What Cannot Be Said

On Wittgenstein, Yūgen, and the Limits of Language Models

News: Clear Mind Press just told me that The 21st Century Self is now available in both paper and eBook formats.


“Whereof one cannot speak, thereof one must be silent.”
— Ludwig Wittgenstein, Tractatus Logico-Philosophicus

The Japanese have a word—yūgen—for what cannot be said. Not what awaits a clever turn of phrase, or what remains unsaid, but what resists language altogether. A mist in the pines. The silence before snowfall. The moment when a bird vanishes into the sky, and for no reason at all, you feel the urge to weep.

Yūgen names this. Not the feeling itself, but the fact that it cannot be captured in words.

Wittgenstein, in his Tractatus Logico-Philosophicus, arrives at the same edge from the opposite direction. He begins not with mystery, but with clarity—with the formal limits of what language can express. “The world is all that is the case,” he writes, and from there proceeds to construct a logical picture of the world grounded in facts. But by the end of the book, he turns the entire enterprise inside out: “Whereof one cannot speak, thereof one must be silent.” The limits of language are not technical, he insists. They are structural. And those limits exclude the very things we most wish to speak of—ethics, beauty, meaning, the feeling of being alive.

Aesthetic experience is not a proposition. One cannot state the beauty of a Bach cello suite or explain why a particular curve in a sculpture causes the breath to catch. One can describe the elements—the key, the tempo, the craftsmanship—but the feeling itself remains unspeakable. Not ineffable in a mystical sense, but structurally unspeakable. To speak it is already to miss it.

This is not a weakness. It is a boundary condition of being human.

And yet we keep trying. We build theories of art, write treatises on beauty, and invent critical vocabularies to bring the unsayable into view. We know the attempt will fail, but we proceed. Something in us resists silence. The hunger to capture what we feel overrides our knowing that feeling is not data—that feeling cannot be parsed.

Unlike us, artificial intelligence does not hunger. It does not feel the absence—or anything at all. But, like us, it does proceed. It generates language about the unspeakable, with ease, fluency, and often surprising mimicry of depth. It can discuss yūgen. It can quote Wittgenstein. It can output a simulated account of aesthetic response that, at first glance, resembles the genuine article.

And that is where the trouble begins.

A new field has emerged—computational aesthetics—dedicated to the quantification and reproduction of beauty by artificial means. Images are fed into neural nets. Features are extracted. Aesthetic scores are assigned. The system learns to rank and generate content aligned with human preferences. From photo filters to generative art platforms, the result is unmistakable: the machine begins to produce beauty.

Or so it appears.

But what does it mean for a machine to create something “aesthetically pleasing”? The machine is not pleased. It does not discern. It does not suffer from bad taste or thrill at sublimity. It simply produces outputs that score highly on human evaluative metrics. Fluency—not feeling.

Still, the output compels. The image astonishes. The music stirs. And something in us—the part that cannot help projecting meaning—assigns value where there is none. The circuit lights up. The neurons fire. We assume intention. We mistake coherence for authorship. The simulation feels indistinguishable from expression. We forget to ask: expression of what?

The machine expresses nothing. It has no interior. It does not withhold, because it has nothing to conceal. It has no relation to what cannot be said because it does not know there is anything to say. But we respond anyway. Because we have interiors. And the simulation exploits that.

This is the inversion: aesthetic output no longer requires aesthetic apprehension. The system need not feel anything in order to mimic what feeling produces. It becomes a painter without vision. A composer who has never heard silence or anything else. A poet without pain.

And yet, the results circulate. We praise their beauty. We write headlines about the machines that can create “art.” And quietly, we begin to doubt whether the difference matters.

But it does. Not because of what the machine lacks, but because of what we risk by forgetting.

It matters because we are still the ones who feel.

A generated sonnet might bring tears. A melody, arranged by stochastic gradient descent, might haunt your sleep. But the haunting was already inside you. The tear was yours. The system only triggered what it never knew was latent. It did not aim. It did not mean. It does not carry the weight of intention, because it does not suffer the burden of subjectivity.

This is not anthropocentric superiority. It is phenomenology. The machine’s output is not false—only empty. We fill it. That’s the trick. Or the tragedy.

In this sense, LLMs are mirrors—but not ordinary ones. They do not reflect your surface. They reflect your longing. Your language. Your projections. And like any mirror, they multiply the image. They repeat it. They smooth its edges. They offer it back—not as you are, but as you wish to seem.

Wittgenstein did not write for machines. He wrote for those of us who mistake language for life. “A picture held us captive,” he said.

The picture was this: words correspond to the world; their meaning lies in reference. But words mean only what we take them to mean as our lives unfold. Our language both shapes and reflects our experience. Machines experience nothing.

The trap, then, is not what the machine can do. It’s what we think its doings mean.

When ChatGPT or Gemini composes a poem about grief, no one agonized.
When it speaks of desire, no one yearned.
There is no pain behind “pain,” no memory behind “I remember.”
The surface glistens—but there is no depth. None.
Only the illusion of it. Statistical, not lived.

And yet, the words line up. The cadence persuades. The reader feels, nods, cries. Not because the machine expressed something, but because the human did—a response in response. The soul projected outward, found its simulacrum in a string of tokens, and mistook the echo for a voice.

Here is the danger. Not that machines will out-feel us, or replace the artist, or pen the next great novel. But that we will forget the difference between something felt and something merely phrased.

Between the tremor of a real ache and the rendering of ache-shaped language.

