Between Promise and Practice: Bridging Ethical Artificial Intelligence Literacy Gaps Across Students, Educators, and Policy

Synthesis: Fekete (2026) synthesizes two European questionnaire studies — one with students (n = 226) and one with instructors (n = 256) — that used parallel five-point scales to measure AI Literacy, motivation, Ethics, intention, and institutional support in Higher Education. Independent-samples t-tests revealed significant group differences: students rated their ethical awareness higher than instructors, while instructors reported stronger willingness to experiment with Generative AI tools and higher behavioral intention — in part because students view Ethics through their immediate coursework practices, whereas teachers interpret it through a lack of institutional clarity and treat it as a question of integrity. Cluster analysis of instructors identified three user groups differing across literacy dimensions, with the Ethics scale providing the clearest distinction, and correlation analyses showed that instructors' moral awareness grows with institutional and social support while students' confidence and readiness correlate mostly with Self-Efficacy and collaboration rather than formal instruction.

Key Findings

• Students report higher ethical awareness than instructors: the largest gap in the study, with students averaging 4.03 versus instructors' 2.44 on ethical awareness (t(480) = 18.143, p < .001). Students tend to frame Ethics as fairness within immediate coursework, while instructors treat it as a question of institutional clarity, human oversight, and academic integrity.
• Instructors report stronger willingness and behavioral intention to use AI: instructors scored higher on willingness (t(480) = −3.879, p < .001) and intention (t(480) = −9.643, p < .001), yet perceived their own ethical confidence as low — a "ready but hesitant" profile that reflects the absence of institutional guidance.
• Both groups report weak institutional support: students rated external support low (M = 2.19) and described largely self-directed, informal AI learning; instructors perceived stronger (yet still limited) support through training or peer exchange, highlighting a responsibility gap in Educational AI Policy.
• Cluster analysis of instructors reveals three distinct profiles: k-means clustering identified Beginner (24%), Intermediate (56%), and Advanced (20%) users based on AI-Technological Pedagogical Content Knowledge (TPACK) dimensions, with the Ethics scale discriminating most strongly between clusters (F = 124.073, p < .001) — ethical awareness rises as professional readiness advances.
• Support drives instructor readiness, peers drive student readiness: for instructors, ethics correlates with facilitating conditions (r = .322) and behavioral intention (r = .353); for students, self-efficacy correlates with willingness (r = .730) and collaboration (r = .653), indicating readiness arises from confidence and peer interaction rather than formal instruction.
• Engagement is asymmetric: students use AI mainly for surface tasks (brainstorming 38.1%, short homework 36.3%, information search 32.7%), while instructors use LMS in 71.1% of classes but only 10.9% mention AI and fewer than 6% actively demonstrate it — students expect guidance teachers have not yet developed.

Study Design & Method

This is an empirical synthesis applying a comparative and interpretive design (Yanow, 2014) to two related, non-representative convenience-sample questionnaire studies conducted in 2024 across the European Higher Education context. The student study (n = 226; mean age 24.96; largest cohorts from Hungary and Austria) used a 15-scale instrument measuring components of AI Literacy and attitude, while the instructor study (n = 256; mean age 47.14; 18.6 years teaching experience) used 18 multi-item scales organized into three blocks: the seven TPACK domains, four AI-TPACK extensions, and seven AI literacy/behavioral constructs. All scales used five-point Likert items and met reliability standards (Cronbach's α ≥ .70). Student data were analyzed with independent-samples t-tests; instructor data additionally supported k-means cluster analysis (with ANOVA and post-hoc Duncan tests) and Pearson correlations. Participation was voluntary and anonymous with ethical approval (Refs #2024/02/3 and #2024/02/4), and analyses were run in SPSS 22 with significance accepted at p < .05.

What this means for practice

• Instructors. Demonstrate AI use in class rather than only mentioning it, and attach structured feedback to it: the study recorded AI mentioned in only 10.9% of class sessions and demonstrated in fewer than 6%, while students reported learning about AI informally and on their own. Evaluate ethical AI use through process, reflection, and judgment rather than detection alone.
• Learners. Build competence through peer collaboration and practice, since students' willingness correlated with Self-Efficacy (r = .730) and collaboration (r = .653) far more than with formal instruction.
• Administrators. Publish clear, context-sensitive guidance on AI use with named points of contact, because instructors' ethical awareness tracked facilitating conditions (r = .322) and the largest gap in the study was students rating support low (M = 2.19).
• Faculty developers. Tier professional development to the three instructor profiles the cluster analysis found (Beginner 24%, Intermediate 56%, Advanced 20%) and lead with Ethics, the scale that discriminated most strongly between them (F = 124.073, p < .001), instead of running uniform workshops.

Limitations

• Both questionnaires used non-representative convenience samples (n = 226 students, n = 256 instructors) drawn largely from Hungary (56.6%) and Austria (25.5%), so the group differences cannot be generalized to European or global higher education.
• All measures are self-report on five-point Likert scales, which the author notes may not capture the full complexity of actual practice.
• The two datasets were collected separately in 2024 and combined as an empirical synthesis, so student and instructor responses are not paired within the same institutions or courses.
• The comparative design captures group differences at a single time point and cannot show how ethical literacy develops through instruction, mentorship, or institutional reform.

Citation

Fekete, I. (2026). Between promise and practice: Bridging ethical artificial intelligence literacy gaps across students, educators, and policy. Journal of University Teaching and Learning Practice, 23(5).