Human-AI Collaboration in Skill-Based Classroom Feedback

Abstract

This study investigates how human-AI collaboration can improve feedback quality and learning efficiency in skill-based classroom instruction. The study is designed to involve approximately 520 students and 36 instructors from practice-oriented courses over a 12-week teaching cycle, generating around 7,800 student practice submissions, 5,200 AI-generated feedback reports, 3,600 teacher feedback records, 2,400 learner self-correction logs, and 1,500 peer-feedback entries. The research measures feedback timeliness, feedback specificity, correction adoption rate, practice improvement score, learner autonomy, teacher workload, and student satisfaction. AI feedback is generated through multimodal analysis of text, image, audio, or task-performance data, while teacher feedback is used as the professional reference for evaluation. Quantitative methods include paired-sample t-tests, ANCOVA, multilevel regression, mediation analysis, inter-rater reliability testing, and learning gain analysis. The study compares three feedback conditions: teacher-only feedback, AI-only feedback, and human-AI collaborative feedback. Evaluation indicators include feedback response time, task revision frequency, performance score improvement, learner self-regulation score, teacher workload reduction, and agreement between AI and teacher judgments. The innovation of this study lies in treating AI not as a replacement for teachers but as a feedback amplifier that improves timeliness and diagnostic coverage while preserving human judgment for expressive, contextual, and high-level learning evaluation.