Human-in-the-loop semantic learning analytics for assessing moral-value reasoning in university AI ethics education

Introduction University AI ethics education addresses fairness, privacy, accountability, transparency, human autonomy, and academic integrity, but the semantic structure of students' moral-value reasoning and its connection to responsible solutions remain insufficiently characterized. This study examined a human-in-the-loop semantic learning analytics framework for task-specific moral-value reasoning. Methods Ninety science and engineering graduate students completed an online explicit-reflective AI ethics module in a one-group pre–post multimethod study without a control group. Assessments comprised an AI ethics knowledge test, a perceived-awareness scale, and open-ended ethical reasoning tasks. Expert annotations informed transformer-based multi-label coding, descriptive value coverage and balance metrics, problem–solution coupling, and baseline-only nested cross-validated prediction of subsequent raw problem-solving change. Results Within-participant pre–post increases were observed in ethical knowledge, perceived moral-value awareness, actual ethical awareness, problem–solving skills, value coverage, value balance, and problem–solution coupling. Automated semantic coding achieved participant-grouped cross-validated macro-F1 = 0.83 (95% CI [0.78, 0.87]; micro-F1 = 0.86; macro-AUPRC = 0.82). The baseline-only prediction model yielded cross-validated R 2 = 0.30 (95% CI [0.11, 0.45]) for subsequent raw problem-solving change. Discussion Expert-supervised semantic analytics can characterize value representation and issue–solution reasoning in AI ethics assessments. Coverage and balance describe response structure rather than moral quality. The absence of a control group, repeated use of one scenario, and modest sample size preclude causal and broad trait-level conclusions.