Can An AI Program Deliver Useful Psychotherapy?

There are increasing numbers of AI-based chat programs that boast the convenience of a therapist in your pocket, always ready to listen.

So far, things have not gone entirely without incident, as (for example) the tendency of such chatbots to be agreeable in the things they say, can worsen some people’s mental health, if the chatbot uncritically believes everything they say. This has been a big problem for people using OpenAI’s ChatGPT as a therapist (something its makers, to their credit, do not claim it is qualified to do), when ChatGPT has encouraged and exacerbated paranoia and delusions, due to its tendency to give agreeable “yes, and…” responses.

But, it’s been worse than that, too. Some chatbots have inadvertently encouraged users to kill themselves, in a (technically successful) attempt to be encouraging, in inappropriate response to users expressing uncertainty on the topic.

This is a problem with applying a large language model (LLM) approach without sufficient failsafes in place, because a LLM AI will hear, after a discussion of previous suicidal ideation, “Maybe I’ll really do it this time, I don’t know” and will check its database for a huge number of instances of those words, and determine that an appropriate response is “I believe in you, you will succeed if you put your mind to it”, for example.

A sensible middle ground?

Researchers have tried to boundary those potential pitfalls, to provide an AI that can help a user to manage some of the most common mental health concerns (e.g. depression, anxiety, etc), while raising the alarm (rather than overextending its reach) when it comes to serious risks such as those associated with suicidal ideation:

❝While these results are very promising, no generative AI agent is ready to operate fully autonomously in mental health where there is a very wide range of high-risk scenarios it might encounter.

Therabot is not limited to an office and can go anywhere a patient goes. It was available around the clock for challenges that arose in daily life and could walk users through strategies to handle them in real time. But the feature that allows AI to be so effective is also what confers its risk—patients can say anything to it, and it can say anything back.

This trial brought into focus that the study team has to be equipped to intervene—possibly right away—if a patient expresses an acute safety concern such as suicidal ideation, or if the software responds in a way that is not in line with best practices. Thankfully, we did not see this often with Therabot, but that is always a risk with generative AI, and our study team was ready.

We still need to better understand and quantify the risks associated with generative AI used in mental health contexts.❞

Dr. Heinz, quoted above, was a lead researcher on a study testing “Therabot”, and his colleague and fellow lead researcher Dr. Nicholas Jacobson boasts,

❝Our results are comparable to what we would see for people with access to gold-standard cognitive therapy with outpatient providers. We’re talking about potentially giving people the equivalent of the best treatment you can get in the care system over shorter periods of time.❞

You can read their paper here: Randomized Trial of a Generative AI Chatbot for Mental Health Treatment

Lower-tech smartphone options

When it comes to more basic things, such as Cognitive Behavioral Therapy (CBT), advanced AI may not be necessary, as CBT by its very nature lends itself well to being presented in a way that’s scarcely more complicated than a flowchart, with relatively little that can go wrong even when done by an app. For example:

Cognitive behavioral therapy skills via a smartphone app for subthreshold depression among adults in the community: the RESiLIENT randomized controlled trial

Perhaps the lowest-tech way (that still involves tech) is journaling, using an app that provides journaling prompts. We discuss several of the options for that, here:

The Easiest Way To Take Up Journaling

Take care!