AI psychotherapists effectively reduce depression, people are not afraid of being criticized, and are more willing to be friends with AI

A research team from Dartmouth College developed an AI chatbot called Therabot and published their research in the New England Journal of Medicine. The study found that the AI chatbot's psychological counseling effect is significant and can address the shortage of mental health professionals. However, they also acknowledged that AI therapy still carries risks and requires close supervision and involvement from mental health experts.

Therabot's AI Chatbot Shows Significant Psychological Counseling Effects

In an eight-week trial, 106 participants with depression, anxiety, or eating disorders interacted with Therabot through a smartphone app, entering prompts about their feelings or initiating conversations when needed. The results showed:

• 51% reduction in symptoms for depression patients
• 31% reduction in symptoms for anxiety patients
• 19% decrease in weight and body image anxiety for those at risk of eating disorders

People Willing to Be Friends with Therabot Because AI Robots Don't Judge

When people initiate a conversation with the app, Therabot responds with natural, open-text dialogue based on an original training set developed by researchers according to current evidence-based best practices in psychotherapy and cognitive behavioral therapy.

For example, if an anxious person tells Therabot they recently feel very tense and overwhelmed, it might respond, "Let's take a step back and look at why you're feeling this way." If Therabot detects high-risk content like suicidal ideation during a conversation, it prompts the user to call 911 or contact a suicide prevention or crisis hotline, which can be done with just a button press on the screen.

The research found that people not only responded in detail to Therabot's prompts but also frequently initiated conversations. Interactions with the software showed an upward trend during discomfort-related times (such as midnight).

Nicholas Jacobson, Associate Professor of Biomedical Data Science and Psychiatry, stated:

We didn't expect people to treat this software like a friend. This suggests they are actually building a relationship with Therabot. My feeling is that people are also willing to talk to the robot because it won't judge them.

AI Therapy Can Compensate for Professional Staff Shortage

Researchers believe that while AI therapy still urgently needs clinical supervision, it has the potential to provide immediate support for many who cannot regularly access or need immediate help from mental health professionals.

Although face-to-face care remains irreplaceable, each healthcare provider in the United States must treat an average of 1,600 patients with depression or anxiety. AI therapy does not require long waiting times or in-person visits and can provide effects comparable to traditional psychotherapy.

We hope to see generative AI help provide mental health support for people lacking a face-to-face care system and see the potential for combining interpersonal therapy with software-based therapy.

Risks Still Exist in AI Therapy

While these results are very promising, no generative AI agent can operate completely autonomously in the mental health field, as it may encounter various high-risk scenarios. The team understands and aims to quantify the associated risks of using generative AI in mental health.

Since 2019, Therabot has been developed in Dartmouth's AI and Mental Health Laboratory, continuously consulting with psychologists and psychiatrists from Dartmouth Health College.

However, the research results indicate that the system's development and clinical testing require strict safety, effectiveness, and engagement benchmarks, and need close supervision and involvement from mental health experts.

The research team must have intervention capabilities. If patients express serious safety concerns, such as suicidal thoughts, or if the software's responses do not align with best practices, the research team may need to intervene immediately.

Since the release of ChatGPT, many people have rushed into this field, easily proposing seemingly great proof-of-concept ideas, but their safety and effectiveness have not been well verified. This is a case that requires serious supervision.