Reconfiguring tutor–peer roles to foster critical thinking in medical PBL assisted by tutor‐managed chatGPT

Background Critical thinking skills (CTS) are essential in medical education. Problem‐based learning (PBL) effectively develops CTS, with generative AI (e.g., ChatGPT) offering enhancement potential. Objectives This study investigates how a tutor‐managed ChatGPT‐PBL model: (1) reconfigures tutor/peer roles and CTS compared to traditional PBL; (2) establishes predictive relationships between tutor/peer roles and CTS and (3) confirms optimal operational parameters for tutor‐managed ChatGPT to enhance CTS. Methods A quasi‐experimental study assigned 170 medical students to either a ChatGPT‐PBL or a traditional PBL group. Data from a 31‐item survey assessing tutor and peer roles and CTS, along with learning logs and tutor observations for 36 participants, recorded AI engagement metrics (e.g., query frequency, multi‐source validation, comparative questioning). The intervention used a pre‐defined “3–2‐60 Rule” (≥3 queries/session, ≥2 authoritative sources/claim, >60% comparative framing), derived from existing literature, to guide ChatGPT engagement. Interrelationships were examined. Results ChatGPT‐PBL improved CTS outcomes, enhancing tutor‐guided constructive/self‐directed/contextual/collaborative learning and peer‐driven information processing/communication/critical analysis. Regression identified collaborative learning and information processing as CTS predictors in both approaches, but ChatGPT shifted their influence: collaborative learning's role decreased while information processing's increased. High adherence to the “3–2‐60 Rule” correlated with CTS gains, with high‐engagement groups (>3 queries, ≥2 sources, comparative framing) outperforming peers. Conclusions Tutor‐managed ChatGPT reconfigures tutor/peer roles in PBL to enhance CTS development. Structured peer‐ChatGPT‐tutor interactions (following the 3–2‐60 rule) contributed to enhancing CTS. Given ChatGPT's socioemotional limitations, future implementations should embed adversarial ChatGPT tasks and ethical transparency protocols.