Remote and hybrid education has spread rapidly, revealing severe limitations of traditional human-invigilated and rule-based systems of examination monitoring. Maintaining academic integrity in online assessments has become increasingly difficult with students devising various ways to avoid scrutiny. In a bid to combat these challenges, this paper proposes an automated proctoring system that combines machine learning classification using YOLOv11-based pose estimation to detect suspicious behavior in real-time. The system takes live video streams or recorded video streams, and detects 17 essential body landmarks in each frame and estimates deviations in posture, trends in head movement, and unusual upper-body motion. A labeled dataset of examination scenarios was used to evaluate several classifiers. Mutual-information-based feature selection retained 30 pose features, and the standard K-Nearest Neighbors (KNN) with k=1 was selected for deployment. The selected model achieved an accuracy of 99.56%, precision of 99.20%, recall of 99.60%, F1-score of 99.40%, and an AUC (Area Under the Curve) of 0.9957 on the held-out test set. The system includes a lightweight user-friendly interface on the web that allows real-time deployment with just a single webcam to allow easy monitoring and immediate behavior feedback without the need to have specialized hardware. On balance, the suggested solution can provide a lightweight, efficient, and scalable solution to the problem of promoting academic integrity in online and face-to-face settings of the assessment environment.
Curriculum Analytics (CA) applies data-driven methods to improve curriculum design, quality assurance, and alignment with workforce needs. A key component of CA is mapping, professional skills to the assessments in a program of study as part of professional accreditation. Traditional approaches to curriculum mapping are labour-intensive, time-consuming, and prone to biases. While machine learning models, such as BERT have been shown to mitigate some constraints, they still face significant challenges, including neglecting varying associations across skills, dependence on extensive training data, and failure to capture the contextual nuances in the curriculum. Generative AI (GenAI) offers a promising alternative to mitigate some of the existing issues. Moving beyond binary mapping and utilising extensive training data, this study evaluates various prompting methods with GPT-based models to generate skill mappings guided by Messick's framework for assessing the validity of mappings. The skill mappings produced by the model are then compared with expert-created mappings and evaluated based on how well the reasoning behind the mappings can be validated, ensuring both accuracy and interpretability. The results show that GenAI models achieved reliable results for explicit and high-frequency skills and less so for implicit skills and still requires expert oversight paving the way for potential for broader applications in educational quality assurance.
This study examines the development of teachers’ AI-integrated teaching competence, defined as AI-Technological Pedagogical Content Knowledge (AI-TPACK), aiming to move beyond additive assumptions toward an ethically grounded model. Based on survey data from 454 Chinese university teachers, a suppression pattern emerged: isolated AI-Technological Knowledge (AI-TK) showed a significant negative direct association with AI-TPACK, while its positive contribution was realized indirectly through pedagogical, content, and ethical knowledge. This pattern suggests that ethics functions as an integrative force rather than a peripheral concern, supporting a conceptualization of AI-TPACK as an ethically mediated process rather than a simple accumulation of knowledge domains. A key implication for teacher professional development is the need to reorient training toward holistic approaches that embed ethical deliberation and pedagogical design, moving beyond isolated technological upskilling. Such pedagogically grounded, ethically informed approaches are essential for fostering responsible AI integration.
Artificial intelligence (AI) anxiety is increasingly prevalent among university students, yet mainstream AI literacy interventions are often infeasible in resource-constrained settings. This study focuses on a law-focused university in western China, where AI literacy education is absent. Students’ AI competence and AI ethical awareness are largely acquired through self-study, and it remains uncertain whether such self-acquired capacities can alleviate AI anxiety. Drawing on the stimulus-organism-response framework, we surveyed 584 law and computer science undergraduates and employed a dual-stage analysis combining partial least squares structural equation modeling (PLS-SEM) and artificial neural network (ANN) to clarify the relationships among AI competence, AI ethical awareness, and AI anxiety. PLS-SEM results revealed that AI competence did not directly reduce AI anxiety; instead, it amplified AI anxiety indirectly by enhancing AI ethical awareness, indicating indirect-only mediation. This mechanism held consistently across both disciplines. ANN analysis further identified AI ethical awareness dimensions as key predictors of AI anxiety. These findings suggest that without teachers’ guidance, AI ethical awareness may be an amplifier of AI anxiety. Accordingly, universities should proactively incorporate AI-related legal education to develop AI ethics course, helping students develop a reasoned perspective on AI ethics, thereby turning these constraints into an opportunity for an autonomous, evidence-based intervention pathway.
