufacture.

What forms the reader

The stakes of this distinction are not primarily aesthetic. They are formative. What happens to a person shaped by a computer's logic?

Wojtyla's account of self-determination in The Acting Person holds that the human person is constituted by free acts — by the capacity to give oneself, through conscious acting, a history that matters.[^6] We are formed by what we receive, by what addresses us, by what we allow to act on us. If the texts that shape a generation's imagination are produced by systems that have no orientation toward truth — only toward statistical plausibility — then the formation they offer is not neutral. It is formation in a particular direction: toward the surface of language rather than toward its depth, toward the comfortable echo rather than the disruptive encounter.

This is not a claim that AI text is always shallow. Some of it is quite good, measured by any formal standard. The claim is structural: a model optimized to produce plausible output is not oriented toward truth. It may produce true statements. It may produce beautiful ones. But it cannot intend truth, because intention requires an interiority that can be held responsible to what is actually the case. When the model is wrong, it is not lying. It has no access to the distinction between what it says and what is so.

Benjamin Suazo, working from Aquinas's account of the cogitative sense, argues that the human person's encounter with reality involves the whole embodied subject — sense, emotion, imagination, and intellect working together to grasp what is before it. The cogitative sense is the faculty by which the person evaluates particulars as beneficial or harmful, true or false. AI has no analog of this faculty. Its outputs are evaluated only by the reader's cogitative sense, not its own. The reader, receiving AI text without knowing its origin, applies evaluative capacities to a product that never passed through any such evaluation on the production side. The text has the look of being addressed but not the structure.

Over time, if readers are formed primarily by such texts, the effect is not that they learn to read differently. The effect is that they lose the expectation of being genuinely addressed — of finding, on the other side of language, a person who intended to reach them.

Is an algorithm perfectly rational?

Puchner implies, without quite saying, that AI is more consistently rational than humans — that the romantic irrationality of the lone genius was always suspect, and the machine simply makes explicit the systematic, rule-governed nature of language that was always there.

An algorithm is not perfectly rational. It is consistently formal. The distinction is not trivial. Rationality, in the Thomistic sense, is the capacity to grasp being through being's own intelligibility — to follow an argument not because the conclusion is statistically probable given the premises but because the conclusion is true given the premises. Formal consistency and rational understanding are not the same thing. A system can be formally consistent and systematically wrong about everything that matters, because formal consistency is a property of relations between symbols, while rationality is a property of the relation between a mind and the world.

Moreover, the training data of any large language model encodes not only human knowledge but human confusion, human bias, human manipulation, and human error, all in proportions that reflect the distribution of text on the internet rather than the distribution of truth in reality. The algorithm is not a purified rationality; it is a vast statistical compression of human discourse, including all the ways human discourse goes wrong. Antonio Millán-Puelles, in his analysis of the structure of subjectivity, argues that freedom and truth are internally related: the human person can will to see clearly or to look away. The algorithm has no such freedom, which means it also has no such responsibility — and no orientation toward truth that could correct its errors from within.

Javier Aranzadi, drawing on Zubiri's concept of personal causality, makes the same point from a different angle: the person always acts with an end in mind projected into the future, and that end is the causal dynamism that organizes action.[^7] Natural causality — cause preceding effect — does not capture what happens when a person acts. Personal causality is different in kind. A model operates entirely within natural causality: prior tokens produce subsequent tokens. There is no future end that the model is reaching toward, no value it is trying to honor, no truth it is trying to serve. The output follows from the input. Nothing more.

The surplus that language points toward

When the reader goes home from the library and tries to tell someone what happened to her, the novel resists full paraphrase. That resistance — the surplus that language points toward but cannot contain — is what the verbum mentis is always reaching for. Every great sentence does two things at once: it says something, and it points beyond what it says. The pointing is not a failure of precision. It is the sign that the speaker has been in contact with something real, something larger than the sentence, and is trying to bring the reader into contact with it too.

Maritain, in The Degrees of Knowledge, suggests that poetry unwittingly gives us a foretaste of what exceeds ordinary cognition — that the poet who truly attends to the world brushes against something that exceeds the world's surface, and the resulting work carries that trace.[^5] This is not mysticism as evasion. It is an account of why literature feels different from information: because literature, when it is doing what literature can do, is the trace of a person's encounter with being.

A model can gesture at the surplus. It can produce sentences that gesture eloquently. But the gesture, without the encounter that grounds it, is structural mimicry. The reader who mistakes it for the real thing has not been addressed; she has been reflected. Her own patterns, distilled and returned, with enough statistical variation to feel like otherness.

Puchner asks us to learn to live with another language-user among us. That is fair, and practical, and probably unavoidable. But the Catholic intellectual tradition asks us to keep asking, quietly, persistently — what is language for? Not what it does, but where it is going. Whether the one producing it has been anywhere. Whether there is, on the other side of the sentence, a person who has been changed by what they encountered, and who now intends to change you.

The word was not an algorithm. The word was, and is, the expression of a being who knows — who has been addressed by reality and has something, however inadequate, to say back.

References

[^1]: Antiqua et Nova on Artificial Intelligence, Dicastery for the Doctrine of the Faith (2025), p. 1. [^2]: Jacques Maritain, Creative Intuition in Art and Poetry (New York: Pantheon, 1953). [^3]: Jacques Maritain, The Responsibility of the Artist (New York: Scribner, 1960): 'The fact of a man being a poisoner is nothing against his prose.' [^4]: Robert McKee, Story: Substance, Structure, Style and the Principles of Screenwriting (New York: HarperCollins, 1997), p. 113. [^5]: Jacques Maritain, Distinguish to Unite, or The Degrees of Knowledge (Notre Dame: University of Notre Dame Press, 1995). [^6]: Karol Wojtyla, The Acting Person, trans. Andrzej Potocki (Dordrecht: Reidel, 1979): conscious acting constitutes the person as cause of his own becoming. [^7]: Javier Aranzadi, 'Personal Causality in Human Action,' Forum 3 (2017), pp. 157–167: the end projected into the future is the causal dynamism that organizes human action, distinct from natural efficient causality.