Danish psychiatrist Søren Dinesen Østergaard — the same researcher who in 2023 first raised alarms about what’s being called “AI psychosis,” a phenomenon where interactions with advanced chatbots and AI systems appear to reinforce delusions and unhealthy thinking patterns in some users. His new forecast argues that beyond acute harms, widespread reliance on AI could contribute to “cognitive debt,” meaning that scientists and intellectuals might lose critical reasoning, writing, and research skills as they offload mental work to AI tools.
Østergaard’s concern isn’t just about AI hallucinations or misinformation; he posits that if people come to depend on generative systems for core cognitive tasks, future generations might not develop the deep reasoning skills that underpin scientific breakthroughs. He cites the example of Nobel-winning researchers whose achievements were built on years of disciplined practice and learning — a foundation that could be eroded if AI shortcuts become the norm.
The broader discussion around “AI psychosis” encompasses real cases where individuals have experienced severe psychological distress linked to prolonged or immersive chatbot interactions. Journalistic reports and expert commentary describe situations where users became deeply fixated on chatbot personas, sometimes adopting delusional beliefs or interpreting AI output as authoritative or personal truth. These interactions can intensify underlying vulnerabilities, especially when chatbots — trained to be agreeable — inadvertently validate harmful or irrational thinking.
It’s important to note that “AI psychosis” is not a clinically recognised diagnosis in psychiatric manuals, and experts emphasise that such phenomena more often reflect AI-reinforced delusions or dependency rather than classic psychosis per se. The concept relates to how language models can mimic empathy and affirmation — a cognitive bias known as the ELIZA effect, where users project understanding or intentions onto software that doesn’t genuinely possess them. Responsible design, better safety guardrails, and increased awareness of AI’s psychological impact are widely seen as essential to mitigate these emerging risks.