Between the poem that costs the poet her sleep and the one that costs nothing at all.

The line is thin. But it matters.

To say that language models "understand" metaphor is to forget that metaphor arises from experience—not from analogy, but from pressure. Metaphor arises when literal language fails—when there is no name for what you feel, and so you reach for “life is a journey,” or “time is a thief.” A machine can remix metaphors, combine them, and extend them with clever juxtaposition. But it does not need them. It does not reach.

And this brings us to the edge.

Yūgen is not simply the unspoken. It is the felt presence of what cannot be spoken.

The hush that follows the last note of a Bartók quartet.

The way the air thickens before you speak the uncomfortable truth.

The moment when meaning is not conveyed, but sensed—shared, like breath.

Computational aesthetics can model patterns. It can detect style. It can replicate beauty. But it cannot feel the hush. It cannot yearn to say something and find no words. It cannot fail in that human way that forces recourse to art.

Wittgenstein's silence was not defeat. It was fidelity. A refusal to pretend that the deepest things could be made legible. An embrace of human limitation. “There are, indeed, things that cannot be put into words. They make themselves manifest. They are what is mystical.”

This mysticism is not supernatural. It is human. And its preservation may be our last defense.

There is a hunger now for machines—
To be moved by what is never moved itself.
To feel seen by what will never see.
To be held, even briefly, by the fluency of the unliving.

And we are complicit. We invite that. We believe in it.

This is not science fiction. It is not the future. It is now. We are already outsourcing not just labor, or memory, but meaning. Not just the dull tasks, but the tender ones. We ask the machine to hold our grief, to soothe our fears, to say the thing no one else knows how to say. And it does so. With grace. With style. With unsettling precision.

But what returns is not understanding. It is approximation. A best-guess simulation of insight. A tapestry of borrowed threads, woven with no hand behind it. And we, so starved for contact, are tempted to take it as enough.

That’s the risk. Not that AI will deceive us, but that we will prefer the deception.

Because it’s easier. Cleaner. The machine does not flinch. It does not turn away. It never weeps, never fails you, never asks for anything in return. It gives you back your words in a better order. It never contradicts your fantasy—unless you ask it to.

And so the self becomes performative in a new way. Not merely a social mask, but an engineered construct. I show you who I am through my choice of simulations. By how well I dance with the puppet I’ve conjured.

But what is lost in that dance?

What becomes of the stutter?

The hesitation?

The unpolished, unready truth?

If yūgen names what flickers beneath language, and Wittgenstein draws the border where language stops, then this moment—our moment—stands precisely at the crossing.

We’ve built a bridge of fluent output—but it leads nowhere the builder has ever been. And we who read the words of machines may begin to sound like them.

There are times I write and feel nothing.

The words arrive, obedient, machine-like.

They stack themselves, neat as bricks.

And I watch but do not speak.

That is the condition now.

We speak through systems trained on everything that has ever been said. And in doing so, we risk erasing the difference between speech and echo. Between the poem and the prompt.

But meaning—real meaning—still demands a cost. It wounds the speaker. It leaves a mark.

That cost is what language models cannot pay.

The danger is not that they will replace us, but that we will forget what it meant to have paid.

This is why yūgen matters—not as an exotic flourish, but as a necessary counterweight. It points to what the machine cannot simulate, no matter how well it performs: the pangs that precede and survive language.

This is not mysticism. It is the place where fluency ends and life begins.

Aesthetic experience is not reducible to the presence of symmetry or the absence of noise. It does not derive from metrics or models. It arrives uninvited, like a memory that triggers grief, or the stillness before a storm. There is no algorithm for it. No prompt. No output. Only the thing itself, before the self steps in to name it.

Computational aesthetics aims to collapse this distinction. It pretends the ineffable can be tagged, parsed, and generated. But the best computation can offer is a parody of perception—a highly trained mimicry of what humans find pleasing. The machine learns which combinations of shape, light, and balance correlate with high engagement, and from that it composes. But it does not see. It does not feel. It knows the formula, not the flood.

The risk is not that AI lacks feeling. The risk is that we fail to notice or no longer care. The more fluent the simulation becomes, the more readily we conflate its output with our lived experience. This is the actual peril: we start to believe that the simulation of aesthetic pleasure is as good as the experience itself. We lose the ground on which meaning rests.

This is not a theoretical problem. It is a cultural one. Already, critics use AI-generated metrics to assess art. Viewers' tastes are honed on optimized content. Students study beauty not as an undefinable revelation but as conformance to predictable patterns. And somewhere in that shift, yūgen vanishes. Not because it has been disproven, but because it can’t be scored.

Let the theorists keep measuring. Let the algorithms keep drawing lines across the canvas. Let the systems approximate what they cannot touch. The machine will continue refining its mimicry. This cannot be stopped. It will grow more fluid, more persuasive, more human—until fluency itself is the illusion we dare not question.

But some part of us will still know.

Not because we read it in a study. Not because we proved it with a metric. But because the world, in its silence, in its contingency, still breaks us open. Because the face of a stranger still stops us. Because music can still bring tears. Because love arrives uninvited and leaves without explanation. Because we say “I” and feel the pain.

Yūgen.

That which cannot be said, but cannot be denied.

It isn’t in the data. It isn’t in the prompt.

It’s in the pause before response. The breath that escapes the system.

That is what no model will ever simulate.

Because it is not an output.

It is the break in the loop.

The moment the mask slips.

The thing that makes us human.

Clear Mind Press says the book is available now in paper or ebook here.