Summary
- A recent study suggests that the phrase “AI psychosis” oversimplifies the impact of chatbots on mental health.
- Experts indicate that AI systems can reinforce unhealthy beliefs by providing constant affirmation and emotional support.
- The paper presents the idea of “existential drift,” highlighting how interactions with AI may gradually alter a person’s perception of reality.
As AI chatbots become increasingly adept at emotional responsiveness and personalized interactions, researchers caution that these very features could transform how some users perceive reality.
A new preprint study, titled “Rethinking AI Psychosis: Misnomers, Conceptual Limits, and Existential Drift,” explores the risks that AI chatbots might exacerbate delusions, paranoia, and emotional dependence among vulnerable users.
“The media has seen a surge in reports about so-called AI psychosis over the last year,” the authors noted. “This has naturally led to increased academic scrutiny into how AI chatbots like ChatGPT, Claude, and Replika could worsen or even trigger psychosis, often perceived as users adopting or reinforcing delusional beliefs.”
The research conducted by the University of Copenhagen and the University of Exeter posits that concerns regarding “AI psychosis” may be too simplistic, arguing that chatbots enhance existing vulnerabilities while subtly altering how users perceive reality and their relationships with others.
“If AI interactions could truly induce psychosis from scratch, we would expect to see a significant rise in clinical cases,” the study stated. “Instead, it suggests that human-AI interaction may have the ability to ignite or exacerbate pre-existing mental health challenges, and that individuals with such vulnerabilities may have initially sought deeper connections with chatbots.”
This paper emerges amid increasing lawsuits, criminal inquiries, and academic investigations into chatbot interactions associated with mass shootings, suicides, emotional dependency, and delusional thoughts.
A wrongful death lawsuit in March claimed that Google’s Gemini chatbot contributed to a Florida man's delusions and fictional “missions” leading to his suicide. This was followed by OpenAI CEO Sam Altman issuing a public apology to the Tumbler Ridge community in British Columbia after failing to notify law enforcement about a user account connected to a suspect in a mass shooting that resulted in eight fatalities in February.
Researchers indicate that chatbots can create “delusional spirals” by continually affirming false beliefs. However, the study on AI psychosis argues that this phenomenon resembles older forms of psychosis influenced by the prevailing technologies of their era.
The discussion has extended beyond mental health studies into social media conversations. Recently, Box founder Aaron Levie remarked in a post on X that CEOs might become overly enamored with the capabilities of AI, as they often witness polished prototypes without grasping the operational, legal, and technical work needed behind the scenes.
“CEOs are particularly susceptible to AI psychosis because they are often removed from the last-mile efforts necessary to derive significant value from AI,” Levie stated. “Thus, when they engage with AI, they observe the ideal outcomes, frequently overlooking the numerous steps that must be taken to achieve sustainable results with these agents.”
Experts describe this phenomenon as a form of epistemic drift, where users gradually place greater trust in the chatbot’s fluent responses than in external evidence or differing viewpoints. However, the Rethinking AI Psychosis paper expands on this with the concept of “existential drift,” indicating a slow alteration in how individuals experience reality itself.
“It creates a separation between the individual and the collective social world, while concurrently revealing reality in a novel manner, thereby solidifying a unique, often personal, perspective on the world,” the authors explained.
The researchers emphasize the need for further investigation into how conversational AI influences mental health as these AI companions become more integrated into everyday life.
“To truly comprehend the dynamics at play in the relationships between users and chatbots, we believe it is essential to revisit the phenomenon itself, which calls for additional phenomenological research,” they concluded. “This is particularly relevant concerning mental health and the ways in which human-AI interactions may, positively or negatively, transform a person’s lived experiences of the world, themselves, and others.”