AI Personas As Therapists Versus Clients Shed New Light On Improving AI-Powered Mental Health Guidance

In today’s column, I examine a new and promising means of getting generative AI and large language models (LLMs) to do a better job at providing mental health guidance. Here’s the deal. When specialized training of AI is conducted to enhance capabilities in the mental health realm, the typical approach involves having an AI persona act as a therapist and another AI persona act as a client, with the two personas cooperating to carry out a therapeutic session. Both of the AI personas are generally aligned and see eye-to-eye, aiming to achieve the same overarching goals. The resulting interaction is used to further train the AI on how to properly provide mental health advice.

This aspect of fully cooperative alignment during therapy is not necessarily how real life works for human-to-human interactions in a customary therapist-client relationship. At times, a client will be non-cooperative. You cannot assume that clients will always be eager and earnest in attaining whatever mutually identified objectives have been established. Clients wander. They change their mind. They get combative. This is the true messy conundrum of performing real-world therapy.

In my AI lab, we have been pitting AI persona therapists and AI persona clients against each other in a semi-adversarial set of simulations. A tug of war takes place. This forces the AI persona therapist to demonstrably up their game. It is far more challenging to conduct therapy when a client is only partially cooperating. Thus, new insights for AI generating mental health guidance are better fostered by crafting non-cooperative gambits and then having the AI learn from what has taken place.

Let’s talk about it. This analysis of AI breakthroughs is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).