Embodied AI-Guided Interactive Digital Teachers for Education
Zhuoran Zhao, Zhizhuo Yin, Jia Sun, Pan Hui
Abstract
Traditional education is considered incapable of providing prompt feedback, facilitating proactive learning, and giving indiscriminate responses.This has been observed in both in-person classes and online courses, especially when students' questions fall outside the instructors' knowledge base or are considered trivial by instructors.Nowadays, the advent of large language models (LLMs) has transformed knowledge acquisition.The LLM-based chatbots enable fast learning through interactive question-answering, which serves as an effective supplement to traditional educational approaches and even shows potential for replacement.To utilize such advancements in education, we propose MAGI, a novel system providing Embodied AI-Guided Interactive digital teachers for education, which integrates LLM-based chatbot technology.To ensure MAGI generates answers without hallucination, we employ a novel retrieval-augmented generation (RAG) paradigm to organize and retrieve useful educational documents for the LLM.Moreover, we create animatable 3D avatars powered by text-to-speech and audioto-motion models to provide students with interactive conversation experiences.We highlight the possibility of MAGI to enhance education accessibility and improve the overall learning experience.