Abstract: [Purpose/Significance] With the development of artificial intelligence, virtual teacher has emerged as new artificial intelligence pedagogical agents, and a new forms of education has also appeared. This study aims to explore the role of knowledge presentation modes and virtual teachers' simulation fidelity in Human-AI pedagogical interaction so as to provide new possibilities for improving learning effect in the immersive environment and provide guidance for curriculum construction. [Design/Methodology] Under the scenario of HumanAI pedagogical interaction, 54 participants were recruited to complete a within-subjects design experiment of 3（knowledge presentation mode: text/picture/video）×2（degree of realism: hyper-realism/cartoon）. Participants were randomly assigned to one of six groups and watched the corresponding instructional videos. The learning effects of participants were measured through pre- and post-test questionnaires. [Findings/Conclusion] In Human-AI pedagogical interaction with virtual teachers, the mode of knowledge presentation has a significant impact on the learning effect and the learning effect of video is obviously better than that of text and pictures. These effects are only observed in virtual teachers with high-fidelity simulation, while there are no significant difference in the learning effects of different knowledge presentation modes in the low-fidelity simulation condition. [Originality/Value] Focusing on virtual teacher, a new artificial intelligence pedagogical agent,this study accurately measures the user learning experience through controlled experiments, which breaks through the limitations of experiential design and provides a scientific basis for the optimization of the virtual teacher visual features and its knowledge presentation modes, and enriches the basic design principles of teaching courses centered on virtual teachers.

Key words: Human-AI Interaction, Virtual teacher, Knowledge presentation mode, Simulation fidelity, Learning effects