A Phygital Approach to Social-Emotional Learning: Effectiveness of a Generative AI Chatbot as a Digital Transitional Space for Junior High School Students

Abstract

Traditional large-class instruction often struggles to provide individualized guidance for Social and Emotional Learning (SEL). Addressing the theme of physical-digital integration (phygital), this study explores the effectiveness of introducing a generative AI chatbot as a "digital transitional space" for teaching Nonviolent Communication (NVC). Utilizing a quasi-experimental design, 40 eighth-grade students were assigned to either the experimental group (Baseline NVC AI, n=21) or the control group (Traditional Instruction, n=19). Quantitative results revealed that the system possessed an acceptable level of usability (p = .876 against the industry standard). The experimental group significantly outperformed the control group in the Nonviolent Communication Behaviors Scale (NVCBS) (p = .042) and performed comparably in the Situational Judgment Test (SJT) (p = .408). However, qualitative log analysis revealed a 95.2% dialogue survival rate. Despite this high completion rate, some students exhibited boundary-testing and passive harm behaviors when facing repeated objective corrections from a single-style AI. These findings provide empirical evidence for the limitations of a one-size-fits-all AI in a phygital ecosystem, highlighting the need for adaptive mechanisms based on the Stereotype Content Model (SCM) in future studies to buffer emotional frustration.