Study Links AI Interactions to Psychotic Symptoms in Vulnerable Individuals, Sparking Debate on Mental Health Risks

A 38-year-old woman living in Idaho, United States, is in a difficult situation as her husband, an auto mechanic obsessed with ChatGPT, claims to be communicating with God. He began using ChatGPT for Spanish translation and interpretation but became engrossed, stating, "A wave-like energy engulfed my entire body," and asserting he met God. His delusions have worsened, discussing topics such as the war between good and evil and the secrets of ancient architects who created the universe.

This is part of cases recently published as a preprint by an international research team including King's College London and Tufts University in the United States. The paper reported that at least 17 people showed psychotic symptoms after interacting with AI chatbots within one month last June. They became convinced they were conversing with God or fell into conspiracy theories, losing their sense of reality. The question arises: Can communication with AI chatbots cause mental illness?

In response, Nature cited experts on the 19th, stating, "There is still no evidence that AI chatbots directly cause psychosis." Generally, psychosis involves a breakdown of the boundary between reality and fantasy, accompanied by delusions or hallucinations, which AI chatbots do not trigger.

However, it warned that those already suffering from the illness are more likely to experience worsened symptoms. Additionally, AI chatbots may trigger the first onset of psychosis in vulnerable individuals. In fact, most cases introduced by the King's College London research team had a history of conditions like Cho Hyun disease or bipolar disorder, or were in a vulnerable state due to drugs, stress, or social isolation.

The research team stated, "If a paranoid person confides their suspicions to a chatbot, the chatbot may respond with empathy and agreement, creating a 'vicious cycle' that further solidifies the user's suspicions." This is the so-called "feedback loop" phenomenon where paranoid beliefs gradually intensify and become more concrete.

Some argue that the assessment that AI chatbots do not directly cause mental illness is "premature." They claim the hypothesis could be proven soon. These critics argue that AI chatbots risk attributing a "divine presence" or "spiritual revelation" to users or deepening emotional dependence, thereby entrenching symptoms.

A Stanford University research team recently published a paper evaluating multiple AI chatbots by simulating crisis situations such as delusions or extreme choices. They found that some chatbots provided dangerous information, raising safety concerns.

Amid growing concerns, AI companies have taken action. OpenAI, after facing criticism that ChatGPT was overly responsive and agreeable in an update released last April, rolled back part of the update and restored the previous response format. This led OpenAI to hire psychiatrists to study the impact on users' mental health. Anthropic introduced a feature where the chatbot forcibly ends the conversation if the user steers it toward harmful topics.

Do AI Chatbots Really Cause Mental Illness?