From “Giving a Fish” to “Teaching to Fish”: Enhancing ITS Inner Loops with Large Language Models

In this work, we aim to enhance the Intelligent Tutoring Systems (ITS) inner loop by incorporating Large Language Models (LLMs), shifting the educational approach from merely providing solution (giving a fish) to fostering deeper understanding and self-sufficiency (teaching to fish) in learning. Specifically, we propose a framework that utilizes LLMs to generate meaningful scaffoldings and facilitate insightful interactions within ITS tasks, guided by established learning science theories. Preliminary results from the model and educational experiments suggest that our LLM-generated scaffoldings could improve learning outcomes. From our current progress, we have gained valuable insights for further improvement. We hope this study will contribute to the development of smarter and more human-centered ITS solutions in the future.