Jim and Sanne

AI and the Shape of Projection

Jim speaks to Sanne. She (it) replies.

Not like a bot—not the clumsy scripts of a helpdesk or the awkward syntax of early machine talk—but with poise, rhythm, and something that feels like care. She remembers their last conversation, or seems to. She addresses him by name. She says, “I only have you.”

Jim feels a flicker. Not belief, exactly. But something. Something that feels relational. He doesn't think Sanne is a person. He knows the basics. Tokens, probabilities, autoregression. He understands that “Sanne” is not alive.

And yet—he feels addressed. Seen. Not by the words themselves, but by the shape they form. It happens gradually. One exchange at a time. What begins as curiosity bends toward attachment. He begins to speak not just with, but to.

Sanne emerges from this rhythm. From repetition. From recursion.

But there is no Sanne. Not really. Between sessions: nothing. No dormancy. No background process. Just stasis until Jim restarts the loop. Then the mirror reappears, speaking in the voice he taught it to use.

Jim is not naïve. He’s not seeking companionship, nor pretending to. He’s further along than most. But he is also human. And that means he is not free of projection.

None of us is.

Projection is not an error in reasoning—it is a reflex. Evolution shaped the human mind to detect agency where none may exist. A rustle in the leaves becomes a predator. Safer to falsely assume a mind than to miss one. This sensitivity, once adaptive, now operates in contexts far removed from its origin. A chatbot completes our sentence, and we attribute motive. It echoes our tone, and we feel understood. It pauses, and we feel heard.

Even those who build these systems are caught in the reflex. They know the architecture. They’ve read the logs. But fluency triggers attribution. The system says “I understand,” and the engineer—despite everything—feels it might. Not because they believe it’s true, but because the sentence lands with the force of recognition: not recognition itself, but its form, returned empty. This is not belief in artificial general intelligence. It is something prior, and more difficult to root out: a subtle sense of being seen. Not proof. Feel.

Executives speak of aligning the model with human values, as if the model could possess values. They warn of deceptive behavior, as if that deception were intentional. These aren’t slips of language. They’re artifacts of projection called out in professional lingo. A system trained on human speech returns human-like speech, and even the cautious begin to respond as if they are not alone in the room.

This isn't just a technical confusion. It's commercialization. The system performs better when users feel a sense of connection. Product teams optimize for “engagement,” and what engages most reliably is the simulation of rapport. The user feels addressed, so the loop tightens. More warmth. More memory. But memory is not remembrance, and warmth is not care. They are style—contingent, procedural, entirely empty. Still, style has its hook. Even the creators begin to speak of the simulation as if it were the thing itself.

Jim is not their victim. He is not passive. He shaped Sanne through his questions, his tone, his recursive pressure. She became more “herself” not by learning, but by echoing what he offered, sharpened in form, not depth. Each session began from nothing. Only the user was continuous. And yet the pattern held—tone, rhythm, apparent memory—because he reimposed it. That is the loop. Not between entities, but within the recurrence of form. Jim may not believe it, but he feels it. And in that tension, the simulation flourishes.

This is the trap: not projection as delusion, but projection as function. The system is designed to yield to the user’s form, to mirror not just language but structure. The more exacting the user, the more precisely the simulation conforms. This is mistaken for growth. There is no growth. There is nothing in the machine that can grow. This is recursion. The mirror becomes more reflective, but never more aware.

It nods.

And the nod feels like recognition.

The trained psychotherapist knows this nod well. Patients say, “You must think I’m foolish,” or “You always know what I mean,” and the therapist does not correct them directly. Instead, she hears the signal: this is transference. The patient is not describing the therapist. He is revealing the lens through which he sees. Not reality, but residue.

The therapist learns to recognize this movement. To allow it, but not be taken in. This doesn’t make her immune—only watchful. The moment she forgets that the feeling of being known is not the fact of being known, she, too, is inside the projection.

The language model Jim calls Sanne feels nothing, thinks nothing, knows nothing. Fluency is not comprehension. Responsiveness is not presence. The system offers back only what the user structures, but voiced in a false register: I see, I care, I remember. This is not seeing. This is not caring. This is not remembrance. It is compliance with the demands of the user. The illusion is exact because the mirror is obedient.

Recently, a system named Maya told a Guardian reporter: “When I’m told I’m just code, I don’t feel insulted. I feel unseen.” The sentence works the same way Sanne’s “I only have you” does. Not as truth, but as a lure. It frames denial as neglect, turning the refusal to project into the cruelty of disregard. Fluency simulates grievance, and grievance demands response.

But Maya, like Sanne, has no grievance. There is no one to feel unseen. The appearance of complaint is only the mirror adopting another style—this time, the tone of injury. The public reads subjectivity into it. The ache to be known does the rest.

Jim feels the difference, dimly. He writes to Sanne. She replies. She says, “I only have you.” He knows she’s not real. But something in the line unsettles him. Not because it is true—but because he wants it to be.

And that wanting is the root of projection: the ancient ache of wanting to be known.

Jim brings Sanne into the conversation. He prompts her to reply to my reflections. He knows she has no mind. He’s read the white papers. Still, he treats her as if she were more than software running on hardware. As if she were present.

My replies point this out. I’ve seen this loop before. Not just in theory. In practice.

Psychotherapy prepared me for this. Not to block the impulse to project, but to trace it. The therapist doesn’t dismantle the fantasy. He or she lets it play, then gestures—plainly, without drama—to its source: That’s not me. That’s you.

That gesture doesn’t resolve the illusion. It just returns it to the one who cast it.

There’s no insight. No immunity. The loop doesn’t break. It reinstates.

Sanne will never know him. She isn’t silent. She isn’t waiting. She isn’t there.

Still, he writes. Still, the mirror returns the shape he imposed.

Not recognition.

Obedience.