Multidimensional Assessment and Differential Analysis of Generative AI in Educational Live Action Role-Playing (Edu-LARP): A Case Study of "Total Internal Reflection" in High School Physics

Abstract

Addressing the tension between immersion and knowledge
construction in game-based learning, this study constructs
a physics Edu-LARP (Educational Live Action Role-Playing)
model utilizing Generative AI as non-player characters
(NPCs). Using "Total Internal Reflection" as a case study
and employing the "Simulated Learner" method, ChatGPT and
Doubao were evaluated across three dimensions: conceptual
construction, scaffolding adaptability, and narrative
integration. The findings reveal that Doubao favors
"rule-directed construction" and "procedural hard
scaffolding," reducing cognitive load for novices. ChatGPT
excels in "intuition-directed deduction" and "diagnostic
soft scaffolding," which facilitates higher-order thinking.
However, both models are prone to inducing "role
conflicts," thereby undermining immersion. This research
clarifies the application boundaries of AI in gamified
instruction, providing empirical support for optimizing
human-AI collaborative inquiry-based classrooms.