Abstract
Background
The rapid expansion of artificial intelligence tools in higher education demands understanding whether students' moral development is associated with more responsible AI use. Moral development, conceptualized through Kohlberg's stage theory, provides a theoretically grounded framework for examining this relationship, yet few empirical studies have tested it in a Latin American university context.
Methods
A quantitative, cross-sectional, correlational design was used with a quota sample of 5,487 Peruvian university students recruited through institutional channels. Moral development was assessed using the Defining Issues Test-2 (DIT-2) and responsible AI use was measured with the EURIA-ES, a 24-item self-report scale with evidence of reliability and factorial structure comprising four dimensions: transparency, academic honesty, critical algorithmic awareness, and responsibility in content dissemination.
Results
Most participants fell in the middle descriptive N2 band (61.4%; N2 mean = 32.14), using descriptive cut points rather than validated Kohlbergian stage diagnoses. EURIA-ES scores were moderate (M = 72.34 out of 120), with academic honesty scoring highest and critical algorithmic awareness lowest. The N2 index showed an important standardized association with EURIA-ES scores (beta = .31). The final model accounted for 23.5% of the variance (R2 = .235), and entering N2 increased explained variance by 9.2 percentage points beyond the sociodemographic block (ΔR2 = .092). Interpretation emphasizes effect magnitude and 95% confidence intervals rather than statistical significance alone. Year of study and field of knowledge showed adjusted associations, whereas regional differences were small in magnitude.
Conclusions
within a cross-sectional design, higher DIT-2 N2 scores were associated with higher self-reported responsible academic use of generative AI. N2 should be interpreted as an important correlate rather than a primary determinant of responsible AI use. These findings pertain specifically to academic uses of generative AI and should not be generalized to all artificial intelligence systems or to objectively observed student behavior.