Training AI To Produce Guiding and Meaningful Feedback To Support Learning and Foster Knowledge-In-Use
Abstract
Deep science understanding, or knowledge-in-use, is the ability to apply learning to explain phenomena. Supporting this skill requires open-ended tasks and high-quality, tailored feedback—both challenging to scale. We address a key challenge in educational AI: training models to provide student-guiding feedback grounded in pedagogical theory. We developed an AI training approach based on learning and feedback principles, along with a validated rubric to assess feedback quality. This method produces cognitively appropriate, personalized feedback and shows promise for designing AI systems that foster knowledge-in-use skills in a developmentally appropriate way.