Key Takeaways |
|
Proactive Disclosure Inquiries: Patients may hide AI use due to fear of stigma, making it essential for clinicians to ask directly and incorporate AI-use questions into intake and sessions. Identifying Care Gaps: A patient's reliance on a chatbot for emotional support often signals an unmet need in their current care that clinicians can address. Diverse Risk Factors: Beyond the risk of anthropomorphizing AI, clinicians must consider other potential harms, including OCD reassurance checking, manic thought elaboration, and unpredictable product-level issues like data privacy changes. |
As co-facilitators and aspiring mental health clinicians, we wanted to hear directly from practicing clinicians about what they are witnessing. Generative AI is now clinically meaningful for many patients: about a third of clinicians report that patients use AI for mental health support [1], yet no protocol exists to recognize or respond to related harm. Research and clinical practice need to move together, but patients aren't waiting for the research. Clinicians need to be prepared to encounter naturalistic AI use in their patients right now, not once the evidence base catches up.
That's what inspired and shaped the model of this collaborative; a virtual training that grounds clinicians in how AI works, paired with a case-based learning curriculum that activates clinical reasoning through real discussion. Here's what came out of it.
Note: Cases discussed each week were hypothetical composites, not real patients.
Patients may stay quiet about their AI use out of fear of stigma. Clinicians need to ask directly, in an open and non-judgmental way, and consider building AI-use questions within their sessions and/or at intake.
When a patient leans on a chatbot for support, that's evidence of a gap in their care that a clinician may be able to address.
Treating AI like a person, especially for younger, developmentally vulnerable users, repeatedly emerged as a potential risk factor for maladaptive AI use. However, plenty of clinical harm could happen even when patients are using AI "just as a tool," from OCD reassurance checking to manic thought elaboration.
Because these tools are products owned by external companies, patients are exposed to risks beyond the clinical relationship, from data practices to sudden changes in how the product behaves.
Some clinicians have considered bringing the chatbot directly into the session. If a patient is open to it, this could be an effective way to explore the patient’s relationship with AI.
Every case discussion brought up conflicting therapeutic approaches amongst experienced clinicians. This begs the question: how can clinicians and the digital mental health field as a whole move forward? As co-facilitators, we were inspired and created several cases due to the conversations amongst clinicians during week one. This is why learning collaboratives are so crucial. Conversation is a key aspect of education, which is why education is one of the three SODP pillars to move digital mental health from research to routine care [2].
No single clinician, training program, or institution is going to resolve these questions alone. SODP wants to take this conversation further, with hospitals, mental health centers, clinical groups, and training programs. Interested in partnering to host a learning collaborative? Reach out to SODP co-leader Kathryn Taylor Ledley at kledley@bidmc.harvard.edu.
References
Subscribe Now