Understanding university students’ AI ethical decision-making in academic contexts: a SOR - social cognitive perspective

Abstract

Introduction The integration of generative artificial intelligence into higher education has reshaped students’ academic practices and blurred normative boundaries around academic integrity. While prior research has focused mainly on attitudes and usage intentions, limited attention has been paid to students’ AI ethical decision-making in academic scenarios and its psychological mechanisms. This study examined the associations among AI ethics guidance, academic ethics course experience, moral cognition, cognitive complexity, and AI ethical decision-making. It tested whether moral cognition mediated the relationships between normative inputs and AI ethical decision-making and whether cognitive complexity strengthened these relationships. The study integrated the Stimulus–Organism–Response framework with Social Cognitive Theory and combined net-effect, configurational, and importance–performance analyses. Methods Survey data were collected from 1,106 undergraduates at Chinese universities. Partial least squares structural equation modeling was used to test direct, mediating, and moderating relationships, complemented by fuzzy-set qualitative comparative analysis to identify configurational pathways to high AI ethical decision-making. Importance–performance matrix analysis identified intervention priorities. Results AI ethics guidance (β = 0.334, p < 0.001) and academic ethics course experience (β = 0.336, p < 0.001) were positively associated with AI ethical decision-making. Both normative inputs were positively associated with moral cognition (β = 0.479 and 0.280, respectively; p < 0.001), which was positively associated with AI ethical decision-making (β = 0.215, p < 0.001). Significant indirect relationships were identified through moral cognition (indirect β = 0.103 and 0.060; p < 0.001). Cognitive complexity strengthened the relationships between the two normative inputs and AI ethical decision-making (β = 0.077 and 0.073; p < 0.05). The model explained 66.3% of the variance in AI ethical decision-making (R2 = 0.663). fsQCA revealed no single necessary condition; however, AI ethics guidance and academic ethics course experience consistently emerged as core conditions across high-performing configurations (overall consistency = 0.934; coverage = 0.793). Discussion The findings advance academic integrity research by shifting attention from attitudes to scenario-based decision quality and clarifying the internalization mechanism through moral cognition. Practically, they highlight the complementary roles of technological prompts and academic ethics course experience in fostering students’ AI ethical decision-making.