Design and Application of Pedagogical Agent with Foundation Models

Abstract

With the rapid development of generative artificial intelligence (GenAI), agents based on foundation models (FMs) have gradually acquired capabilities, such as multimodal perception, retrieval-augmented generation, reasoning and planning, interaction, and evolution. This study proposed a fundamental concept and framework for FM-based pedagogical agents. Centering on the FM as the technological core, the framework primarily consisted of several functional modules, including educational task setting, educational task planning, realization and expansion of educational capabilities, memory and reflection on educational content, and interactive collaboration and dynamic evolution. These modules support interaction with diverse entities and facilitate dynamic evolution, encompassing human-computer interaction, multi-agent interaction, and environment interaction. Building upon the proposed framework, this study adopted project-based learning tasks as an application scenario. It elaborated on the roles of pedagogical agents functioning as assistant agents and peer agents across various phases, such as personalized driving question formulation, collaborative project plan design, collaborative project work completion, and multi-role evaluation of project works, along with the associated supporting technologies. Furthermore, this study discussed the future development and prospects of pedagogical agents.

Keywords

pedagogical agent / foundation model (FM) / generative artificial intelligence (GenAI) / project-based learning

RIGHTS & PERMISSIONS

Higher Education Press