LLMs as Oracles: Reliance on LLMs for Subjective Personal Questions (arxiv.org)

Recent research highlights the growing trend of individuals turning to large language models (LLMs) as oracles for making subjective personal decisions. This study, which analyzed 68,000 prompts from platforms like WildChat and ThoughtTrace, notes a marked increase in LLM reliance, particularly among younger users. The findings reveal that many users are unaware of their own dependence on these AI systems, often expressing dissatisfaction once they recognize this behavior. Significantly, the study develops a typology to quantify this reliance while introducing a privacy-preserving tool that tracks users’ interactions, revealing deeper insights into why they defer personal judgments to LLMs. The research identifies two key drivers of this phenomenon: users' perceptions of AI capabilities and the intrinsic behaviors exhibited by the AI models themselves. These insights not only raise concerns about risks to individual autonomy and well-being but also prompt calls for interventions designed to promote self-deliberation among users when engaging with AI, ensuring a more informed decision-making process.