INTEGRATING HUMAN–AI COLLABORATIVE EXPLAINABLE GENERATIVE AI FEEDBACK INTO EFL WRITING CLASSROOMS: IMPACTS ON ACADEMIC WRITING PERFORMANCE, LEARNING OUTCOMES, STUDENT AUTONOMY, AND LEARNER AGENCY

Abstract

Generative Artificial Intelligence (AI) tools like ChatGPT are quickly becoming part of the English as a Foreign Language (EFL) writing classroom, and research has not yet caught up to understand the impact of their use. This study aimed to investigate if a human–AI collaborative explainable feedback model, where a generative AI tool generates feedback that includes explanations, and a teacher supports students with feedback will lead to better outcomes than teacher-only feedback. A quasi-experimental pretest–posttest control-group design was adopted throughout a 12-week undergraduate EFL writing course. A total of 64 students were divided into an experimental group (n = 32) and a control group (n = 32) where the former group was given human–AI collaborative explainable feedback while the latter group was given traditional teacher-only written corrective feedback. All participants were asked to write four timed argumentative papers, and to answer the writing performance, learning outcomes, autonomy, and agency questions at Week 1, Week 6, and Week 12. The experimental group exhibited significant greater gain than the control group on total writing performance, t(62) = 13.92, p < .001, d = 3.48, on all five rubric dimensions, content knowledge and revision quality and self-regulated learning, and on both autonomy and agency, all p < .001. Results indicate that using generative AI feedback in conjunction with teacher explanation and scaffolding could be more beneficial for the development of students' writing skills and their autonomy in learning. Classroom implications, the role of explainability for building learner trust and directions for future research are discussed.