Reproduce “AI-Mediated Feedback Improves Student Revisions: A Randomized Trial with FeedbackWriter in a Large Undergraduate Course” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xinyi Lu, Kexin Phyllis Ju, Mitchell Dudley, Larissa Sano, Xu Wang
This study from the University of Michigan developed and evaluated FeedbackWriter, a human-AI collaborative system, demonstrating that AI-mediated feedback leads to significantly higher-quality student revisions in knowledge-intensive essays compared to human-only feedback, while maintaining similar learning outcomes. The system facilitates more comprehensive and desirable feedback by leveraging AI suggestions under human oversight.
Key insights
- AI effectively complements human expertise by enhancing rubric coverage and generating feedback messages that possess desirable properties like actionability, promotion of independent learning, and a supportive tone.
- Effective human-AI interaction design, which aligns with human cognitive processes for reviewing and editing AI suggestions, is critical for efficient auditing, correction, and adoption of AI-generated content.
- Iterative refinement of assignment rubrics in collaboration with instructors is crucial for improving AI's judgment accuracy and enhancing consistency among human evaluators, turning tacit knowledge into explicit criteria.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.