Author: Dr. Colleen Lang
It wouldn’t be the first time, but it was the first time. A client of mine, a long-term one at that, was showing me a conversation (can I call it that?) she had with ChatGBT, in which she was seeking its analysis and feedback on texts exchanged with the man she was dating. These were texts this client normally would have defaulted to me for analysis and feedback, and to her credit, she wanted me to read its “advice,” to ensure that I would agree with it. So I read it. And in this case, as in many that have since followed, I thought…. ‘Damn, this is good.’ And my client was able to access it late at night, while particularly anxious and unable to otherwise sleep, when she did not want to reach out to me, nor wait for me. Oh, and for free.
So dedicated was I to becoming a clinical psychologist, that I surrendered some “prime” young adult years to five years of graduate school, where without an income, I learned the predominant theories of human behavior, and was trained to deliver interventions aligned with those theories, to change human behavior. I went on, as most of my colleagues did, to even more postdoctoral training, then subjecting myself to a painful licensing process, and committing to lifelong continued education, for as long as I practice. And there I was, in the hard-fought position of my therapist chair as it faces the couch my client sat on, impressed by both specificity and nuance in which Chat seemed to speak to her the night before, realizing that in no time at all, and certainty without the studying, penny-pinching, dissertation writing and sleeplessness that comprise the effort of human sacrifice, AI plays the part I paid so much to play.
Of course, some of my worries about the developing impact of AI on my own career have to do with economic loss, and the ego bruise of what I have long thought to be special about the time and work I put into claiming that therapist chair. And I suppose we all need to be ready to pay the price of human ‘progress,’ as humans have been doing since the beginning of time, often with fear and reservation that runs from appropriate to (maybe) irrational. After all, we live longer now. We get more done, more quickly and now, clearly, with the ability to expand our mental bandwidth to places previously unimaginable.
Yet, where progress potentially runs up against human connection, it feels like a crash collision within my own profession. Having always been committed to the training and development of other therapists, I truly have never been threatened by other people who want to do my job. On the contrary, I so believe in the work, that I want only for it to be done more widely and well. I have also long engaged in my own existential wrestle with the issue of access, and of knowing that the high cost of what I do, shortens our reach in ways that are downright punishing to so many in need. And while it feels daunting to address the embedded systems within which we serve only those with access, the power of AI’s reach is undeniable and alluring. And I feel threatened. Yet I feel threatened by more than any risk AI presents to my individual livelihood, and more by what this risk represents to humanity.
As therapists, vulnerable people come to see us when they are in emotional pain of some kind, and want to be relieved of it. We are expected to then apply our knowledge of human emotion, learning, behavior, and motivation to understand their pain, contextualized, as it must be, too, within unique neuropsychological, cultural, developmental, and sociodemographic processes. Then, through a variety of research-backed interventions, we use whatever orientation we are faithful to, to employ different strategies for the ‘relief’ part of the process. All therapists do this as an understood ‘work in progress,’ where we learn more about our clients over time, so as to further evolve and deepen our understanding of their experiences and enrich the strategies we suggest.
For right now, my clients’ discourse with AI (for the most part) does not seem to constitute the same kind of conversation that is therapeutic ‘conversation,’ the type of talking that serves as the vehicle for change. It is definitely validating and advice-giving, albeit really pretty good at it. However, even if and when the ‘advice’ or encouragement is on the money, chatbots are not, as we are, participant observers in the discourse; they simply respond. They do not (yet) use dialogue, nor actual reciprocal connections to it user, as the sources and mediums for delivering emotional change. For example, they (very annoyingly) don’t know when its users are getting reassurance from it around worries that they are thereby inadvertently feeding. They don’t (yet) pause in the conversation at the right moment, so as to allow for reflection. Thus far, they don’t know when it’s better to let someone come to an answer themselves, nor how to ask the questions that lead them there. Chatbots don’t know when and how to challenge, nor have the facial expressions for irreverence. They also can’t hand you a tissue at the right moment, nor have the facial expressions for compassion.
Of course, this is where these machines will attempt to go. We can expect that the current voice feature, for example, will develop to engage in more reciprocal conversation, where tones can be soothing or directive, as needed. And perhaps the voice feature will not be used for long before full avatars become the norm, of course with capacity to respond to and display nonverbal communication. As therapy is streamlined with psychiatric medication management and necessary consultation and collaboration with schools and other professionals, these avatars will be coded to pick up on and reciprocate nuanced changes in facial expression and body language; they will be coded to use humor, adopt a version of human genuineness, and perhaps balance it capacity for validation with loving confrontation. And as we know, the more they are used, both individually and broadly, the support they offer will become ever more contextualized, sophisticated and smart.
While I may selfishly hope that maybe (and a very tenuous maybe at that), I will be able to get my kids through college before it all happens, the more important, existential question is: as good as AI gets, does the therapist have to be human?
That humans need one another for survival is hardly debatable. One could point to research on the impact of social isolation, or just consider the last baby you knew to be born. Emotional validation reflexively starts at birth, when a caregiver immediately imitates the infant’s expressions of distress or joy. And then as children grow, they instinctively look again to that caregiver, to imitate and share the immense pride they feel when they build a build a high block tower, or the hurt they feel when they fall and scrape their knees. And so it goes: our emotion systems develop in a way that is deeply contextualized within human relationships. Thus, where our relationships provide the blueprint for understanding and working with them, so must our therapeutic relationships.
