Mental Models of Mental Health
Our view of AI therapists hinges on what we think "therapy" is
This essay is cross posted from Capita.
Whether we want to or not, we all have to reckon with the idea of AI therapy.
The most hyped and best-funded AI therapy startup to date just launched. It’s backed by serious people and a compelling pitch: that it is designed with real psychology and delivers outcomes that rival or exceed traditional therapy. Dozens of other well-funded companies are racing to develop new models of care.
AI therapy makes economic sense in a world where demand for mental health care has exploded—driven by everything from reduced stigma to social media-fueled anxiety to the chronic uncertainty of political and planetary instability—while the supply of traditional, human-powered approaches struggles to keep up.
As you'd expect with any potentially transformative and disruptive moment, reactions to AI therapy have been all over the place. Some voices are exuberant, celebrating it as an accessible and stigma-busting breakthrough. Others are more skeptical, pointing to serious ethical concerns and glaring safety gaps. Some take an apocalyptic view, seeing it as an existential threat to the profession that the greedy industry will be all too eager to exploit.
There’s no clear consensus yet, and there may not be for a while. The unsatisfying reality is that we probably just need to let it play out. Already, wildly divergent policy poles are emerging: Illinois has moved to stop AI therapy before it begins, while the federal government seems to be taking a “damn the torpedoes, full steam ahead,” approach to AI innovation in general.
We disagree strongly because we aren’t actually arguing about the same thing. Before we can talk about whether a chatbot can do therapy, we have to ask: What do we think therapy actually is?
We all have different mental models of therapy, informed by culture, economics, experiences, and much more. These are not well-researched or carefully reasoned viewpoints, but rather more gut level and free associative: What do you think about when you hear the word “therapy”?
What follows is not a definitive taxonomy, but a set of recurring elements that shape our mental models. Whether a person finds the idea of AI therapy promising, troubling, or absurd often has less to do with evidence and more to do with which models resonate most for them.
1. Therapy as Self-Optimization: “Doing the Work”
This model sees therapy as part of a broader regimen of self-sacrificing self-improvement. You wake up at 5 a.m,, bike to the sauna, plunge into freezing water, journal your intentions, skip breakfast, and then log into BetterHelp.
Under this model, therapy isn’t necessarily a response to anything wrong. It's something anyone who is serious about bettering themselves needs to add to their routine. It might be painful, but “no pain, no gain.”
2. Therapy as Tactical Problem-Solving: “Fix the Thing”
You’re anxious about work. You’re fighting with your partner. You can’t sleep. So you go to a therapist to troubleshoot, like bringing your car to the mechanic.
This model is largely transactional and pragmatic. The therapist is a consultant you hire for insight, a new frame, and maybe a cognitive tool or two. Once the problem is solved, the transaction is over.
3. Therapy as Science: “Evidence or Bust”
The mantra for this field is evidence based. Under this model, therapy is first and foremost a branch of science-backed medicine: modalities like CBT, DBT, and IFS, all delivered to fidelity by highly skilled and rigorously trained professionals.
Subcategory: Therapy as Treatment for Mental Illness
A version of this model is therapy as a necessary intervention, usually delivered alongside medication, for diagnosable mental illnesses like depression, anxiety, PTSD, or bipolar disorder. Therapy here is less about self-exploration or personal growth and more about stabilization, functionality, and medical necessity.
4. Therapy as Relationship: “Someone Who Listens”
This is therapy as a nonjudgmental presence. Under this model, the most valuable part of therapy is just space: the ability to complain about your own unique problems for 50 uninterrupted minutes without your own guilt or minimization tendencies kicking in. From time to time, the therapist might gently chime in with a reminder or redirection..
The therapist is not there to “fix” anything. They are just…there.
5. Therapy as Health Care Business: “Billing Code 90837”
This model understands therapy as a service industry nested inside a massive (and often dysfunctional) health care system. It’s coded, billed, and reimbursed like any other medical appointment. It is rationed by payers and gatekept by licensure and credentialing. It might be a cold way to think about such a person-centered endeavor, but after all, “no margin, no mission.”
6. Therapy as Spiritual Repair: “The World Is Sick, So I Am Too”
This model sees therapy primarily as a way to cope with civilizational malaise: an antidote to loneliness, isolation, loss of faith and meaning, or the trauma of late capitalism. Therapy is not a solution as much as a tool to help achieve minimum viable functionality in a broken world.
7. Therapy as Childhood Excavation: “It’s Your Parents”
This model has gone from Freud to punchline and back to a certain level of cultural ubiquity. The text or subtext of virtually every movie or TV show about parenting is the presumption that parents inevitably mess their kids up. How often does Tony’s mother come up in The Sopranos therapy sessions? This model holds that present maladies are invariably symptoms of too much, too little, or the wrong kind of love from the people whose voices we still hear and whose patterns we often repeat.
There’s a certain fatalism here, but also a certain liberation.
8. “It’s Not Your Fault” Therapy
We all know the scene from Good Will Hunting: Robin Williams, as the therapist, faces a defiant Matt Damon. “It’s not your fault,” he says, over and over again, until Damon breaks and starts crying in his arms.
This is therapy as exorcism, where beneath all our anger and self-sabotage is something old and wounded. Something that, unless confronted and excised by the firm hand of a skilled and compassionate professional, will continue to hold us back from the life we are supposed to live.
Most people hold a blend of these mental models. Therapy can be a space for emotional release and a tool for problem-solving. It can feel both sacred and scientific or both relational and results driven. Our perspectives are shaped by personal experiences, cultural cues, and the particular needs we bring into the (real or virtual) room.
People don’t just disagree about whether AI can replace therapy. They disagree about what’s being replaced. AI therapy might work, or might not. But before we ask whether it can do the job, we have to agree on what the job actually is.
Jay, I really appreciated this taxonomy. Totally agree that our arguments often bypass each other because we’re working from different mental models of what therapy is.
It strikes me that the same is true for “AI therapy.” For some, it means psychoeducational question-and-answer; for others, Rogerian-style dialogue; for others still, a delivery channel for CBT. On top of that, the tools themselves differ in architecture: some are fine-tuned on clinician–client data, some are specialized models, and some are a layer over general-purpose systems. Add in shifting regulatory framings, divergent user and practitioner expectations, and the pace of technological change, and it isn't surprising that consensus is elusive. Until we surface those assumptions clearly, people aren't even debating the same thing when they talk about “AI therapy.”