Responsible AI-Governance Judgment in LLM-Mediated Higher Education: A Four-Occasion Mixed-Methods Study of Performance, Vigilance, and Implementation

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

Responsible use of large language models (LLMs) in higher education requires students to verify evidence, identify affected stakeholders, calibrate reliance, and defend human-owned decisions. This four-occasion mixed-methods study examined responsible AI-governance judgment across two coded quantitative groups. The quantitative strand comprised 900 participants (450 per group) and used adjusted T4 models; the qualitative strand drew on source-linked written and interview evidence from 14 coded cases. G1 had small adjusted advantages in AI-governance judgment (HC3 b = 2.01, 95% CI [0.72, 3.30], p = .002, Hedges g = .21), decision-dossier performance, socioscientific reasoning, and practical-wisdom orientation. G2 had higher epistemic-vigilance/calibration, AI-literacy orientation, verification-ledger, scaffold, and fidelity scores. Qualitative themes emphasized visible reasoning, proportionate verification, human-owned defense, self-scaffolding, and implementation boundaries involving burden, privacy, access, language confidence, and team agency. Because the retained quantitative records did not include a verified key linking G1/G2 to named learning conditions, and no person-level quantitative-qualitative link was available, the findings support a multidimensional coded-group profile rather than a condition-labeled causal claim.

Links and sources

Comments