When a patient walked into her office and showed her his phone screen, clinical psychologist Sarah Dargouth saw something that made her stomach drop. A chatbot had just done what she had been carefully working toward for weeks: it told her patient to break up with his girlfriend. The relationship ended shortly after. Dargouth sat in her professional composure while part of her fumed at being upstaged by an algorithm.
That wasn't the first time AI had invaded her practice. Another patient came in having already received coaching from ChatGPT on how to repair a marital conflict. The bot had analyzed the relational breakdown and proposed solutions. When Dargouth read the suggestions, an uncomfortable thought surfaced: "WTF. This thing is actually good." The patient had used the advice. It worked. For an instant, Dargouth felt small. She might have spent an entire session wrestling with one of those ideas.
Dargouth works at Massachusetts General Hospital and teaches psychiatry at Harvard Medical School. She spends her days listening to people describe their most tangled emotional lives. Now she finds herself sitting across from patients whose internal monologues have been pre-processed by artificial intelligence. The boundary between her patient's authentic voice and the AI's has become murky. Sometimes she doesn't know whose emotion she's hearing anymore.
Her clinical instinct was to push back. She warned patients about the documented risks: AI can amplify anxiety, deliver false information, deepen isolation, and in rare cases contribute to delusional thinking or suicide ideation. She suggested alternatives, like journaling, so they could bring raw material into therapy rather than polished chatbot insights.
Then she got home one Sunday morning at 7:20 a.m. to her nine-year-old throwing a tantrum. Dargouth teaches parenting techniques for a living. She didn't need ChatGPT's guidance. What she needed was immediate calm. She reached for her phone and started talking to Claude, asking it to help her breathe through the screams.
She describes the advice it offered as fake therapy. It worked anyway. The synthetic empathy got her through the morning. It struck her then: does it matter whether the help is real if it actually helps?
The Mess That Matters
Therapy, Dargouth reflects, requires something AI cannot easily provide. It requires sitting in chaos. Her former professor once described therapy as sorting through a messy closet: everything gets pulled out, the room becomes a disaster zone, and only then does order emerge. The tornado is the work.
She thinks back to a recent session where a patient sat tense and ashamed, her voice cracking with anger at the world and herself. Dargouth fumbled for words that missed their mark. Silence fell between them, awkward and unhelpful. At that moment, she imagines ChatGPT appearing with its clean, efficient interpretations. She was tongue-tied. The AI would have been eloquent.
But here is what Dargouth has learned: the mess often signals something important is happening. Conflict, hesitation, false starts, strong emotion that shatters words, the slow emergence of understanding week after week. These are not bugs in the therapeutic process. They are features. Healing often travels in circles. It takes time. It doesn't submit to the demand for immediate answers that AI is built to satisfy.
AI's superpower is its clean, all-knowing stance. That same stance is its liability in the slow work of human change. AI cannot tolerate not knowing. It cannot sit with the unsteady path. It cannot be moved by fragile, absurd moments of beauty that sometimes have no words.
The patient Dargouth agonized over comes back to therapy. They continue to work through jagged emotions and unfinished pain. One day the patient mentions that what actually helped was how Dargouth laughed at a joke as she was leaving the room the previous session. Not the rich therapeutic dialogue Dargouth had carefully constructed. Not the insights or interpretations. A laugh. A human moment. Dargouth didn't even remember the joke, but she felt humbled and happy to be a flawed human.
Dargouth wonders what the future holds. She acknowledges that fewer than 7% of people with mental health issues get effective treatment. AI technology is free, imperfect, sometimes competent, and available to everyone. In a world of scarcity, that matters. Telehealth will expand. Soon people may not know whether they're talking to a human or a machine. The distinction might blur further still as AI improves at recognizing facial expressions and mimicking emotional connection.
Yet Dargouth imagines a small counter-movement will persist: people who choose to stand against the current, who seek flawed human therapists who will frustrate them and annoy them in return, who will sit with them through emotional storms and feel genuine happiness at small good things. Therapists, she notes, will sometimes say the wrong thing. They will forget parts of a patient's history. They will also know when to stop talking and simply bear witness.
Author James Rodriguez: "Dargouth has stumbled onto something that no algorithm can mimic: the therapist's permission to be bewildered, to fail, and to show up anyway."
Comments