Abstract
INTRODUCTION: This experimental study investigates how engagement modes with AI-related information-structured courses, group discussions, and self-directed research-influence attitude polarization and policy preferences among 132 Chinese undergraduates at a northern Chinese university. Methods: Participants were randomly assigned to conditions over a six-week intervention, with cognitive load and perceived reliability assessed as key mechanisms. METHODS: Participants were randomly assigned to conditions over a six-week intervention, with cognitive load and perceived reliability assessed as key mechanisms. RESULTS: Hierarchical regression revealed structured courses, marked by high cognitive load and reliability, significantly reduced polarization (β = -0.32, p < 0.01, η(2) = 0.11), while self-directed research increased it (β = 0.45, p < 0.01, η(2) = 0.15). Self-reported polarization strongly correlated with pre-to-post-test shifts (r = 0.68, p < 0.001), validating the General Attitudes Toward Artificial Intelligence Scale (GAAIS). Policy preferences mirrored these shifts, with structured courses fostering balanced stances (mean change = -0.15, SD = 0.40, p < 0.05). DISCUSSION: This study suggests structured, reliable, cognitively demanding interventions mitigate polarization, offering theoretical insights into attitude formation and practical guidance for AI education and policy design.