Development and Preliminary Evaluation of a Teacher-Supervised AI-Assisted Closed-Loop Learning Model for an Undergraduate Animal Physiology Laboratory Course

Animal Physiology Laboratory requires students to integrate physiological knowledge, experimental procedures, animal handling, real-time signal interpretation, and post-class reflection. To address the fragmentation of conventional laboratory teaching, this study developed an AI-assisted closed-loop learning model integrating online preparation, virtual simulation, predictive interpretation, hands-on animal experimentation, real-time physiological signal acquisition, troubleshooting-oriented review, and teacher-supervised AI-assisted feedback. Two evaluations were conducted at class-level: (1) a non-concurrent historical cohort comparison of students in Animal Science and Veterinary Medicine who entered the university in 2020 or 2021 (pre-reform cohorts) and 2022 or 2023 (post-reform cohorts); and (2) a cluster-randomized comparison involving 386 Veterinary Medicine students from 14 classes, allocated to either traditional teaching (7 classes; n = 190) or the AI-assisted group (7 classes; n = 196). The lab course lasted 8 weeks, while the teaching team, instructional content, laboratory resources, assessment criteria, and grading rubrics were matched between groups. The historical cohorts had higher reported final grades after the curriculum reform in both majors (p < 0.05). In the cluster randomized comparision, baseline prerequisite-course scores were comparable between the traditional and AI-assisted groups (81.93 ± 0.65 vs. 81.13 ± 0.59, p = 0.3807), and the final laboratory scores were higher in the AI-assisted group (94.27 ± 0.27 vs. 89.09 ± 1.16, p < 0.01). Although postrequisite-course scores and score gains from prerequisite to postrequisite courses were slightly higher in the AI-assisted group, neither difference was statistically significant (p = 0.1470 and p = 0.3022, respectively). These findings suggest that the AI-assisted closed-loop learning model can improve course-specific learning performance in veterinary physiology laboratory education. Its longer-term effects on subsequent-course performance require further investigation.