The Integration of Artificial Intelligence-Powered Psychotherapy Chatbots in Pediatric Care: Scaffold or Substitute?
In April 2024, the United States Food and Drug Administration approved the first digital application to treat major depression in adults 22 and older.1 The app—Rejoyn—joins a growing list of artificial intelligence (AI)-based platforms designed to treat mental illness.2 These tools range from chatbots to gamified cognitive behavioral therapy (CBT), to machines that emulate human therapists. Given the significant barriers to accessing mental health care, these technologies have been pitched as a means of addressing the current mental health crisis among youth.3 Although there is a growing literature exploring the ethical implications of the use of AI in mental health care, their use in pediatric populations remains underexplored.4-7
This commentary explores the use of large language model (LLM)-based digital applications in pediatric mental health care, with the goal of beginning to map this terrain. LLMs are a form of AI designed to generate human-like responses to natural language prompts. These models use deep learning techniques, specifically transformer architectures, to analyze patterns in language and create coherent, contextually relevant responses. We will refer to LLM-based “conversational agents” colloquially as “therapy bots” or “AI chatbots” interchangeably throughout. These are systems that use LLMs, or other advanced natural language processing systems, to simulate the utterances of a human therapist as part of a conversation with a pediatric patient.
After providing a brief overview of the basics of different therapy bots, we turn to how their use intersects with accepted standards in pediatric ethics and practice. First, we argue that the use of these AI chatbots not only fails to satisfy several interests children have with respect to their mental health, but also could potentially constitute a setback to those interests. Second, we argue that insufficient attention has been paid to the role of others in supporting children who are experiencing mental health challenges, and how this important social context intersects with therapy bots. Third, we argue that who is authorizing “treatment” decisions is obscured by these technologies, undermining responsibility and shared decision-making processes in pediatrics. Finally, these technologies could further entrench existing inequities in pediatric mental health care. We argue that, together, these considerations cast doubt on the idea that therapy bots will be “better than nothing” in the context of pediatric mental health care. We will use the terms ‘child’, ‘children’, and ‘pediatrics’ throughout to refer to any individual under the age of 18 and the care provided to that population.
-with support from Working Group, DH11: AI and Human Values