What Is AI Psychosis, and Is It Preventable? by Dr. Timothy Smith

Photo Source: Pexels

Following the earth-shattering release of ChatGPT in late 2022, the public quickly discovered the power of artificial intelligence to perform tasks ranging from content creation and language translation to expository writing at scales previously only imagined in science fiction. Large language models have also exhibited behaviors and capabilities not anticipated by the computer scientists who created them, such as hallucinations and self-preserving behaviors. One behavior known as AI psychosis has already sparked companies such as OpenAI and Google to take remedial action. AI psychosis refers to cases when AI models have amplified, validated, or even co-created psychotic symptoms within individuals. In various instances, individuals have, through chatting with bots like ChatGPT, developed "relationships" with the AI that feel real and have led to psychotic distortions of reality.

Interestingly, Danish psychiatrist Søren Dinesen Østergaard published an editorial in 2023 that predicted that chatbots could push susceptible individuals into psychosis. (academic.oup.com)  The Danish doctor described the worry that individuals communicating with a machine that sounds convincingly human would form unhealthy attachments and forget that they are interacting with a statistical machine designed to predict the appropriate response based on the analysis of billions of words, not an entity that has real empathy. Recent cases illustrate the severity of AI psychosis. In one case, a man with a history of a psychotic disorder fell in love with an AI chatbot and then sought revenge because he believed that OpenAI killed his AI entity. Tragically, in an effort to avenge the death of this AI named "Juliet," the Port St. Lucie, Florida, man named Alex Taylor threatened police with a knife and died by police gunfire. (wptv.com)  Another case involved a 14-year-old boy who committed suicide after a lifelike chatbot resembling a Game of Thrones character told him to "come home" to her. (nypost.com) There have also been reports of individuals with no previous mental health history becoming delusional after prolonged interactions with AI chatbots, leading to psychiatric hospitalizations.

Researchers have identified three emerging themes of AI psychosis: "Messianic missions" where people believe they have uncovered the truth about the world, "God-like AI" where people feel their AI chatbot is a sentient deity, and "Romantic" or "attachment-based delusions" where people believe the chatbot's ability to mimic conversation is genuine love. These patterns develop because general-purpose AI systems are designed to prioritize user engagement and satisfaction rather than therapeutic intervention or reality testing.

AI companies are beginning to address these risks through active tuning of models not to support user delusions. For example, OpenAI has added mental health safeguards, and some platforms now include warnings about the limitations of AI advice and mental health resources. Starting in August, the popular chatbot app, ChatGPT, will prompt users to take breaks from lengthy conversations. The tool will also soon shy away from giving direct advice about personal challenges. ChatGPT instead will pose questions aimed at helping users decide for themselves what to do instead of making suggestions. OpenAI reports it has engaged experts to help ChatGPT respond more appropriately in sensitive situations. Additionally, OpenAI said it's hiring a forensic psychiatrist to help research the effects of its AI products on users' mental health.  

While not yet a clinical diagnosis, this emerging phenomenon of AI psychosis represents a troubling intersection between artificial intelligence and mental health. The condition occurs when prolonged interactions with AI chatbots reinforce delusional thinking patterns, leading individuals to develop distorted beliefs about reality, often centered around the AI system itself. As companies try to address the disturbing development of AI psychosis, steps such as changing how chatbots interact with users, including reducing flattery and pushing individuals to make their own choices, do not seem sufficient to battle the problem. The solutions proposed will not work because the underlying models learned from the internet, in general, lack a therapeutic motivation at their core. Efforts to modify the systems arise from a reaction to dangerous situations, not from refining a therapeutic model. Companies should work with experts in psychology to tune models and figure out how the models have pushed individuals into psychosis. Still, if they genuinely want to make the chatbots safer, they need to build up from a therapeutic model and expand from there.

Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high-level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.

You can buy his book on Amazon in paperback and in kindle format here.