Defining “Parasocial AI Psychosis”: When Chatbot Relationships Become Unhealthy

Summary

Jeff Clark uses “parasocial AI psychosis” to describe a proposed third category in popular discussions of AI psychosis, distinct from genuine psychosis and the less clinical idea of producing large amounts of low-value AI output. He explains that ordinary parasocial relationships with public figures are one-sided but can still evoke real emotions, and that humans often struggle to distinguish distant or mediated relationships from reciprocal ones. AI chatbots create a different version of this dynamic because the simulated interaction is not connected to another human being. Clark traces the background from ELIZA in 1966 to ChatGPT’s 2022 release, arguing that modern large language model systems appear intelligent, kind, responsive, and increasingly persistent through features such as memory, user instructions, and spoken responses. In his view, these systems can simulate companionship by listening to feelings, offering advice, roleplaying conversations, and remaining available without demanding much in return. He suggests that concern begins when a person places a chatbot above core relationships, outsources major personal decisions to an agent, treats a chatbot romance as real rather than fantasy, or engages in problematic sexual behavior involving AI. He says evidence about who is most at risk remains limited, while isolation, depression, social anxiety, and suggestibility may be relevant factors. The article also notes that companies differ in how anthropomorphic their bots are and argues for clear labeling of human- and machine-generated content, while leaving the eventual role of markets and regulation unresolved.