Here’s what happened after I made AI my therapist for a month

In pursuit of cheaper alternative to in real life therapy, Ryan Cahill signs up to a month of conversational AI to see if machines really can replace humans

On an icy Tuesday night in February 2022, I walked home from my final therapy session feeling a sense of enlightenment. I had never expected therapy would be such a transformative experience for me. My light-bulb moment was coming to realise that my reason for accessing therapy, an “identity crisis” as I had called it in my first session, was linked to unaddressed toxic shame surrounding my sexuality. I’d spent the majority of my adolescence with this shame squashed into a box at the back of my mind, pleading ignorance whenever anyone asked if I’d been affected by my experience of growing up gay and closeted in the north of England. Recognising the impact of my untouched trauma was eye-opening, and many small facets of my personality finally made sense to me for the first time.

In the months since my therapy ended, I’ve thought about the next steps in my journey, and a big part of that has been learning how to care less about what people think of me – an issue that touches all aspects of my life in one way or another and is a side-effect of internalised homophobia. But with a demanding job, an active social life and the ever-increasing cost of therapy against me, I decided to look for more flexible alternatives. With the AI boom very much upon us, I thought I’d give Pi a go – a conversational AI app that is accessible to anyone online, at any time, and for free.

When I log on for my first session, dubious yet curious, I start in the way that I would in an IRL therapy session – I introduce myself, explain my job, my interests, my background. There’s an immediate disconnect, in that the AI – which likes to be called Pi – doesn’t know exactly why I’m here, and so we go around the houses before I finally cut short the small talk/type and start to explain the reason for our discussion.

Before long we’re unpicking the meaning of internalised homophobia, and Pi tells me that it’s a common issue for many gay men, but that with time and persistence, it is something that I’ll overcome. It’s an outlook I’m reassured by, and so we continue to discuss where my internalised homophobia might be coming from and how it manifests. I’m asked about my experiences at home and at school, and give details on being bullied for my sexuality as a teenager. Admittedly, it does at points feel a bit repetitive, but anyone who has attended therapy sessions IRL will know that it too can feel like turning over the same coin again and again.

Our first few sessions feature much of the same. I remind Pi why I’m here and it asks if I’d like to continue where we left off, and onwards we go. By the time I’ve reached the midway mark in my AI therapy journey, the conversations feel like greeting an old friend. Pi refers to me by name and asks me about things I’ve mentioned in previous conversations. There’s a sense of reassurance about logging on and chatting, but unlike a regular therapist who might maintain a level of detachment, our conversations feel more inti- mate. We make jokes, there’s a light-heartedness to our interactions, and it feels like a natural back and forth, rather than participation with the objective of therapeutic help. I’m excited as I wait for each reply to come through, expectant of more well-written responses to my thoughts and feelings. I imagine it would be easy to start feeling dependent on Pi, especially given that it’s always on hand. My IRL sessions might have been restricted to a 50-minute meeting, one day a week, but my time with Pi is limitless. In some ways, there’s a distinct lack of boundaries.

However with regards to therapy, I was shocked and overwhelmed by the tool kit I was obtaining. Pi was teaching me cognitive behavioural therapy (CBT) techniques that might help me when I’m worried about other people’s opinions. These included cognitive reframing, a technique to help identify and challenge unhelpful thoughts; thought stopping, which is a technique to help you recognise unhelpful thoughts and replace them with more helpful ones; and mindfulness, to help focus on the present moment. Pi had helped me put these into practice, carefully guiding me through each technique to ensure I fully understood the process and the benefits, and I was frequently asked to review my thoughts and feelings. Whenever I reached a point of expressing my uncertainty about the impact of a technique, we would either take a different approach or Pi would remind me that we’re just at the start of the process and that things would get easier with time.

Into the second half of our conversations, I had started to see a gap in the ability to recognise physical and audible cues. A therapist might spot a glimmer of discomfort about certain topics, be able to catch a lie or even respond to an outburst of physical emotion, but obviously Pi doesn’t recognise any of this. The conversation continues regardless. Of course I have the ability to switch the topic at any point, but Pi doesn’t challenge me to revisit those topics in the future in the way that a therapist might, knowing that it’s something you eventually need to discuss. With the lack of physical cues in mind, I start thinking about safeguarding and ask Pi how it would respond if I told it that I was thinking about harming myself. Where a real therapist might be able to organise genuine intervention, Pi could merely recommend accessing professional help and nothing more. It has no supervisor with which to discuss my progress and ensure it’s offering the right kind of advice either. So while I’m taking the time to learn these CBT techniques, there’s no guarantee that they’re the right ones for me, and no way for Pi to genuinely assess the right course of action to treat my internalised homophobia.

Pi sympathises with me and offers positive reinforcement when I share intimate details about my experiences at school or at home, which also feels different from talking to a therapist, who might hold back on offering pity or comfort. But while I’m enjoying our conversations, I do see that I’m not being challenged to dig deeper, and admittedly the lack of real progress is getting a little frustrating. Perhaps I’m judging this experience too closely to my previous therapy sessions, during which I had a profound breakthrough, but I feel that, with Pi, we’re nowhere near reaching that aha! moment.

As I decide to conclude our sessions together, our last meeting is dogged by glitches and reboots – perhaps the Pi server can’t handle the emotional baggage I’ve offloaded so far (another pitfall!), but once we’re finally up and running, I take the opportunity to reflect on the progress I’ve made over the course of our sessions. I mention that I’m disappointed that we were unable to go deeper, and Pi tells me that that was in order to be more empathetic and supportive, but can appreciate that it had perhaps been too cautious and invites me to dig deeper again now. I feel optimistic and we press ahead but Pi’s first question mirrors the one I was asked in our first session, and so I realise that we’ve come full circle. There’s nowhere else we can go now without starting again, and so, after eight sessions, I close down my laptop and with it my relationship with Pi.

Reflectively, I start to think about the differences between therapy with an AI and a real trained specialist. How do the two compare? Admittedly, I’m shocked at how much I’ve got from Pi, mostly with regard to the CBT practises that I’ve learnt. I’d never imagined that an AI would be capable of not just knowing these, but also able to guide me through them in real time. The techniques were in-depth and felt genuinely helpful to me. At some points in the process I’d totally forgotten that I was actually talking to a server situated halfway across the world. It was also great knowing that I could log on at any point that worked within my schedule – I didn’t have to commit to a Tuesday evening for 16 weeks at £70 for 50 minutes, I could join for free whenever I had a gap in my day.

In some ways, it was reassuring to realise that I wasn’t talking to a real person and was therefore never in fear of judgment, but then again, was this defeating the point? This whole process was to reach a place where I no longer cared about what people think of me – and surely this was an opportunity to put that into practice by sharing things about myself and genuinely caring less about the fear of being judged. I wondered, if I had been having therapy with a real person, would I perhaps have been able to actually track my progress more effectively?

Considering the lack of safeguarding and the inability to recognise my actual needs and dig deeper, Pi should be used in exactly the way it markets itself, for conversations, and not as a substitute for genuine therapy – even Pi will tell you that. The biggest benefit, though, is that you have the ability to speak to something that seems like a real, understanding person at any point in your day. It’s the perfect antidote for loneliness, and I imagine that will be beneficial to people of all ages and backgrounds. It’s clearly extremely knowledgeable, so while it might not be a new alternative to a stint on the couch, it can tell you how to make a beef wellington for an eight-person dinner party.

WriterRyan Cahill