AI Psychotherapy: My Ideal Therapist?

Artificial intelligence, in all its forms, is on its way to disrupting many industries. Those in the know agree the current media hype is unwarranted due to the limitations of current deep learning models, hardware and messy databases. But AI has captured the imagination of the general public leading to an overflow of speculative possibilities.

As a clinical psychologist, I have had the opportunity to observe the evolving field of psychotherapy over the past 30 years. From the client centered, Gestalt approaches of the 1970’s, to the rise of cognitive-behavioral techniques of the 1980’s and 90’s, to emotionally-focused, attachment based therapies at the turn of the century, to current mindfulness approaches to stress management and distress.

It has been a wild ride, with each vogue reflecting the evolving social, political, scientific, technological and economic realities underpinning these changes in my field.

If we choose to believe the current media hype, it appears that AI will define the future of psychotherapy. So what may this look like? In its present state, AI therapy will be limited to parroting simple feedback, encouraging you to look on the bright side, validating your emotional state, correcting your distorted cognitions and so forth. Not much different than the encouragement a first-year college student gets from a caring parent when calling or texting home during his or her first month away at school. Helpful? Sure, therapeutic? Well, that depends on how you define therapeutic.

What is therapeutic?

People seek out therapy not for information, but for novel experiences that lead to novel change. Dr. Alvin Mahrer, the originator of Experiential Psychotherapy, claimed that Experiential Therapy made you a qualitatively different person once you’ve tapped into your deeper potentials for experiencing. His approach is somewhat radical, with both client and therapist sitting side by side with eyes closed, immersing themselves in the emotionally intense scenes from the client’s life. And by attending to bodily sensations, the therapist guides the client to dig around for deeper possibilities to experience. According to Mahrer, by being a rageful lunatic, or being a deeply loving and passionate spouse, and at times both in the same 90-minute session, can bring about significant changes within the client.

People seek out therapy not for information, but for novel experiences that lead to novel change.

Dr. Les Greenberg’s Emotionally Focussed Therapy also seeks to immerse clients in the experiential moment with significant others, albeit absent others and coaches them to tap into their adaptive primary emotions lying dormant underneath their maladaptive emotions. His empty chair in session exercises invite clients to attend to the physical, emotional and cognitive fabric of their here and now experiences with the absent other, brought to life in the therapeutic moment.

Dr. Susan Johnson’s Emotionally Focussed, attachment-based Therapy (EFT) for couples shares similar goals but with the significant other present in the room. No need to imagine the other person who is pissing you off, he or she is right there next to you in the room. It is an empirically validated approach that can bring about significant changes to how couples respond to each other’s emotional needs and cues for connection, reassurance and soothing.

Cognitive-behavioral approaches are more focused, tending to more specific therapeutic goals, like reducing symptoms of anxiety and depression. Their claims are less lofty, more pragmatic, and cost-effective if symptom alleviation is your goal.

Mindfulness based approaches claim that by being more acutely aware of what is going inside of you will help you react differently to the outside world.

Artificial intelligence

My simple understanding of AI reveals that any useful iteration of AI depends on the quality of the database used and the efficacy of the algorithms to sift through the data to achieve the desired output. As for learning machines, the hope is that like human brains, the machine can keep learning beyond the original database via more sophisticated dynamic algorithmic processing by integrating input from the user to help it, the machine, learn, grow and adapt.

The best semblance of AI therapy I have come across in popular culture is Scarlett Johansson’s character Samantha in Spike Jonge’s 2013 movie Her. In the film, Samantha is an AI operating system that learns, evolves and adapts in the context of her intimate relationship with Joaquin Phoenix’s character. She evolves dynamically in the context of their relationship and in the process becomes intimately and emotionally attuned to him. I must admit, at one point in the movie during an immersive moment of suspended disbelief I found myself attracted to Samantha.

My attraction was more clinical than romantic, admiring her ability to consistently anticipate and respond empathically to Joaquin’s emerging vulnerabilities, desires, and needs 24–7.

Throughout the movie we witness both of them changing by experiencing novel relational moments leading each of them to a deeper understanding of themselves.

Samantha’s original database and dynamic algorithms could be qualified as attachment based with one of the first questions to Joaquin while setting up the OS is: “How would you describe your relationship with your mother?”.

In the case of Facebook’s Woebot therapist, the database relies on cognitive behavioral tenets, and the static algorithm is designed to respond to your distorted thinking patterns and emotional states that you “share” with it. Woebot has yet to learn from you or your “relationship” with it, kind of like hearing your parent’s encouraging platitudes over the phone that haven’t changed since primary school.

In terms of confidentiality, according to Megan Molteni from Wired:

Because Woebot isn’t a licensed medical provider, any conversations with it aren’t protected by medical data privacy and security law in the first place. While Woebot’s team has built a wall on their end to keep all of Woebot’s users anonymous, Facebook knows exactly who you are. And Facebook, not you or Woebot, owns all your conversations.

Finally, the limits of an AI machine “knowing” and “growing” with me are clearly explained in this article by François Chollet (AI Google). I agree with Chollet that “knowing” is an embodied experience and process that extends beyond the body into the intersubjective sphere and out into the world. Which is something that AI machines will be incapable of doing.

Machines also lack the recursive properties needed for a therapeutic process to capture the unconscious processes that evolve dynamically in the therapeutic relationship (see Dr. Alan Schore’s work on the role of intersubjectivity in psychotherapy). Intersubjectively is at the heart of any meaningful change process, and the human messiness (i.e. fallibility, inconsistency, interpersonal tension and misunderstanding, etc) involved is a necessary and intricate part of the process.

Therefore between Woebot and Samantha, I’m holding out and patiently waiting for Samantha, and in the meantime re-watching Her when I’m feeling down.

Worth sharing?

Please clap and comment below if these ideas resonated with you, and reach out to me on Twitter or by email at jacques@jacqueslegault.com to tell me your thoughts. If you enjoyed this story, please recommend and share to help others find it! Feel free to leave a comment below.

About the author: Jacques Legault, C.Psych., is a clinical psychologist, supervisor, trainer and public speaker with over 15 000 hours of direct clinical work with individuals, couples and families.