AI-Assisted Grading and Feedback in International Baccalaureate Middle Years Programme Physics: A Mixed-Methods Investigation of Learning Outcomes, Feedback Quality, and the Moderating Role of English Language Proficiency

Abstrak

This sequential explanatory mixed-methods study examined the impact of two artificial intelligence (AI) assessment modules on student learning outcomes in International Baccalaureate (IB) Middle Years Programme (MYP) Physics at an IB World School in Shenzhen, China. Participants comprised 101 Grade 9 and Grade 10 students, of whom 91% were non-native English speakers, alongside three science teachers, over a 24-week implementation period. Revision Village was deployed for Criterion A assessments, while ChatGPT supported Criteria B, C, and D. Quantitative data from pre- and post-intervention MYP criterion-referenced assessments, two Feedback Quality Assessment Rubrics, and Technology Acceptance Surveys were analysed using paired-samples t-tests, one-way ANOVA, and Hayes’ PROCESS macro (Models 1 and 4). Qualitative data from semi-structured teacher interviews, student focus groups, and classroom observations were analysed using Braun and Clarke’s six-phase thematic analysis. Results revealed statistically significant improvements across all four MYP criteria, with effect sizes ranging from Cohen’s d = 0.43 (Criterion B) to d = 0.74 (Criterion D), and overall MYP grade improvement from 4.58 to 5.27 (d = 0.64). Feedback quality characteristics significantly mediated 48.3% of the relationship between AI implementation and learning outcomes (bootstrap 95% CI [0.802, 2.047]). English language proficiency significantly moderated implementation effectiveness (B = −0.782, p = .010), with Low- and Mid-proficiency students demonstrating significantly greater gains than High-proficiency students. The study contributes evidence-based guidance for integrating AI assessment modules in linguistically diverse IB Physics classrooms and proposes a revised AI–Human Feedback Partnership Model in which teacher mediation functions as the central bridging mechanism between AI-generated feedback and learner outcomes.