A Comparison of AI-Assisted and Manually Scored Student Projects in Physiology Education

As class sizes grow, providing high-quality, accurate, timely feedback on student projects becomes increasingly difficult for instructors. This research explores the potential of using artificial intelligence (AI) to assist instructors in providing student feedback. The authors hypothesized that AI Large Language Models can provide valuable, accurate, and timely feedback on student papers in a physiology class. The tested sample comprised thirty-two written papers of around 1.5 to 3 pages in length, each evaluating two medical TV drama episodes across seven point scale rubric categories. Since the papers were sampled from a pre-existing data set, they had previously been scored with a point scale rubric once (human existing) by a single reviewer. Papers were de-identified and rescored by three independent reviewers (human new, scores averaged) and Claude 3.5 Sonnet (July 1, 2024 version). Investigators looked at the time to completion, written feedback, and rubric point scale scoring. Point scale scoring was the only consistent comparison across all three evaluations. This is because Human existing scores did not initially include written feedback and time to completion. Human new scores did not include written feedback in order to mimic the original scoring workflow. We prompted AI to include written feedback as an experimental extension since this required no additional effort. A one-way ANOVA revealed no statistically significant differences between the existing human scores (human existing), the new human scores (human new), and the scores generated by Claude 3.5 Sonnet (F = 0.12, p = 0.89), indicating high accuracy for each scoring method. Human new scoring took an average of 3 minutes and 19 seconds for each paper with no written feedback. AI completed its assessment, which included written feedback, in under 10 seconds per paper. The instructor judged AI feedback to be robust, accurate, and highly relevant. The results suggest that AI could enhance academic performance by directly providing students with quality feedback before submitting an assignment. This could also free the instructor to focus on teaching and curriculum development. Future research should explore productive workflows and methods of assessment for AI-assisted student work that promote targeted learning outcomes. This project was funded by the Department of Physiology at Michigan State University This abstract was presented at the American Physiology Summit 2025 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.