Socratic vs. Expert Agents: Conversational style effects on Programming Performance, Learning Experience, and Higher-Order Thinking
Authors
Shiqing Peng
Faculty of Artificial Intelligence in Education, Central China Normal University
Xueying Tao
Higher-order thinking has become the core mission of programming education in the digital transformation era, serving as a key competency for junior high school students to adapt to digital environments and achieve sustainable development (Lee et al., 2024a). However, traditional programming lessons often emphasize syntax over thinking, making it difficult for teachers to meet higher-order thinking cultivation needs (Patrick & Campbell, 2025). The integration of artificial intelligence offers a new pathway where educational agents can externalize students' implicit programming thinking through real-time natural language interaction, building a bridge for personalized cultivation (Ma et al., 2024). The effectiveness of these agents depends not only on their presence but also on how they interact with students. Among design features, conversational style is particularly important because it shapes students’ thinking and engagement. In this study, we focus on two representative styles: Socratic Agent and Expert Agent.；To address this gap, this study examines the effects of Socratic and Expert conversational styles in junior high school programming education. A total of 82 seventh-grade students were randomly assigned to learn C++ programming with either a Socratic Agent or an Expert Agent. The research questions are as follows:
Heng Luo
Higher-order thinking has become the core mission of programming education in the digital transformation era, serving as a key competency for junior high school students to adapt to digital environments and achieve sustainable development (Lee et al., 2024a). However, traditional programming lessons often emphasize syntax over thinking, making it difficult for teachers to meet higher-order thinking cultivation needs (Patrick & Campbell, 2025). The integration of artificial intelligence offers a new pathway where educational agents can externalize students' implicit programming thinking through real-time natural language interaction, building a bridge for personalized cultivation (Ma et al., 2024). The effectiveness of these agents depends not only on their presence but also on how they interact with students. Among design features, conversational style is particularly important because it shapes students’ thinking and engagement. In this study, we focus on two representative styles: Socratic Agent and Expert Agent.；To address this gap, this study examines the effects of Socratic and Expert conversational styles in junior high school programming education. A total of 82 seventh-grade students were randomly assigned to learn C++ programming with either a Socratic Agent or an Expert Agent. The research questions are as follows:
This study examined how educational agent conversational styles influence seventh graders' programming learning. In a between-subjects experiment, 82 students were randomly assigned to learn with either a Socratic Agent or an Expert Agent on the AI agent platform. Results showed that: (1) Students using the Socratic Agent demonstrated greater performance improvement on coding tasks but not on objective questions, indicating that agent effects vary by task type; (2) The Socratic Agent group reported significantly higher emotional engagement and behavioral engagement, while no differences emerged on other learning experience dimensions; (3) Agent interaction style significantly influenced problem-solving ability and creativity after controlling for pre-test levels, with the Socratic Agent showing advantages; (4) Both agent styles benefited students across high and low prior-coding skills, with no significant differences between groups. The study provides empirical evidence for designing and applying educational agents in programming education.
