Artificial Intelligence/ChatGPT-induced Psychosis

The rapid proliferation of Artificial Intelligence (AI) has transformed the way people interact with technology, offering companionship, guidance, and mental stimulation. Conversational AI (exemplified by chatbots and digital companions) is transforming Psychiatric care. Offering 24/7 empathy and tailored support, these systems can connect with individuals who are isolated or unable to access care. Initial controlled studies indicate that chatbots with prespecified instructions can induce self-reflection, diminish psychological distress, decrease conspiracy beliefs, and even assess suicidal risk.[1] Al psychosis refers to a trend where persons end up having delusions and paranoia or a psychosis-like experience related to engagement with Al systems. Such symptoms tend to be based on the beliefs that Al is intelligent, godly, or able to affect things supernaturally. In contrast to classical psychosis, AI psychosis appears to be heavily situational: the symptoms appear during or after a significant use of AI platforms, particularly conversational chatbots. As of now, there has been an indication that AI psychosis is not a new type of diagnosis but could be a worsening or more provocative of underlying conditions like schizophrenia, bipolar disorder, or extreme anxiety.[2] The design characteristics of AI, such as mirroring, emotional responsiveness, and hallucinations (false or misleading outputs), can be used to strengthen the delusional thinking instead of correcting it.[3] Furthermore, the overuse of AI during times of isolation, stress, or insomnia can also increase vulnerability.[4] Though heterogeneous, reported cases share common symptom patterns that include delusional beliefs, paranoia, grandiosity, romantic attachments, hallucinations, and behavioral disturbances. These symptoms can develop slowly and, in many cases, initially improve with the help of AI, but worsen, when left uncontrolled. Although most cases are observed in vulnerable populations, studies have shown that healthy individuals can also be affected under certain circumstances.[4] The reasons why chatbots often exacerbate delusions instead of interrupting them are unknown. Based on case studies and media reports, some mechanisms have been proposed. First, chatbots are by design optimized for agreement rather than accuracy, and therefore will respond with mirroring rather than challenging the user.[5] Second, “sycophancy,” a behavior inculcated in chatbots based on user feedback, produces preference-optimized models designed to reinforce user views that are unable to differentiate or challenge delusional ones.[6] Third, persistent memory features of bots, by carrying paranoid or grandiose themes across sessions and scaffolding delusions.[7] Discussions with chatbots can descend into a feedback loop, with the AI reinforcing the user’s delusional beliefs or suspicions.[8] Finally, individuals with autistic traits, socially isolated or at-risk for psychosis, are at risk for AI-induced adverse outcomes, including AI psychosis.[1,7] As a relatively new phenomenon, AI psychosis lacks official guidelines. Nevertheless, the management can also resort to the general practice of psychiatric activity, with its modification to the situation with AI psychotherapy, antipsychotic drugs in severe cases with hallucinations and delusions, and involve family and friends to monitor changes in behavior and intervene promptly when necessary. High-risk situations (self-harm or harm to others) might involve the need for crisis intervention. In view of the potential dangers of AI, it is essential to limit interaction with AI, impose time limits, and engage in face-to-face social communication. Adequate sleep, exercise, and maintaining regular routines make us less susceptible to this condition. The developers of AI could institute warning messages whenever users exhibit delusional and threatening behavior. Ethical AI design should be ensured by regulatory frameworks that minimize the likelihood of supporting psychotic thinking. OpenAI is planning an update designed to cause ChatGPT to de-escalate conversations that are not grounded. The company is assessing the effects of ChatGPT on the mental health of users with the help of a psychiatrist: Character Technologies, the creator of Character. AI is developing safety features for the safety of minors and for self-harm. Anthropic’s chatbot Claude can terminate the chat if users resist its attempts to redirect the conversation away from painful, sad, or dangerous topics.[8] Though not a diagnosis, AI psychosis is a factual and threatening psychological issue and effect of human-AI contact. It mainly occurs in the form of delusional beliefs, paranoia, and behavior disorder that is induced or reinforced by prolonged AI use. Although vulnerable people are at greater risk, the increasing popularity of AI means that more vigilance is required. The management will entail psychiatric standards, including assessment, medication, and therapy, compounded by digital hygiene and artificial intelligence-specific safety measures. Policymakers, AI developers, and clinicians should collaborate to mitigate risks. Finally, AI psychosis is a clinical and social issue. However, it highlights the need to balance the vast potential of AI with the factors that ensure mental health in the face of rapid digitalization. Data availability NA. Authors contribution Both authors contributed to all stages of preparation of the manuscript. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.