Can an AI provide what talk therapy needs to work? Or are humans required?

I expect that human therapists, counselors, and coaches are not going to be replaced by artificial intelligence any time in the near future. In my estimation, AI therapybots are just about as likely as fully self-driving cars. There are things humans can do that robots can’t.

For example, I don’t believe that we can build bots right now that:

  • Express true empathy, in a way that makes someone feel seen and validated
  • Form a genuine relationship that provides a space for someone to grow and evolve
  • Provide accountability so that someone can progress towards their goals, knowing that someone important is there to cheer them on when they succeed and boost their confidence when their energy flags

At the same time, I hope that AI/ML and other data-driven techniques can help us scale human-based counseling and coaching in a way that improves people’s lives. Right now, it’s really hard to get the guidance you need right when you need it.

Is there any academic theorizing that can illuminate these issues? Why yes, there is! In this post, I’ll take a look at two different ways of conceptualizing how counseling works to improve people’s lives: the more widely accepted empirically-supported treatments model vs the intuitively appealing common factors model.

How does talk therapy work anyway?

Let’s start by investigating how talk therapy works. Let’s assume for the sake of this post that it does work, as much research evidence suggests.

There is debate among mental health researchers and clinicians about exactly how different varieties of mental health counseling work. Does cognitive-behavioral therapy work because of its attention to repetitive negative thoughts, and its training in how to dispute them? Does motivational interviewing succeed in helping people manage their substance use due to its special emphasis on strengthening the “whys” behind a proposed change? Or do the various talk therapies work mainly because they bring together a common set of elements: empathy, acceptance, goal-setting, accountability, and skills training?

The common factors theory in clinical and counseling psychology suggests that the various evidence-based approaches in psychotherapy share a set of common elements which drive much of their effectiveness. The alternative to the common factors theory is the empirically supported treatments (EST) movement, closely aligned with the evidence-based medicine perspective. EST says that different approaches work via their special and unique tactics. This perspective drives research that aims to demonstrate that one approach is better than another.

One version of common factors theory, the “contextual model” developed by Bruce Wampold, emeritus professor of counseling psychology at the University of Wisconsin–Madison, proposes that there are three pathways via which psychotherapy works, regardless of its exact methodology.

These three pathways are:

  1. The real relationship between therapist and client
  2. The creation of expectations in which the problem is explained and the treatment is justified
  3. The enactment of health promoting actions, which may be specific to a particular approach, but are designed to move the person towards the envisioned goal state

You can see similar underlying factors in this alternate formulation of a common factors approach:

  • Therapeutic alliance
  • Empathy
  • Goal consensus and collaboration
  • Positive regard and affirmation
  • Mastery
  • Congruence/genuineness
  • Mentalization
  • Emotional experience

Is there good evidence for common factors theory? It’s certainly intuitively appealing, but many of the studies completed to date have presented only correlational evidence. Research on talk therapy is made difficult by a variety of complexities: differences in therapist effectiveness, inconsistencies in uses of manualized treatments, the placebo effect where expecting that therapy will benefit you actually drives benefits, and the self-limiting nature of many mental woes. (Many people don’t stay depressed for long, even without intervention). These complexities combine with the difficulty of teasing out underlying mechanisms to make it almost impossible to know what really drives the benefits of counseling.

For some of the same reasons, we also don’t have good evidence to say that one particular counseling approach, e.g., cognitive behavioral therapy works better than another, e.g., acceptance and commitment therapy whether overall or for specific concerns. While it hasn’t suffered the same crisis of confidence as social psychology, the research base of talk therapy may be just as shaky. I say that as a trained PhD-level statistician and research methodologist who is all too aware of how questionable much research evidence is today.

Implications for AI in digital mental health

It seems that if you put all your chips in on the empirically-supported treatments model, you could possibly argue for eliminating the human from it, since EST emphasizes unique features of each treatment, and how they combine together to make an effective evidence-based intervention. The EST model seems, to me, more favorable to prospects for an all-AI therapist.

I wonder how much the human element matters. To what extent does a one-to-one relationship with a counselor or coach drive someone’s willingness and commitment to take on health promoting actions? And if this is a critical component of how therapy works, could an AI provide empathy, positive regard and affirmation, and a therapeutic alliance? I am skeptical.

The answer to whether specific elements of treatment approaches or common elements such as the therapeutic relationship matter most for creating improvement. The likely answer: it’s not either/or. It’s both.

And I remind myself that it is possible to build a chatbot that is emotionally engaging. In fact, the first therapybot ELIZA was known for its ability to make people feel validated.

Interesting historical note: the common factors model was first proposed in 1936 by US psychologist Saul Rosenzweig. He wrote:

Given a therapist who has an effective personality and who consistently adheres in his treatment to a system of concepts which he has mastered and which is in one significant way or another adapted to the problems of the sick personality, then it is of comparatively little consequence what particular method that therapist uses.

If you’re interested in staying up to date with my explorations about how we can use AI/ML to improve mental health and wellness, check out my new newsletter Emotion / Know.