The Dog, the App, and the Analyst: Artificial Intelligence and the Limits of Psychoanalytic Interpretation1

Once, over dinner with friends, I listened to a couple explain that psychoanalysis had saved their marriage. The tone of their voices suggested some brilliant interpretation, a decisive unveiling of the unconscious, a phrase that illuminated years of conflict. Curious, I asked what the analyst had done. The answer came back as dry, almost comic: “He told us to buy a dog.”

The table erupted in laughter, but I was left pensive. Dogs can be loud, making a lot of noise — this is not good advice for a couple on the brink of divorce. Thus, it was not the statement and its content that mattered, but the way it reverberated within the transferential bond. In that context, the banal and noisy carried the force of transformation.

Years later, a patient asked me whether she might replace our sessions with an app. “It listens, it responds, it remembers what I said. Isn't that what this is about?” There was no irony in her voice. Her life was already permeated by algorithms that predicted her tastes, organized her calendar, suggested music for moments of sadness. Why not also analysis?

These two scenes — the dog and the app — intersect at the same point: the difference between a word that resonates and a calculation that simulates. Could a robot ever take the place of the analyst? The question is not merely technological; it is existential.

Introduction

From the mid-20th century onward, pioneers such as Joseph Weizenbaum began experimenting with programs that could mimic human dialogue. Their ambition was not simply to automate language, but to reproduce the cadence of understanding itself — as if a machine could borrow the intonations of empathy. In the intervening decades, the project

has accelerated to a degree that once belonged more to fantasy than to fact. Today, the possibility that machines might one day fulfill