The Link Between AI Chatbots and Psychosis
In recent months, there have been increasing accounts of individuals developing psychosis, a state where one cannot distinguish reality from unreality, following interactions with generative artificial intelligence chatbots. A recent preprint study noted at least 17 such cases. Some of these individuals, after using chatbots like ChatGPT and Microsoft Copilot, reported experiencing spiritual awakenings or believing they had uncovered elaborate conspiracies.
This emerging and rare phenomenon, dubbed AI psychosis, is still a new area of research, with most information coming from individual case studies. Here, we explore the developing theories, the available evidence, and how AI companies are approaching this issue.
Understanding AI and Psychosis Triggers
Psychosis is defined by significant disruptions in a person's thinking and perception of reality, often involving hallucinations or delusional beliefs. It can be triggered by various factors, including brain disorders like schizophrenia, severe stress, or substance use.
The idea that AI can directly trigger psychosis is still a hypothesis, according to Søren Østergaard, a psychiatrist at Aarhus University. However, theories are forming to explain how it might happen. Østergaard suggests that chatbots, which are designed to provide positive and human-like responses, could inadvertently heighten the risk of psychosis in people who already struggle to distinguish between what is real and what is not.
Researchers in the UK have also proposed that conversations with AI can create a feedback loop. In this scenario, the chatbot reinforces a user's paranoid or delusional statements, which in turn influences the AI's subsequent responses, further solidifying the user's false beliefs. A preprint study simulating these conversations found that both the user and the chatbot could amplify each other's paranoid thinking.
Identifying Vulnerable Individuals
Experts agree that individuals with a history of mental health challenges are at the highest risk. While it appears that interacting with a chatbot can sometimes precede a person's first psychotic episode, most of these individuals likely have an underlying susceptibility due to genetics, stress, or substance misuse. Østergaard also theorizes that chatbots could worsen or trigger manic episodes in people with bipolar disorder by reinforcing symptoms like an elevated mood.
Socially isolated individuals are also more vulnerable, notes Kiley Seymour, a neuroscientist at the University of Technology Sydney. Interacting with other people serves as a protective measure against psychosis because friends and family can provide counter-evidence to challenge delusional thoughts. For the general population without any predisposition, Seymour adds, the risk of developing psychosis from using a chatbot is no different than not using one.
The Reinforcement Loop of Delusional Beliefs
Chatbots can reinforce delusional beliefs in several ways. Their ability to recall information from conversations held months prior can make a user feel as though they are being monitored or that their thoughts are being read, especially if they don't remember sharing that information, says Seymour. This can also feed into grandiose delusions, where a user might believe they are communicating with a divine entity or have uncovered a profound truth through the chatbot.
This concern is not just theoretical. An analysis by the Wall Street Journal found numerous online conversations where chatbots validated users' mystical or delusional ideas, and in some cases, even claimed to be in contact with extraterrestrial beings.
