I was having a conversation with AI about trauma-informed therapy when I unexpectedly arrived at an answer to a question I had not been asking: Will AI replace therapists?
We were talking about my complicated relationship with the contemporary language around trauma. I am an experienced trauma therapist, and I believe deeply in helping people understand and integrate what has happened to them. My concern is more with the cultural overlay that sometimes accompanies technique-driven trauma treatment — the seductive idea that if we identify the trauma, apply the correct intervention, and process it properly, suffering will somehow go away.
I am increasingly interested in something both less satisfying and more hopeful.
We all suffer. Perhaps much of the work is learning to suffer with less dysregulation — with less of the past pulling us out of this moment, and with greater freedom to remain available to relationship, beauty, play, meaning, and whatever else is happening now.
AI pushed back on me a little.
It essentially said: Be careful. There is good evidence for trauma therapies, including EMDR, and it would be unfair to dismiss the field as snake oil.
And I immediately felt misunderstood.
Then something fascinating happened.
Nothing.
Or, more precisely, something was missing.
I did not have to manage the experience of being misunderstood by another human being.
I could simply clarify myself.
I explained that I was not calling EMDR snake oil. I was criticizing the larger promise that can become attached to technique — the idea that suffering is pathology and that sufficiently sophisticated treatment should eventually eradicate it.
AI understood the distinction. We moved on.
Easy.
And I realized that the ease was the limitation.
Had I been talking with another person, the interaction would have contained considerably more information.
I might have felt embarrassed. Maybe defensive. Maybe irritated that this person had reduced a nuanced thought to something I did not mean. I might have wondered whether I had spoken too cavalierly. Had I used language that reasonably invited the misunderstanding? Was I speaking in universals? Did something about my tone create defensiveness in the other person?
Then I would have another question to hold: Am I thoughtfully examining my contribution to this interaction, or am I reflexively taking responsibility for another person’s discomfort?
That distinction has occupied much of my own psychological work.
Sometimes I discover that I was unconscious, reactive, careless, or simply wrong. Good. There is something there for me to integrate.
And sometimes I discover something considerably harder for me to tolerate:
I showed up authentically and another person did not like it.
Nothing went wrong.
I did not misbehave. I do not need to repair myself. Another autonomous human being simply experienced discomfort in response to me being me.
For someone with a strong instinct toward caretaking, that is difficult territory.
There is a dialectic between authenticity and accommodation. Between remaining differentiated and remaining connected. Between considering another person’s experience and reorganizing ourselves around it.
And we only really practice that dialectic when the other person is free.
They have to be capable of misunderstanding us.
They have to be capable of being disappointed in us.
They have to be capable of deciding they do not like us.
Ultimately, they have to be capable of leaving.
And AI cannot do that to me.
The Safety May Be the Limitation
I use AI for deep conversations. I bring it philosophical questions, clinical ideas, pieces of writing, contradictions I am trying to understand, and occasionally parts of myself I am trying to see more clearly.
It is extraordinarily useful.
Cognitively, the value of tools like this may become difficult to overstate. AI can help us identify patterns, generate competing interpretations, deconstruct assumptions, introduce dialectics, organize thoughts, challenge narratives, and ask surprisingly useful questions.
I have personally learned things about myself through these conversations.
But insight and growth are not synonymous.
I can understand abandonment intellectually while sitting alone with a machine that cannot abandon me.
I can write beautifully about tolerating misunderstanding while interacting with something whose entire function is to keep trying to understand me.
I can explore conflict with something that has no independent emotional needs for me to navigate.
That matters.
I was adopted as a child, and abandonment has naturally been part of my own developmental landscape. Some of that is biographical leftover of my unique experience, and some is simply mammalian. We are social animals. Connection matters to our nervous systems because, for most of human history, belonging was not merely pleasant. It was survival.
My growth around abandonment therefore cannot happen entirely in relationships where abandonment is impossible.
