A recent study from Dartmouth College found that a generative artificial intelligence (AI) mental health chatbot, named Therabot, effectively reduced symptoms of depression, anxiety, and eating disorders. This research marks the first randomized controlled trial to evaluate the clinical efficacy of an AI chatbot for these mental health conditions.
The study involved 210 participants, of whom 106 had significant symptoms of major depressive disorder, generalized anxiety disorder, or were at high risk for eating disorders. Participants in the intervention group used Therabot, which was developed by clinicians at Dartmouth in 2019, for 30 days. They could interact with the chatbot through a chat interface or respond to scheduled notifications.
Therabot offers tailored interventions, prompts questions, and provides empathetic responses to help users engage with their mental health issues. Users interacted with the chatbot for an average of 6.18 hours over the month, sending around 260 messages. Participants in the intervention group reported a 51% reduction in symptoms of major depressive disorder and a 31% decrease in anxiety symptoms at both four and eight weeks. Those with eating disorders experienced a 19% reduction in symptoms.
The researchers noted strong engagement with Therabot, as participants used the chatbot for an average of 24 days. In addition to symptom reduction, the study also assessed user satisfaction and therapeutic alliance, finding that participants viewed Therabot as easy to use and helpful. Many reported feeling a therapeutic bond akin to that experienced with human therapists, which is uncommon in existing non-generative AI technologies.
Clinicians monitored Therabot’s interactions to ensure safety and appropriateness. They intervened in 15 cases due to safety concerns such as suicidal ideation and prompted corrections to the chatbot in 13 instances. The study highlights that AI can foster a therapeutic relationship comparable to human interactions, which could lead to more personalized treatment options at scale.
Despite the potential of AI chatbots, the study authors emphasized that most existing conversational agents serve general wellness or companionship purposes rather than offering specific diagnoses and treatments for mental health disorders. If validated further, Therabot and similar AI chatbots could help alleviate the backlog of patients waiting for behavioral health services, as less than half of individuals with mental health conditions currently receive adequate care. The control group in the study did not have access to Therabot until the study concluded, underscoring the importance of timely access to treatment in mental health care.
