AI’s power is undeniable despite its magnificent bloopers (Fahim et al., 2025). It organizes written material with perfect punctuation, taking progress notes off the clinician’s plate. I don’t believe there is anything useful to be learned about clinical work from writing notes, so for me, this is a free lunch.
Like all computers, or even all lists, AI never skips a step in what it is taught to consider important. An AI-informed supervision is constantly reminded to treat speech as metaphorical and not literal, since humans, including therapy patients, are very good poets and very bad reporters. AI never forgets to inquire why at this moment this thought occurred to the patient. It never forgets to consider projective identification—the communication of an intolerable feeling by getting the therapist to feel it. It never forgets to consider a lose-lose comment during a frame deviation.
AI can equalize the power imbalance in supervision, if you think that’s a good thing. With AI, supervisees—and their patients, for that matter—have as much factual knowledge as the person with more power. To me, that’s a disadvantage, because I think the cure for the effects of misused power imbalances is not to eradicate power imbalances but instead not to misuse them. This exposes supervisees—and patients—to power imbalances and helps them discriminate the exploitive ones.
In short, AI is smarter than the therapist in the therapy and smarter than the supervisor in the supervision. Its allure emphasizes intelligence in the clinical process, but is intelligence the thing that is wanted? It’s certainly advantageous in discriminating psychogenic from biologic disorders, and in certain kinds of assessment, like personnel selection and violence prediction.
But how often have we blown up therapies by telling the patient the answer to the problem rather than leading them to it? How often have we tried to tell patients how to be healthier rather than focusing on understanding which of their peculiar expectations interfere with healthy functioning? Once it is determined that psychotherapy is the best course of treatment, AI’s utility diminishes. At that point, AI is all about getting it right, but getting it right might not be the right focus.
Much problematic behavior on the part of patients, supervisees, and supervisors follows from the aversion to being stupid. Patients behave in patterns that come to define them and the world; to change, they have to confront the fact that they’ve been doing it wrong for a long time. Therapists are similar. They hold on to their first clinical theory like it’s their mother tongue. When they first learn the importance of frame management, they are confronted with how many therapies they conducted less than ideally, and many therapists conclude instead that they have been doing it right all along. Supervisors resist confusion and resort to trite prescriptions.
Perhaps the most constructive role AI can play in supervision is stupidity exposure. This can facilitate therapists and supervisors entering both kinds of sessions with a beginner’s mind rather than with the mantle of expertise.
One way to look at supervision is to cast the supervisor as Sacagawea, with the therapy dyad playing the parts of Lewis and Clark exploring the patient’s internal landscapes. The supervisor is familiar with some of the customs and languages of that landscape, but is not so knowledgeable as to provide a clear roadmap for the journey.
Apparently, one of Sacagawea’s important contributions to the expedition was her very presence as a woman with an infant. This signaled to indigenous tribes that the expedition was not a war party, and indeed, one of the remarkable accomplishments of Lewis and Clark was that they made it all the way to the Pacific and back without killing anyone. Supervisors can partially replicate this advantage by bringing their whole selves, their emotions as well as their wisdom, into the situation, not as a source of childish demands but as a display of humanity.
Supervision is a place where therapists can learn to probe without eliciting an aggressive or defensive response. Many therapists, instead, avoid probing, and while it is true that Lewis and Clark also would have not killed anyone if they’d never left St. Louis, it would not have been worth paying them to do so.
Watzlawick et al.’s (1974) model of case conceptualization is relevant here. They said we settle on an understanding of the problem that generates ideas of what to do. If what we do resolves the problem, the case is closed. If not, we formulate a new understanding. What is not relevant is whether the formulation was correct.
Richard Feyman, the Nobel-prize-winning physicist, named a book, The Pleasure of Finding Things Out. Intersubjectivity, at the heart of relational therapy, emphasizes “making sense together,” which Buirski, Haglund, and Markley used as the title of their book. I think our watchwords as supervisors ought to be “the pleasure of finding things out together.” Supervision can teach what it’s like to engage in that sort of enterprise from the subordinate side in hopes of fostering that sort of enterprise from the authoritative side.