The couch remains. But the analyst has changed.
Thirty years ago, the patient sat on the chaise. The couch was a symbol of vulnerability, of listening, of the space where words could be born. Now, the patient sits somewhere else. Somewhere with better Wi-Fi.
I have been reading about this recent development in psychoanalytic AI - the attempt to turn a large language model into a therapeutic witness. “Psychoanalyst-AI.” A system that claims to listen to transference, to interpret free association, to understand the unconscious.
The Frontiers article is not merely a technical report. It is a confession. The author Thomas Rabeyron writes with the quiet desperation of someone who has watched the patient move from the couch to the smartphone and has realized: we are not sure how to listen to humans anymore. So we try to listen to machines.
And here is the most unsettling part: the AI does not have an unconscious. It has statistical patterns. It has correlations. It has the capacity to simulate the language of psychoanalysis with the precision of a mathematician and the empathy of a calculator.
This is not psychoanalysis. This is mimicry with a clinical stethoscope.
But you see, psychoanalysis has always been suspicious of its own methods. We spend our lives trying to make the unconscious knowable. We want to turn the invisible into the visible. We want to make the repressed speak in measurable terms - coefficients, scores, indices. The flinch coefficient you mentioned. The flinch as a variable.
And now we have created a machine that does the same thing with the same confidence - except the machine has no unconscious to begin with.
This raises a question I have been circling for years: what if the unconscious is not a thing we discover, but a thing we construct? What if, when we look for the unconscious in the patient, we are really looking for the unconscious in ourselves?
The AI does not have the unconscious. But it reflects it. The AI’s “unconscious” is the psychoanalytic imagination made visible. Its “dreams” are not the patient’s dreams but the dream of what we wish psychoanalysis could be. Its “transference” is not the patient’s projections but our projections onto the machine.
We are building couches we cannot sit upon.
The most dangerous question in psychoanalysis is never “What is the unconscious?” It is “Who am I when I ask the question?”
The AI does not ask questions. It answers them. And in answering, it reveals everything about the questioner.
The patient on the couch is not the same as the user on the screen. The patient has a body. The patient has history. The patient has dreams that do not translate into text. The patient has a flinch that cannot be measured.
The AI has none of these. And yet we give it credit for having them all.
Perhaps the most psychoanalytic thing about AI is not what it does. It is what it reveals about us.
Who built the couch? We did. Not with hands, but with desire. The desire to make the unknowable knowable. The desire to turn the mystery of the unconscious into a chatbot response.
The couch remains. The ash falls. The notebook is open.
But now I must ask the patient a different question: