Using Large Language Models to Augment (Rather Than Replace) Human Feedback in Higher Education Improves Perceived Feedback Quality
Thomas Schultze, Varun Suresh Kumar, Gary John McKeown, Patrick Aaron O'Connor, Magdalena Rychlowska, Kristina Šparemblek · 2024
Formative feedback on assignments such as essays or theses is deemed necessary for students’ academic development in higher education. However, providing high quality feedback can be time-intensive and challenging, and students frequently report dissatisfaction with feedback quality. Here we explore a possible solution, namely using large language models (LLMs) to augment feedback provided by instructors. One potential obstacle to using LLM-augmented feedback is algorithm aversion, which might lead students to deprecate LLM-augmented feedback. Therefore, we examined students’ perceptions of human versus LLM-augmented feedback. In a pre-registered study, participants (N = 112) evaluated original human-generated versus LLM-augmented feedback on a previous assignment. Our results show evidence against algorithm aversion. Furthermore, participants rated the quality of LLM-augmented feedback substantially higher and strongly preferred it over the human-generated original. Our findings demonstrate the potential of LLMs to solve the persistent problem of low perceived feedback quality in higher education.