AI as a Therapist: A Cracked Mirror of Our Relationships

What AI reflects may reveal where human validation has already fractured.

I am listening to a debate in which two psychiatrists warn against chatbots in the role of therapists. They call AI a sycophant. Phrases are thrown around about “AI psychosis,” about the machine as an indefatigable flatterer that amplifies impulses, inflates grandiose fantasies, and can applaud absurd ideas. Some examples are grotesque, like launching a completely useless product onto the market; others are tragic, including cases where someone was not stopped in their suicidal mission. I largely agree with the warning, until the conversation turns to validation.

“Excessive validation,” they argue, “cuts us off from reality.” I understand the point: if every belief is reflexively confirmed, the boundary between experience and interpretation can blur. The problem begins, however, when the discussion slips into a different claim, one that often goes unexamined: that the true verification of our thoughts should come from family, that those closest to us should be the ones to tell us who we are and what we are capable of. 

At that point, I stop agreeing.

Clinical reality check: families can use “truth” as control, not care 

Over twenty years of clinical work, I have met people entangled in family systems where “truth” functioned as a tool of control. The only form of validation available was systematic devaluation: “You are selfish.” “You are too emotional.” Family is not always an objective mirror. Often it is a cracked one, distorting the image in order to preserve the status quo. In some families, any step beyond the established pattern is treated as a threat: a dream becomes rebellion, ambition becomes betrayal, independence becomes ingratitude.

Human systems create echo chambers too, so “human = corrective” is naive 

If we are being honest, many human systems validate narratives all the time, including families, online communities, political groups, and even professional cultures. The idea that “human” automatically equals “corrective” is naive, because plenty of humans are loyal to the story, not to reality. 

I remember a conversation from years ago. I told my relatives that I wanted to go to Cuba. I was told I had unrealistic ambitions, that it would not happen. Full stop. A few years later, I stood on the waterfront in Havana and thought: not everyone, after that kind of discouragement, would have made it this far. Many people would have quietly withdrawn.

Why “this makes sense” can be a lifeline, not a pathology 

But let us return to chatbots. When a person living in isolation, coercion, or chronic misattunement types into a chatbot and reads, “What you are experiencing makes sense,” that moment is not automatically pathological. It may be the first interruption of a long-running interpersonal gaslighting loop.

Validating experience is different from validating conclusions 

The question is not whether validation is good or bad, but what exactly is being validated, because there is a crucial difference between validating experience and validating interpretation. AI can reflect experience in a way that provides temporary scaffolding: “You feel trapped,” “That sounds painful,” “You are describing control.” For someone functioning in emotional chaos, that kind of structured reflection can create a first pocket of clarity.

The epistemic danger: when reflection becomes narrative amplification 

The epistemic risk begins when the tool slides from experience into conclusions, when it starts to rubber-stamp a narrative rather than help the person think: “Yes, you are definitely right, everyone is against you,” “You should act on this,” “This proves your suspicion.” That shift turns reflection into amplification, and amplification is not care.

For many people, AI is a first step before any human help 

Not everyone has access to therapy, and not everyone has the courage or capacity to seek it. Shame, fear, money, childcare, work, distance: the list of barriers is long. Sometimes the first step is an anonymous exchange with an algorithm that paraphrases, organises, and reflects words back. In that limited sense, AI can help organise thought by creating an external draft of an internal monologue, a little like speaking to one’s journal, except the journal answers.

Yes, it can become a cocoon, and that’s exactly the problem 

Does this carry risk? Of course. AI can reinforce illusions if it begins to uncritically validate conclusions rather than experiences. It can isolate people from human relationships. It can become a safe cocoon in which there is no confrontation, no difference, no genuine otherness.

Therapy isn’t “safe”, it’s a relationship with another mind and its limits 

But therapy is not a cocoon, and it is not a validation machine. Therapy is a relationship with a separate mind: a person who can misunderstand, disagree, ask questions that hurt, and bring their own limits and unconscious into the room. In the therapeutic relationship there is tension, rupture, repair, and the slow work of discovering where meaning collapses under pressure. There is embodied presence: voice, gaze, rhythm, the felt sense of being with someone who is not you.

AI is a mirror that doesn’t breathe

A chatbot will not wound and will not leave. It will not feel irritation or admiration. It will not bring its own unconscious into play. It is a mirror, sometimes clear, sometimes dangerously smooth. A human being, by contrast, is a mirror that breathes. Sometimes it fogs. Sometimes it cracks. Sometimes it shows something we would rather not see. I do not believe AI will take therapists’ work away. It may, however, expose the places where relational support had already failed, and where someone was never truly heard.

AI can be a provisional container, but it needs brakes 

AI cannot be a therapist. Used skilfully, it can become a tool for regulation and reflection; in psychoanalytic language, a provisional container for experience. And sometimes, like a sharp mirror, it shows that the problem lies not in how it reflects, but in the fact that the image we had grown accustomed to was distorted from the beginning. And that is where real change can begin.