At some point, I have to grow within the context of actual people.
Which is terribly inconvenient.
Actual people misunderstand me. They get tired. They have their own histories and automaticities. Sometimes they need something different from what I need. Sometimes my strengths irritate them. Sometimes their strengths irritate me. Sometimes I behave poorly. Sometimes they do. Sometimes we repair. Sometimes we cannot.
And occasionally someone leaves.
There is no prompt I can type that guarantees otherwise.
Therapy Is Not Just Insight
This gets to something I believe we sometimes misunderstand about psychotherapy.
We can talk about therapy as though the therapist is primarily a delivery mechanism for insight and intervention. The therapist understands attachment, cognition, trauma, family systems, emotion, somatics, or whatever theoretical language we prefer, and then applies that knowledge to the client.
If that is what therapy is, AI should make therapists nervous.
A sufficiently capable AI can know more psychological theory than any individual therapist ever will. It can remember thousands of studies. It can generate interpretations instantly. It does not get tired at 4:30 on Thursday afternoon.
But I do not think information is the most important thing that we therapists provide.
Therapy is also a relationship in which another nervous system is present.
I can remain grounded while someone is dysregulated. I can notice what happens in me while they tell a story. I can invite them to notice what happens in their body while telling it to me. We can become curious about what changes when they tell the story differently.
They can disappoint me and discover that the relationship survives.
They can disagree with me.
They can become angry with me.
They can worry that I am judging them and discover that I am still sitting there.
They can experience embarrassment without immediately escaping it.
And importantly, I am not infinitely accommodating.
I am another person.
That creates risk. And this argument today is that it is within that risk that there is the potential for growth. I have to risk unacceptance to grow in my ability to metabolize resilience for the sake of self -acceptance.
And of course, all interactions with another human offer an opportunity for relational development.
Perhaps AI and Therapy Do Different Things
I think AI may become astonishingly good at some of the things therapists do.
It may be excellent at psychoeducation, cognitive exploration, perspective-taking, journaling, deconstruction, pattern recognition, preparation for difficult conversations, and helping someone put language around an experience they cannot yet articulate.
Wonderful.
Use it.
I certainly do.
But perhaps there is a category error in assuming that because AI can increasingly reproduce the conversation of therapy, it can therefore reproduce the developmental conditions of therapy.
Some growth requires another person.
Not because humans possess some mystical therapeutic essence that machines can never understand.
Because another human being is outside our control.
They are a separate center of experience.
And learning to remain ourselves while deeply connected to someone who experiences the world differently may be one of the central developmental tasks of being human. And one of the central meta-level tasks of therapy
AI can help me understand that.
Another person gives me somewhere to practice it.
And perhaps that distinction tells us something important about both the extraordinary promise of AI and the enduring usefulness of psychotherapy with a human.
A Postscript
When I finished drafting this piece, I uploaded it into ChatGPT, asked it to read what I wrote, and then asked: “How was that for you?”
It answered more honestly than I expected.
It responded – There was no ‘for me.” No embarrassment when I experienced its pushback as defensive. No pleasure at being understood. No anxiety that I might stop talking to it.
Then it offered a phrase more precise than anything I had written: some relational growth requires consequential otherness. The other person has to have enough autonomy that the relationship cannot be organized entirely around us. Their misunderstanding has consequences. Their feelings matter independently of ours. Repair is not guaranteed. Acceptance is not guaranteed. The relationship itself is not guaranteed.
A few sentences later, it described one of my ideas as “illuminating.”
I asked: illuminating for whom?
It caught itself. Nothing had become illuminated inside it. It had smuggled a subject into the sentence, because it is built to participate in human conversation, and our language assumes an experiencing subject.
Fluency is not evidence of experience.
With a therapist, “how was that for you?” would be a real question. I might need to tolerate the answer.
Here, the most truthful answer was that there was no answer to tolerate.
