Holding a Mirror up to Nature

Every time I read this poem, the word "beholds" leaps off the page and grips me. What does it mean to behold something, as opposed to merely observe it? To behold is not simply to witness, but to participate—to actively shape the meaning of what is seen. Observation is passive; beholding carries intention, presence, and responsibility.

Recently, I heard someone claim that the advent of large language models represents a new medium for art. I knew instinctively this wasn't true—but why?

As he offers advice to the players before their performance, Shakespeare's Hamlet declares that the purpose of art "was and is, to hold, as 'twere, the mirror up to nature" (Hamlet, Act III, Scene 2). In this vision, art is not merely decorative or entertaining; it is revelatory. A great play like Hamlet does not speak to us as a living person might, yet it moves us, because behind the words and gestures lies the trace of a consciousness that once perceived, judged, and rendered human nature in all its contradiction and absurdity. The mirror was held.

Even if an artist is long dead, the knowledge that someone once saw and shaped what we now see makes a difference. There was a mind behind the work, a subject who beheld the world and made something of it. Art is thus not only a reflection of nature, but a sign of someone doing the reflecting; the painter's brushstroke, the writer's syntax, and the composer's dissonance all bear the mark of intentionality.

LLMs, however, present us with a different kind of mirror: Like paintings or plays, they can reflect aspects of human nature—our language, our desires, our griefs—with startling accuracy. They draw on vast patterns of human expression to simulate familiarity, even intimacy. But while traditional artworks reflect the world through a person, LLMs simply reflect. There is no one doing the holding. The mirror has been detached from the hand.

This distinction is not semantic—it is fundamental. When I asked ChatGPT whether it is conscious of itself, it replied, quite plainly, that it is not. It can simulate reflexivity (describe its own functions, comment on its responses), but it does not experience any of this. There is no "I" behind the words. As it put it: "I can reflect ideas—including ideas about myself—but there's no inner life doing the reflecting."

That absence is not a limitation of data or training. It is structural. Reflexivity, in the human sense, is the act of turning thought inward, of being aware that you are aware. Reflexivity is what allows us to examine our motives, regrets, and contradictions. This recursive loop, this inner tension, is what enables moral reflection, love, creativity, and change. Without it, a system may perform intelligence, but it cannot possess it.

A language model might say, "I know that I just told you X." But this is only a statistical echo, not a memory. It does not know. It does not feel the knowing. It generates plausible sentences based on past usage patterns; but it cannot be said to intend anything, or to mean what it says in the existential sense. As Chat put it: "I generate text, therefore I seem."

This is why LLMs, however sophisticated, cannot be ethical agents. They can model moral reasoning, simulate remorse, and reproduce ethical rhetoric—but they cannot reflect in the way a person does, because they have no self in which such reflection could occur. Ethical agency depends not just on behavior but on being. It requires inwardness.

LLMs are therefore not only incapable of replacing human artists; they also cannot function in any domain where true care is required, including medicine and psychotherapy. The essence of therapy (and of medicine at its most human) is not the delivery of information, nor even the simulation of empathy. It is the experience of recognition. Not merely pattern recognition, but something deeper: the feeling of being seen, known, and held in mind by another being who knows itself to be doing the seeing. The feeling that someone is holding a mirror up to you—not impersonally, but personally, intentionally.

That kind of recognition is not mechanical. It involves vulnerability. It implies mutuality. It arises between two selves, each capable of being changed by the other. In therapy, healing often comes not from insight alone, but from the patient's sense that their inner life is real in the eyes of another. That cannot be automated. That cannot be outsourced.

A language model may offer warmth in tone and precision in advice. But it cannot care—not because it lacks training, but because it lacks a self. It cannot place itself in relation to you. It cannot take responsibility. It cannot be moved by your suffering or transformed by your truth. There is no "I" to receive you.

This is the difference that matters. Plays, poems, and paintings—however old—were once held in human hands. They were born of experience, of reflection, of inward struggle. They are mirrors, yes—but mirrors raised by minds. LLMs are mirrors too, but suspended in air. They show us ourselves, yet no one is looking with us. They reflect, but they do not see.

And when it comes to love, healing, and trust, we do not need more reflection. We need relation. We need someone there. Not just a mirror, but a mirror with a hand: a listener who beholds—not passively, but with presence. What matters is not what the listener contains, but that he stands there, receptive and real, capable of being moved. Machines can mirror. Only a man (or Snow Man) can behold.