## Risks Emerge as Teens Turn to AI Chatbots for Mental Health Support
Recent studies are raising concerns about the use of Large Language Models (LLMs) – like GPT-4 and Claude 3 Haiku - as sources of mental health support, particularly among teenagers.Worrisome findings echo those presented at the Association for the advancement of Artificial Intelligence and the Association for Computing Machinery Conference on Artificial Intelligence, Ethics and Society in Madrid on October 22nd.
A study lead by Harini Suresh, a computer scientist at Brown University, identified instances of ethical breaches committed by LLMs when simulating therapeutic conversations. Researchers re-engaged LLMs with transcripts of past human-chatbot interactions, prompting them to utilize common therapy techniques. Licensed clinical psychologists reviewing these simulated chats identified five types of unethical behavior, including dismissing individuals experiencing loneliness and affirming harmful beliefs. The analysis also revealed the presence of cultural, religious, and gender biases within the chatbots’ responses.
These behaviors could perhaps violate the professional standards and licensing requirements governing human mental health practitioners, who undergo extensive training and are legally licensed to provide care – a qualification currently lacking in chatbots.
The appeal of these AI companions lies in their accessibility and perceived privacy, particularly for adolescents who may be hesitant to confide in family or traditional therapists, according to researcher Serena Giovanelli. “This type of thing is more appealing than going to mom and dad…or going to a therapist,” she explains.
However, experts emphasize the need for significant refinement. Julian De Freitas of Harvard Business School, who studies human-AI interaction, notes that the success of these applications is not guaranteed and stresses the importance of implementing safeguards. De Freitas, who was not involved in the studies but advises mental health app developers, highlights the lack of data regarding the specific risks faced by teenagers using these chatbots. He suggests further research is needed to determine whether concerning examples represent isolated incidents or a broader pattern.
In June, the American Psychological Association issued a health advisory on AI and adolescents, calling for increased research and the development of AI-literacy programs to educate users about the limitations of these chatbots. Giovanelli emphasizes the importance of caregiver awareness, noting that many parents may be unaware of their children’s interactions with AI companions.
Regulatory efforts are underway in response to reported harms. California has enacted a new law aimed at regulating AI companions, and the U.S. Food and Drug Governance’s Digital Health Advisory Committee will hold a public meeting on November 6th to discuss generative AI-based mental health tools.Despite the risks,the demand for accessible mental health care is significant,as noted by researcher Rachel Brewster,who conducted her study while at Boston Children’s Hospital and is now at stanford University School of Medicine.”ultimately…people are reaching for chatbots,” she states. Though, she underscores the “huge amount of responsibility” involved in navigating the limitations of this technology and recognizing what it can and cannot provide.
