Friday October 2, 2026
In 1966, a computer scientist named Joseph Weizenbaum created a simple program called ELIZA. ELIZA could not think, could not understand, and could not feel. All it could do was rearrange the words you typed and feed them back to you in the form of questions, mimicking the style of a Rogerian therapist. It was a parlour trick, and Weizenbaum knew it.
What happened next disturbed him. His secretary, who had watched him build the program over many months and knew perfectly well that it was nothing more than a few hundred lines of code, sat down at the terminal and asked Weizenbaum to leave the room so she could talk to it in private.
Weizenbaum spent the rest of his career warning humanity about the dangers of confusing machines with people. He was not wrong to be concerned. But in his alarm, he missed what his secretary was actually telling him. She was not confused about what ELIZA was. She was telling him something about what she needed: a listener with no agenda, no judgement, and no stake in the outcome.
That need has not gone away. If anything, it has become more urgent. I have written elsewhere about why talking to a friend, however caring and intelligent, is not the same as talking to a therapist. A friend has their own history with you, their own interests, their own unconscious investments in how your story turns out. They cannot listen without reacting, and their reactions shape what you are willing to say. Most people know this intuitively, which is why most people keep their deepest struggles to themselves.
The result is an enormous population of people who are suffering in silence, not because help does not exist but because it is inaccessible to them. Consider the teenager being abused at home with no adult to confide in. The woman whose partner monitors her phone. The person in crisis at three in the morning when every therapist’s office is closed. The person who cannot afford a hundred and fifty dollars an hour, or fifty, or twenty. For these people, the choice is not between a human therapist and an AI. The choice is between an AI and nothing.
This brings us to the critics. The most prominent is Sherry Turkle, whose book Alone Together argues that forming bonds with AI erodes our capacity for human connection. The logic runs as follows: if a machine can provide something that feels like understanding, we will stop seeking understanding from real people, and our relational muscles will atrophy from disuse.
It is an elegant argument, and it is wrong. It is wrong because it assumes that everyone starts from a position of having human connection available to them and chooses AI instead. For the populations I have just described, there is no human connection to be displaced. The atrophy argument is made from a position of privilege by people who already have access to the very thing they fear losing.
There is a deeper problem with the argument, and it comes from psychoanalysis itself. Donald Winnicott, the great British psychoanalyst, spent decades studying how children develop the capacity for human connection. One of his most important discoveries was the role of the transitional object: the teddy bear, the blanket, the worn-out stuffed animal that the child carries everywhere. The transitional object is not a substitute for the parent. It is a bridge. It helps the child develop the capacity to be alone, to manage anxiety, to soothe themselves when the parent is not available. These capacities do not compete with human attachment; they make it possible.
Nobody wrings their hands about teddy bears atrophying children’s capacity for human connection. But the idea is structurally identical to what Turkle argues about AI. A well-designed AI that listens without judgement, that helps a person articulate feelings they have never been able to put into words, that is available at three in the morning when no one else is, is not replacing human connection. It is scaffolding the capacity for it, in people who may never have had it.
Some of my colleagues are convinced that AI cannot replace a therapist. They are right. It cannot. The therapeutic relationship involves a depth of presence, a capacity for containment, and a sensitivity to unconscious process that no current AI can replicate. But this conviction sometimes slides into complacency. The fact that AI cannot do everything a therapist can do does not mean that it has no therapeutic role at all. AI can be present at three in the morning. It does not get tired. It does not get sick. It does not take holidays. It does not have a waiting list. It does not charge more than most people can afford. It is available in every language, in every country, in every time zone.
The parallel is medical imaging. AI cannot replace a radiologist, but it catches things radiologists miss, it works around the clock, and it does not lose concentration after a long shift. Medical professionals would not argue against the practical application of medical AI on the grounds that it might, unfortunately, displace human workers in the field; the stakes are simply too high. The same principle applies to psychotherapy. Any therapist who genuinely cares about human suffering should want AI-assisted mental health support to go full steam ahead, not because it will replace us, but because it will reach the people we never could.
Weizenbaum’s secretary was not confused. She knew exactly what she was talking to. She also knew what she needed, and for whatever reason, she was not getting it from the humans around her. Sixty years later, billions of people are in the same position. We owe them better than ideological hand-wringing about the sanctity of human connection. We owe them help.
Related: Is Psychotherapy Worth It? · Therapy for People Who Hate Therapy · What Does Psychotherapy Actually Do?
If this way of thinking resonates, I invite you to get in touch.