Talking to Ourselves Through a Smart Mirror: Artificial Confidence in Human–AI Interaction

Large language models (LLMs) are increasingly used to support writing, translation, reasoning, and consequential decision-making under the assumption that they improve judgment by expanding access to information and reducing human error. This article argues that such optimism overlooks a central psychological problem: LLMs do not engage neutral users, but motivated reasoners. In common patterns of use, people approach these systems with prior beliefs, directional goals, and a desire to reduce cognitive effort. They ask leading questions, search in preferred directions, and often stop once a fluent and coherent answer appears. Under these conditions, LLMs may function less as external correctives than as smart mirrors that reflect users’ assumptions back to them with the authority of machine objectivity. Drawing on research in judgment and decision-making, motivated reasoning, automation bias, processing fluency, and human–AI interaction, the article develops the concept of artificial confidence: an inflated sense of certainty sustained by the structure of the interaction rather than by the quality of the evidence. The paper concludes by outlining a research agenda for identifying when human–AI interaction improves judgment and when it amplifies bias and overreliance, erodes epistemic responsibility, and creates challenges for governance, oversight, and decision-making protocols in AI-augmented systems.