Zoe Stanley
Towards a Mechanistic Understanding of Delusional Belief Formation during Interactions with AI Chatbots
Recent advances in large language models have generated widespread use of AI chatbots as tutors, assistants and emotional support tools for individuals around the world. Despite their newfound pervasiveness, extended use of AI chatbots is increasingly implicated in a phenomenon sometimes referred to as 鈥淎I psychosis,鈥� the development of delusion-like belief states as a result of long-term interactions with AI chatbots. While existing research has focused heavily on chatbot sycophancy, or the tendency of chatbots to agree excessively with users鈥� statements, little is known about the underlying mechanisms by which chatbot use may be linked to AI-associated delusions. Through a combination of short-term behavioral study, computational modeling and magnetoencephalography (MEG), my project seeks a mechanistic account of how dimensions of a chatbot鈥檚 affective richness may transiently alter human perception.