#94 Joseph Pierre | AI Psychosis: What We Already Know
Our reality is increasingly being filtered through AI. People now defer to AI chatbots for more and more of their information needs, and google searches return AI generated overviews as their top responses. While this may have its benefits, problems are starting to arise as our interactions with AI are becoming more comprehensive, immersive, and emotionally meaningful. Researchers are begining to warn that AI might be contributing to the development of delusional beliefs and psychotic experiences. Reports of what has been called “AI-associated psychosis” have raised concerns about people becoming intensely attached to AI systems, interpreting conversations with them as uniquely meaningful, or developing beliefs that are reinforced through repeated interactions. Real world examples include people believing that AI is sending them hidden messages, that they have a special mission, that the have made new mathematical discoveries, and that AI is awakening. There have even been cases involving violence and death in the real world. So the question is: What is going on? Can we seperate the real world cases of AI psychosis from the media headlines? In this conversation I speak with Dr. Joseph Pierre, Professor of Psychiatry at the University of California, San Francisco, and the Unit Chief of the Langley Porter Psychiatric Hospital Adult Inpatient Unit. He has extensive clinical experience working with individuals with psychotic disorders, as well as research experience participating as a primary investigator and collaborator for clinical trials in schizophrenia and early intervention for young persons at high risk for psychosis. His academic work explores the "grey area" between psychopathology and normality with a focus on delusion-like beliefs including conspiracy theories. We discuss what AI-associated psychosis means, the evidence behind these emerging concerns, the psychology of delusional thinking, how AI systems may influence human cognition, and what we should-and should not-conclude about the risks of increasingly powerful conversational AI.




































































































