Impacts of an Artificial Intelligence Tutor in Foundation Physics: Cognitive Load, Confidence, Interest, Calibration, and Performance

Abstract

This study investigates whether a GPT-based tutor can improve learning and self-monitoring in introductory physics. Two parallel sections studied Newton’s Second Law: An experimental class used the tutor artificial intelligence (AI) and a control class used a textbook (CO). Students then took a quiz with per-question confidence ratings and completed brief questionnaires on extraneous cognitive load, intrinsic cognitive load, self-efficacy, situational interest (SI), and affect (EMO). We examined quiz performance, confidence patterns, and metacognitive calibration, how closely confidence matches accuracy. AI students scored higher and showed better calibration: they tended to be confident when correct and less confident when wrong. CO students reported higher overall confidence but showed weaker calibration. Questionnaire reliability (Cronbach’s α) was generally higher for AI, with SI and EMO most consistent. Overall, the GPT-based tutor improved performance, supported motivation, and strengthened metacognitive judgment in a foundation-level physics setting.