The role of teachable agents’ personality traits on student-AI interactions and math learning

This study explores how the personality traits of pedagogical agents, particularly teachable agents, influence students' math learning experiences. Grounded in the Big Five personality traits framework, students were randomly assigned to tutor a teachable agent that either emphasized one of the five personality traits or did not emphasize a specific trait for mathematics problem-solving. Students interacted with agents of varying personalities across different problems, with each program featuring a single, designated personality for the teachable agent. Results indicate that openness-emphasis agents elicited more student explanations (i.e., cognitive engagement) than extraversion (b = 0.32, p < .001) and agreeableness-emphasis agents (b = 0.28, p < .001) during the teaching process. In contrast, extraversion-emphasis agents facilitated polite expressions (i.e., affective expressions) compared to conscientiousness-emphasis (b = 0.31, p = .002) and openness-emphasis agents (b = 0.29, p = .003). Similarly, students interacting with agreeableness-emphasis agents exhibited significantly more polite expressions than those engaging with conscientiousness-emphasis (b = 0.31, p = .002) and openness-emphasis agents (b = 0.29, p = .004). Additionally, non-personality-emphasis agents were more effective in fostering cognitive engagement, such as providing explanations, compared to agents emphasizing agreeableness (b = 0.29, p < .001) and extraversion (b = 0.25, p < .001). Furthermore, these non-personality-emphasis agents enhanced students' conceptual knowledge application more effectively than agreeableness-emphasis agents (β = −0.09, p = .04). These findings suggest that pedagogical agents do not necessarily need to rigidly embody a single personality trait to be effective. Instead, non-personality-emphasis agents that adapt their responses dynamically based on student interactions may better support learning in problem-solving contexts than emphasizing certain personalities. To optimize instructional effectiveness, pedagogical AI agents should be designed to align with diverse learning goals and adjust their responses flexibily to meet students’ needs. Future research should explore how adaptive pedagogical agents facilitate student engagement and learning across different educational contexts.