The ther-AI-peutic relationship

AI is everywhere right now. It’s on our phones, our TVs, our smart speakers, and even in our fridges. It promises to make our lives easier, make us smarter, and make us more efficient. In amongst debates about its ethical use or its potential impacts – both positive and otherwise – is often a well-founded response that it simply represents the logical next step in a world that has been driven towards efficiency and automation since the beginning of the Industrial Revolution.

And yet for all its benefits, there are its costs; not least in its potential effects on the prospects of gaining and maintaining employment. This too is well-founded, with financial powerhouse Goldman Sachs cheerily predicting that 50% of our work might be fully automated by 2045, costing approximately 300,000,000 jobs and affecting 25% of the global labour market. 

So what then of the use of AI in the therapeutic world, which, give-or-take, has remained largely unchanged for the last 150 years?

In its steadfast commitment to its founding principles and the raw reality of human connection, the therapeutic world has found itself ever more out of step with the digital society that surrounds it. Technologies which have otherwise been broadly utilised in our globalised economy – automation, increased efficiency, outsourcing, the use of cheaper labour abroad, etc – have neither been applied nor are they particularly applicable. Maybe, maybe, your therapist writes client notes on an iPad, if they’re very fancy.

And while stigmas around therapy (by which I include counselling and all other types of mental health support) and those that use it are ever more proudly out of the therapeutic closet, the industry nonetheless suffers from selling a product which is ever more at odds with its market. And despite my own enthusiasm for therapy and admitted vested interest in wanting to continue to live with a roof over my head, even I have to admit that summarising it doesn’t make for a particularly great sales pitch:

“Got some problems? Come bare your soul to a stranger for no guaranteed results for an unspecified period of time! Brilliant! How much is it, you ask? That’s the best part – extremely expensive! And yes, of course I offer low-cost services, but no, I don’t have any space currently.”

But here’s the thing: the human limitations and failings of therapy aren’t just logistical; they’re baked into the necessary structure of therapy itself. One therapist, one client, one hour. A finite resource with built-in scarcity and finite mental capacity. Therapy is slow, private, emotionally costly, inconvenient, and inherently unscalable. Where inclination is increasingly geared towards immediacy, automation, and always-on access, therapy begins to look like a product crying out for disruption and modernisation.

Because its deliberate slowness is such a crucial part of the therapeutic process, it took something seismic to force a much-needed and transformative technological shift with the widespread adoption of online sessions – something I, like many in my field, had dismissed as an inferior product. So inflated by self-importance and our principles were we (therapists, you say? Self-important? Well, I never…) that my five years of training, completed well after the advent of online meetings through Skype and the like, had included exactly zero minutes addressing the significant differences that come with working with clients remotely.

Indeed, it took nothing less than a global pandemic, and the day after the announcement of the national lockdown in March 2020, and facing an empty office, I was suddenly much keener on the idea.

As the months rolled on, I was taught a big lesson. I’d been dismissive; arrogant, even. Online sessions were working, and I was forging connections with people whom geography would otherwise have kept out of reach. And now, with memories of queueing outside supermarkets and rushes on toilet paper and dried pasta an ever-fading memory, it remains a huge percentage of my ongoing practice.

It was an important lesson; not just about adaptability, but in thinking about the nature of connection itself. I’d assumed that removing the room removed key aspects of the work. It didn’t. It just changed the container. But as I and others like me creakily embrace technology in our practices, we are suddenly faced with settling a debate about the use of artificial intelligence (AI), before AI settles the debate for us. And this debate doesn’t just ask how and if therapy can adapt, but whether we’re needed at all. Articles like this are common, even daily occurrences. Increasingly, it’s a subject in sessions, too. Clients ask for transcripts from AI captioning software. There are (alarmingly) stories of therapists using AI for note-taking, summarising and diagnosing(!). Guidance is scant. Those doing the guide-ing seem frankly unprepared and ill-informed. 

For the consumer, there’s something understandably alluring about the idea of AI therapy. Platforms like Wysa and Character.ai offer an almost-too-good-to-be-true alternative – Waiting time? None. Availability? Permanent. Location? Everywhere. Cost? Zero. By almost all metrics, human therapy appears to be a worse product – more expensive, unreliable, you might get a duff one, and that’s all if they even reply to your initial enquiry email in the first place, which all too many don’t. When and if it arrives, that human connection is unpredictable and inconsistent, and, even if it works, can leave one facing a great deal of discomfort and difficult emotions – sometimes sold as “it gets worse before it gets better”.  AI offers a workaround to all of that: availability, affordability, and the comforting illusion of control. AI won’t judge you, there’s no need to sit in a poorly-furnished office; no public transport delays, and you don’t have to pretend you’re in a meeting/yoga class/seeing your mother (delete as applicable).

