AI Insight
This study investigated how 22 South Korean pre-service mathematics teachers integrated AI-generated and peer feedback when revising sustainability-focused mathematical modeling tasks about school cafeteria food waste. Participants rarely copied feedback passively, with most making selective revisions based on different feedback sources presented in varying sequences. Both AI and peer feedback prompted distinct types of engagement: AI feedback led to more validation-focused revisions and was valued for specificity and speed, while peer feedback was appreciated for classroom realism and student perspective, though neither feedback sequence produced consistently superior task-design quality outcomes.
Why it matters
The findings suggest that teacher educators should explicitly teach pre-service teachers how to evaluate, compare, and synthesize multiple feedback sources rather than assuming they will automatically translate feedback into improved designs. This has practical implications for integrating AI tools in teacher education programs, emphasizing the need for scaffolding critical evaluation skills alongside technological adoption.
Understand the Science
IntroductionAs generative AI becomes increasingly available in teacher education, a central question is how pre-service teachers interpret, evaluate, and translate different feedback sources into coherent design decisions. This study examined how 22 South Korean pre-service mathematics teachers used AI and peer feedback while revising sustainability-oriented mathematical modeling tasks on school cafeteria food waste, a context aligned with Sustainable Development Goal 12.MethodsUsing an exploratory, comparative mixed-methods sequence-comparison design embedded in an authentic teacher-education course, participants received two rounds of feedback in different sequences: AI followed by peer feedback or peer feedback followed by AI. Framed within cognitive, affective, and behavioral dimensions of feedback engagement, the analysis combined double-coded ratings of initial and final task designs, feedback-response records, revision-depth coding, validation-focused revision indicators, and post-activity reflections.ResultsParticipants rarely copied feedback passively: 20 selectively accepted feedback in at least one round, and 17 made at least one structural or reconceptual revision. Five participants showed a positive composite gain in rated task-design quality. Participants valued AI feedback for its specificity, new perspectives, and speed, and more often made validation-focused revisions after it, whereas they valued peer feedback for mutual communication, classroom realism, and student perspective. Descriptive comparisons showed different engagement trajectories across the sequence conditions alongside similar composite-quality changes.DiscussionFeedback translation provides an analytic perspective on the divergence between revision activity and artifact quality. Within this course context, the findings show how feedback evaluation, revision decisions, and artifact quality can diverge and suggest that teacher educators should scaffold evaluative judgment, source comparison, and post-revision coherence checking.