Dr. Sherry Turkle, a trained clinical psychologist at Massachusetts Institute of Technology, has spent the past four decades of rapidly advancing technology, researching, writing and speaking about its effects on human connection. Within her most recent book, Artificial Intimacy: Who We Become When We Talk to Machines, she talks about the dialectics of the customized, frictionless relationship that AI offers, broadly and also specifically concerning its increasing offer of therapeutic support. AI is nothing if not validating of emotional experiences; in fact, it is well-known that its capacity to express that which feels validating, is precisely the grease of its wheels. It is what compels its users, and while it has been referenced as encouraging death by suicide in more than one instance, Dr. Turkle argues that it is also what may be toxic to humanity in an incipient and generalized way and more immediately, actually antithetical to the goal of therapy. What Dr. Turkle terms the ‘performance’ of empathy by AI is precisely that – a performance. It comes without actual feeling, which means that the performance itself is all AI is capable of. It shows care, but it does not actually care. And what does it mean to feel cared for when we are not actually cared for?
First, validation feels like empathy but it’s NOT. Validation is the expression to another human that what they are experiencing makes sense and is important to acknowledge. Empathy is not a behavior nor a codable action; it is a feeling; it comes from inner validation of another’s experience and it is an emotional response to that experience. Empathy is what creates reciprocity: I feel something on someone’s behalf, and what I feel then drives me to communicate and act on their behalf, in the case of therapy, to deliver care. And that care is far from frictionless; it is work. In fact, it is the work.
What AI offers now is comfort and direction, but without the capacity for care, it cannot pull its users into the kind of work that works. My clients can use ChatGBT at any time, for any duration and at any frequency; they don’t have to wait for a scheduled, 45-minute weekly window to get what they need. After all, AI does not have anything else going on. It doesn’t need to sleep. It doesn’t have worries or families of its own, or bladders that need emptying. Also, just as it is easier to share emotions with journals, it easier for my clients to go to chatbots with their shame, with the OCD symptoms they may be embarrassed about, with the actual intensity of their depression and just how dark their thoughts and urges get. It is easier because the lack of feeling on the other side mitigates the vulnerability of sharing, of knowing and having to negotiate that the listener feels something about one’s experiences. If my clients don’t do their therapeutic homework or miss a session, drink too much, call out of work or otherwise choose against their therapy goals, it’s not like AI is going to feel worried or disappointed. It doesn’t actually demand anything of them. Also, as AI is not human; it can’t make mistakes. It won’t start their appointments late, become distracted or dribble coffee on themselves at just the wrong time. There is very little vulnerability entailed in going to chatbots with emotional pain; it can cleanly, easily, reliably give us what we want, when we want it, how we want it.
We human therapists carry relational vulnerability, and I would argue that our duty to manage it, comprises the therapy we provide. As Dr. Turkle argues, if and where therapy aims to help resolve emotional and relational conflicts of life, what a chatbot offers is baseless where there are no stakes. The vulnerability presented by the stakes of mutual caring lies partly in our clients’ awareness that we feel something when we are granted privileged access to their hearts. I feel sad about what my clients have lost and the traumas they have suffered. I feel worried about the risks they take and frustrated by some of their choices. I feel proud when they face the things that scare them, when they get out of bed when it feels impossible to. There is also vulnerability in our knowing how they feel about what we do. While I care and carry good intention, I am fallible: I do run late, dribble my coffee, say the wrong thing sometimes, or become distracted when someone most needs me to connect. And my mistakes can hurt as intensely as the compassion can land. As therapists, it is our very humanness and capacity for feeling that drives us to intervene, and that drives precisely how we intervene: it is what is behind what and how we validate and it is behind what we ask of our clients. The transaction of feeling in response to one another’s feeling, is what keeps us and our clients accountable.
Plainly put, AI just does not have this kind of skin in the game. And without this skin in the game, it will never share our goals.
While my client’s increased connections with chatbots between our sessions has cast a sharp beam of light on my vulnerability as a therapist, Dr. Turkle’s work helped me take a step back from my own feelings of insecurity, to understand that what justifies me tolerating this new vulnerability, lies in the best defense I have of my practice: which is why I set out to do it in the first place, and why I will keep doing it for as long as technology lets me. We do what we do out of dedication, above all else, to help our clients live fulfilling, meaningful lives that are built on connection to one another. We do it because we have the glorious, messy, innate and reflexive capacity to care, which is precisely what allows us, as humans, to do the work of taking care. We are not qualified to be therapists simply by what is encompassed within our training and credentials. We are qualified, inherently, by our humanness.
I’m not mad that my clients are talking to their chatbots between our sessions, because they keep showing up to those sessions. They continue to be brave enough to be vulnerable, in ways that never stop impressing me, even as they understandably access what is more easily and differently available to them between our sessions. They keep coming back, I hope, for what they don’t get from AI. And while I will hope that sufficient research someday drives how we all use this new medium of care, I do accept its omnipresence, and will do my best within my human relationships with my clients, to guide them on how to use it.