And for some people, as with the adoption of online counselling, perhaps for those with significant social anxiety, trauma, or who simply live in remote or underserved areas, the access offered by AI companions can feel like a lifeline. This isn’t a piece written in opposition to such tools – some of them may have real value. But it’s a mistake to conflate what we seem to term as ‘therapy’ with what is therapeutic. 

I’m asked often if I’m worried that AI is coming for my job. And if my job was to solve people’s problems, or to take the sum total of their presenting symptomology and give it a name, then I probably would be. But if we are to determine whether AI can ultimately replace therapists, or whether we therapists should be afraid of an oncoming technological client-grab, we must first correctly determine what therapy is.

There is something innately human about being in a therapeutic relationship. At its heart – sometimes referred to as ‘the frame’ or the ‘core conditions’ when we train – therapy is about being held. Not physically, but structurally. It’s about boundaries. It’s about facing discomfort. It’s the silent moments of reflection and wondering what you should say next; sometimes breaking that silence with an anxious filler, but sometimes with something profound. It is in the safety of that construct that one can begin to explore.

As the content of our exploration begins to test that structure, the rigidity of those boundaries and the limitations of that relationship play a crucial role in the process of growth, working in part because of its inconvenience, because of its limitations and its faults; because of the pauses, the absence, the waiting. It forces and enables reflection. It forces and enables resilience. The therapeutic work begins long before words, and continues long after them. 

AI can’t perceive silence. It can’t sit with discomfort. It can’t engage in transference or counter-transference – the essential relational dynamics that so often drive the deeper work. It can’t rupture, and therefore it cannot repair. It can simulate care, but not presence.

When thinking about how I would frame this article and illustrate the innate human-ness of therapy, my thoughts again returned to March 2020. Lockdown had brought with it an influx of new client enquiries. With uncertainty, loss and fear, and time to fill, much of those extraordinarily sunny days were spent staring into a laptop screen, with full days of client work. When restrictions were lifted, people in my profession were deemed ‘essential workers’ – hello, ego boost – and in-person work returned. One particular client elected to come in to meet me in my office, situated peacefully at the bottom of my garden. I remember being surprisingly apprehensive about ‘meeting’ them for the first time, as apparently was he, nervously declaring ‘I’ve got legs!’ as he walked down the garden path.

There was something very different about experiencing him as a whole person rather than just as a face in front of a blurred background, and while over time I came to appreciate the differences between working online and in-person with clients rather than considering one better than the other, it was as he began to talk about more sensitive issues that I became aware of him nervously shaking his leg; the more anxious he became, the more pronounced the shake. This repeated itself, and reflecting this back to him opened a Pandora’s Box of physical associations with his anxieties. I suspect our shoulders-and-up virtual interactions would have otherwise left firmly closed.

Energy isn’t data. To try and quantify in ones and zeros the space that exists between two people, be it in love or indifference or anything in-between, can only serve to cheapen or somehow corrupt it. The Swiss artist and sculptor Alberto Giacometti said “the object of art is not to reproduce reality, but to create a reality of the same intensity” and in a similar context, too, sits the unique intensity of the therapeutic frequency; a necessarily changeable and projected facsimile of one’s mother, one’s father, one’s place in their world, and one’s most ideal and most loathed selves.

And here is where AI falls short, and why I’m largely free of fear. Positive change or ‘help’ cannot be found in an echo-chamber, and convenience comes at a (therapeutic) cost. An online meeting might be easier, but one loses their journey to and from a session; a time for constitution and gathering of thought, and latterly for constitution and gathering of emotion. An AI therapist might reflect and listen and be permanently available, but they also foster dependence and lack the ability to challenge or work with the ‘space’. And while there are innumerate business cases in innumerate industries, and an aching need for the tools AI can offer, they can’t give a client what they really need. Ultimately, AI is little more than a sugary snack for the hungry; filling but not nutritious. 

At its best, one’s therapeutic experiences should open our minds and challenge our thinking, helping assert ownership of our thoughts and actions and enabling better congruity of intent and outcome. A therapist, if nothing else, should be a mirror, reflecting your thoughts and behaviours to help you understand yourself in a deeper and more objective way. As I thought about this article, I was thinking about the kinds of positive experiences I’ve had, and the positive reflections of others who I’ve talked to about theirs over the years, and it struck me that we were joined in one uniting ingredient – when we talked about what worked, we all referred to the person not the practice.

While the practice and the methodology of therapy can be summarised and codified, the transformative magic of therapy's value lies in the unquantifiable, uncodifiable fallibility and wonderful messiness of human connection.