Background. Equipping learners with AI-ethical literacy is a priority for higher education, yet item-level self-reports leave two observables unrecorded: the structure of lexical co-use in learners’ written responses and the polarity of their free text. Neither is a measure of learning. A further inferential hazard is that shared-term statistics in participant–morpheme networks are sensitive to response length, so an apparent rise in shared vocabulary can reflect how much students write rather than lexical alignment; this confound is the target of the present study. We present Multimethod Ethics-Learning Evaluation (MELE), a multimethod evaluation workflow integrating paired post-reading/post-discussion statistical analysis, length-controlled inference on bipartite participant–morpheme networks, and BERT-based sentiment classification, demonstrated here as a single-group proof-of-concept on a facilitated case discussion in AI ethics.Methods. Twenty graduate students from diverse disciplines at Waseda University, Japan, deliberated on an authentic AI job-screening case grounded in the Japanese shūkatsu (job-hunting) context. Because T1 was administered after individual case reading, the paired contrast reflects change associated with the facilitated discussion rather than with case exposure as such. Three complementary analyses were applied: Holm/Bonferroni-corrected Wilcoxon tests, an exact conditional within-participant token-permutation reference test for discourse convergence with Monte Carlo-approximated p-values, and BERT sentiment classification (exploratory).Findings. The shared-term share of the participant–morpheme networks rose descriptively between administrations, while post-discussion responses were substantially shorter. Under the length-controlled within-participant token-permutation test, neither observed increase was statistically significant relative to its length-conditioned null distribution (two-sided Monte Carlo-approximated p-values; Q6 p=.62; Q7 p=.15), whereas a naive whole-bag permutation test would have reported both as significant; smaller effects cannot be excluded. Perceived fairness (Q2) rose significantly and survived multiple-comparison correction (W=15.5, p=.006, Holm/Bonferroni p=.031; rrb=−.77, large); this single-item effect is, however, sensitive to floor-responder exclusion: it attenuates to non-significance (p=.11, rrb=−.53) when the five participants who began at the scale floor are removed, so we report it as preliminary rather than established. The other four items showed non-significant directional gains; participant-level model-estimated polarity shifted positively but non-significantly.Implications. Methodologically, MELE contributes a length-controlled discourse-inference procedure for participant–morpheme graphs: we show that the conventional degree-preserving null is degenerate for the star configurations underlying the shared-term share s3B and that naive convergence permutation tests are confounded by response length, and we provide an exact conditional reference test with Monte Carlo-approximated p-values and a bounded simulation study. Under within-participant exchangeability the test is finite-sample valid by construction for any fixed configuration of response lengths (length imbalance can reduce power but cannot legitimately inflate the false-positive rate), with the simulation study serving as an implementation check. In one of the ten null cells, the exact 95% confidence interval for the false-positive rate lay wholly above the nominal .05 (point estimate .062), so the simulation does not by itself demonstrate calibration at that level, and performance beyond the evaluated cells remains to be characterized. Substantively, the fairness item showed a floor-sensitive, case-bound shift in this brief, single-case discussion. In this proof-of-concept, then, the workflow’s principal yield is the length-controlled discourse-inference procedure; the fairness shift is a single preliminary observation, and the discourse and sentiment components are hypothesis-generating rather than confirmatory, and we make no claim of cross-component corroboration. In this single-group proof-of-concept (N=20), all findings require controlled, longitudinal, cross-cultural replication; key limitations are discussed.
