Research Article
Artificial intelligence ethical awareness of university students in Ghana: A network and latent profile analyses
Synthesis: Acquah and colleagues (2026) assessed the artificial-intelligence ethical awareness (AIEA) of 509 undergraduates at the University of Cape Coast, Ghana — 295 (58 percent) male, 214 (42 percent) female — across human autonomy, beneficence, and fairness, using a nine-item scale developed in Hong Kong. Confirmatory factor analysis supported the three-factor structure, though human autonomy's average variance extracted of 0.382 fell short of the 0.50 threshold. Network analysis linked the items through 30 of 36 possible connections, with the item holding that AI "can never replace ultimate human responsibility" bridging most strongly to concerns about benefits outweighing risks. Latent profile analysis retained five profiles, from "Comprehensive Very High" (26.1 percent) to "Low Ethical Awareness" (4.5 percent, beneficence 2.06), while gender made no meaningful difference. The authors frame AI ethical awareness as coherent yet unevenly distributed, with implications for AI ethics curricula in Ghana.
Key Findings
- The three-factor structure replicates, with one weak factor. Confirmatory factor analysis supported human autonomy, beneficence, and fairness as distinct dimensions among the 509 undergraduates, but average variance extracted reached 0.592 for beneficence and 0.574 for fairness against only 0.382 for human autonomy, below the 0.50 threshold.
- Awareness sits in the high band, unevenly. Construct means were 4.17 for human autonomy, 4.06 for beneficence, and 4.12 for fairness, placing the sample just inside the top band while leaving beneficence the weakest dimension.
- The dimensions form a relay-like network. The network contained nine items and 30 nonzero connections out of a theoretical maximum of 36; the item asserting that AI "can never replace ultimate human responsibility" carried the highest cross-construct weight at 0.273, linking autonomy to the benefits-outweigh-risks item, while the harm-minimization item bridged toward social justice.
- Five latent profiles span comprehensive to low awareness. Profiles were "Comprehensive Very High" (133 students, 26.1 percent), the largest "Very High Autonomy-High Ethical" (180, 35.4 percent), "Fairness-Oriented High" (78, 15.3 percent), "Moderate Awareness" (95, 18.7 percent), and "Low Ethical Awareness" (23, 4.5 percent), whose beneficence score of 2.06 fell in the rejecting range.
- Close to a quarter of students fall below the headline averages. The two lowest-scoring profiles together account for close to a quarter of the sample, so aggregate means obscure a meaningful minority well below the construct averages.
- Gender adds no meaningful variation. One-way MANOVA found no statistically significant difference between male and female students (Pillai's trace 0.005, F = 0.869, p = 0.457).
Measuring Ethical Awareness Across Contexts
The paper's first contribution is measurement: it tests a Hong Kong instrument (Kong and Jinyu, 2025) in a Global South setting, where the authors stress that cross-cultural transfer cannot be taken for granted. Overall reliability was strong and the three factors stayed empirically distinguishable, but human autonomy failed the convergent validity threshold and the RMSEA of 0.100 exceeded the recommended ceiling, which the authors read as likely cultural variation in how Ghanaian students conceptualize autonomy.
Uneven Awareness and the Role of Beneficence
The profile and network analyses add a distributional story that a mean would hide. Students cluster into five subgroups, so aggregate figures conceal a spread from comprehensive awareness to a small group the authors describe as rejecting the principles, with beneficence the weakest dimension. That the responsibility item is the hub linking to benefit-versus-risk concerns suggests students who see autonomy at stake also question whether AI serves their good — an argument for teaching the principles connected rather than in isolation.
Five Profiles and the Case for Tiered Teaching
Because profiles differ in shape, not just level, the authors argue a uniform curriculum fits almost no one well. The largest profile pairs strong agency endorsement with lower beneficence, the "Moderate Awareness" group sits near the neutral midpoint, and the smallest group scored 2.06 on beneficence. Recommended responses differ accordingly: dilemma-based modules for the middle group, more intensive and possibly mentored engagement for the smallest, and modules connecting autonomy to outcomes for the largest. The authors also flag the retained five-class solution's entropy of 0.816 against the four-class solution's 0.929, and suggest consolidating profiles for a more stable grouping.
What this means for practice
- Instructors and curriculum designers. Weight AI ethics teaching toward beneficence, the weakest dimension, and connect it explicitly to autonomy rather than assuming it follows from general ethical awareness.
- Instructors. Plan at least two or three intervention tiers rather than one compulsory module, since the five profiles mean a single lesson will land very differently.
- Researchers. Use the validated instrument to compare Global South contexts, collect richer background data than gender alone, and treat the responsibility–benefit bridge as a theory-building target.
Limitations
- The scale captures self-reported attitudes, not how students actually reason through ethical dilemmas; the authors note no validated instrument yet measures stages of ethical decision making in concrete AI situations.
- Data come from a single university in Ghana at one point in time, so representativeness for the wider Ghanaian student population is not established, and the design cannot show whether stated awareness corresponds to actual AI use.
- Gender was the only background variable collected — age, academic discipline, educational level, and AI literacy were not measured — so the study cannot identify which student characteristics predict profile membership.
- The smallest profile contains only 23 students, and the authors themselves question its stability and suggest consolidating it with the moderate group.
Connected Concepts
Connected Articles
- Situated AI ethics: a cultural-historical and ecological framework for education — Situated AI ethics: a cultural-historical and ecological framework
- Understanding ethical dimensions of AI in higher education: insights from faculty members and students — Understanding ethical dimensions of AI in higher education
- Beyond Compliance: A Proposed Framework for Ethical Governance of Student Data in Learning Analytics — Ethical governance of student data in learning analytics
Citation
Acquah, B. Y. S., Salifu, I., Arthur, F., Inkoom, M., Suradji, S. D., Inkoom, C., Quayson, E., Quaye, S. A., Gyedu, F. O., & Nortey, S. A. (2026). Artificial intelligence ethical awareness of university students in Ghana: A network and latent profile analyses. Computers and Education: Artificial Intelligence, 100652.