Can new research improve the safety of AI therapy for teens?

The University of Cincinnati's Stephen Rush joined WVXU's Cincinnati Edition for a discussion on the growing trend of young people turning to artificial intelligence (AI) for mental health support and therapy.

Rush noted there are structured AI chatbot tools specifically designed by mental health professionals that function "more like digital coaching programs," but many teens turn to generative AI tools like ChatGPT that are designed to create responses in real time based on learned patterns.

"The caveat though is that generative AI is built to produce agreeable responses that are validating to the people using the service," said Rush, MD, associate professor of clinical psychiatry in UC's College of Medicine, medical director of ambulatory services and a UC Health physician. "So the chatbot experience is like working with a GPS, and the generative AI experience feels more like having a passenger in the seat telling you where to go who can also say anything else that they are so inclined to say."

Recently, families have testified before Congress asking for tougher regulations after their children died by suicide following the use of AI chatbots for therapy. Rush stressed that most generative AI tools are not equipped like a professional to respond to a crisis.

"There is to my knowledge one model that can do some risk stratification in terms of crises," he said. "But they’re not trained, nor do they have the credentialing, to deal with crises that might involve safety to one’s person or safety to somebody else."

At the same time, access to mental health services remains a significant hurdle, particularly with a scarcity of licensed youth mental health supports. Rush said the World Health Organization is currently working to expand access to mental health care in low- and middle-income countries through the use of AI chatbots.

"They would describe AI as a tool, but not a replacement for therapists," Rush said. "A therapist brings something to the conversation that an artificial intelligence can’t, and that’s things like real time emotional understanding, the ability to notice subtle signs of distress and the ability to step in when somebody is a danger.

Listen to the Cincinnati Edition segment.

Featured image at top: iStock Photo/vertigo3d.