Dr. AI Will See You Now: The Opportunities, Challenges, and Risks of ChatGPT, Gemini, and Other Large Language Models in Psychiatry

The rapid advancement of artificial intelligence (AI) over the last decade has led to dramatic innovation. Over the past two years, a novel generation of AI models, known as large language models (LLMs), have gained prominence due to their impressive capabilities in processing and interacting with natural language. These models have particular potential in the field of psychiatry, given the primacy of language in psychiatric assessment and treatment. However, considerable concerns exist regarding these models’ potential for misinformation, bias, and other forms of harm. Here, we present an introduction to AI and LLMs, an overview of current use cases within behavioral health, in-depth discussions on some of the latest investigational research on LLMs in psychiatry, and a robust panel discussion on their risks and opportunities within our field.

First, we will provide an orientation and introduction to the concepts of AI and LLMs, with definitions of frequently used AI-related terminology. We will then present an overview of how LLMs are being used in healthcare, with a focus on behavioral health, including risk assessment, diagnosis and monitoring, documentation, and workflow tools. We will then present an overview of current investigational applications of LLMs for psychiatry. A straw poll will be used to assess the audience’s comfort with and use of LLMs and other AI tools in their day-to-day clinical practice.

Following the introduction, the expert panelists will provide a series of in-depth explorations of recent research in the use of LLMs for psychiatry. Dr. Kaitlin Hanss, a psychiatrist and AI researcher at UCSF, will present the latest work in assessment of the breadth of psychiatric knowledge encoded in state-of-the-art LLMs, their ability to assess the risk of uncertainty in this knowledge base, and the sources of errors made during question-answering. Dr. Karthik V Sarma, a psychiatrist and AI researcher at the Bakar Computational Health Sciences Institute, will share the latest work in harnessing LLMs for psychiatric reasoning and diagnosis, including an in-depth comparison of performance across the models released by the five largest vendors. Dr. Marina Tolou-Shams, a child psychologist and digital health equity researcher at UCSF, will present ideas for how youth-serving systems, such as juvenile probation and child welfare, can use LLMs and other AI tools to enhance youth behavioral health screening, assessment and treatment referral practices to increase access to care. Dr. Isaac Galatzer-Levy, a senior staff research scientist focusing on LLMs in behavioral health at Google, will present on the strengths and limitations of LLMs in clinical reasoning, focusing on advanced applications for LLM interactions with patients. Finally, the panelists will hold an interactive discussion, including Q&A with the audience, focused on the opportunities, challenges, and risks posed by LLMs in psychiatric practice.