AI-Assisted Assessment of Experimental Physics Laboratory Reports: Potential, Limitations, and Support for Teaching Practice
The incorporation of AI into the assessment of physics laboratory reports offers opportunities to support the review of student work and the provision of feedback. This article revisits an experience involving the use of ChatGPT models to analyze laboratory reports and examines the conditions that influence their usefulness in teaching practice. The analysis considers the evolution of AI models, document-processing procedures, and different modes of interaction, with particular attention to the effective availability of evidence contained in student reports. The findings indicate that differences between AI-generated and teacher assessments may be related not only to model capabilities but also to difficulties in retrieving and interpreting equations, graphs, tables, units, and other visual elements. The possibilities of systematic processing of sets of reports and conversational review of cases requiring more focused analysis are also examined. Based on these observations, a working strategy is proposed in which AI serves as a complementary resource for organizing assessment, supporting feedback, and identifying recurring difficulties in student work. The analysis highlights the importance of document-processing conditions and evidence availability, as well as the teacher's role in interpreting AI-generated information and making evaluative and pedagogical decisions.