Universities increasingly tell students that AI should be used responsibly. The difficulty begins when we ask what “responsibly” means in a course, what students are expected to learn, and how a teacher can assess it. General principles such as fairness, transparency, privacy, and accountability sound important. They do not automatically become teachable knowledge.
This is the problem Andrea Aler Tubella, Marçal Mora-Cantallops, and Juan Carlos Nieves take up in their study of responsible AI in higher education. Their work fits naturally within my broader coverage of AI and teaching practice because it moves the discussion from ethical guidelines toward curriculum, learning outcomes, and classroom activity.
Summary of the Study
The researchers combined a structured literature review with interviews involving 11 higher education experts affiliated with institutions in five European countries. The review examined work on teaching ethical competencies across disciplines, while the interviews explored how the European Commission’s seven requirements for trustworthy AI could be introduced into university programmes.
These requirements cover human agency and oversight, technical safety, privacy and data governance, transparency, fairness and inclusion, societal and environmental wellbeing, and accountability.
The clearest finding is the gap between policy language and classroom practice. Experts valued the European guidelines, yet described them as abstract, technically worded, and difficult to translate across disciplines. Ethical issues were sometimes already present in courses, but they were rarely connected explicitly to a complete responsible-AI framework.
Teachers also lacked concrete learning objectives, ready-to-use cases, assessment criteria, and time to develop the necessary expertise. This is a familiar AI-literacy problem: naming competencies is much easier than building instruction that develops them, a concern I discussed in my review of AI literacy frameworks.
The authors organize responsible-AI learning into three levels. Ethical appreciation means recognizing that an AI application contains an ethical issue and identifying who may be affected. Ethical analysis asks students to examine possible actions, relevant principles, and likely consequences.
Ethical application requires them to select and implement a response, such as a privacy-preserving data practice, a bias-mitigation method, an audit procedure, or a form of human oversight. I find this progression especially useful because ethical education can easily stop at discussion. Students may learn to say that an algorithm is biased without learning how bias should be investigated, reduced, documented, or communicated.
The evidence still has limits. The literature search used one database, covered work available in early 2021, and included studies of ethics education that were not always specifically about AI. The expert group was small and heavily represented computer science cases, while the paper did not preserve enough discipline-level information to compare perspectives systematically.
Interviews identified needs and informed recommendations; they did not test whether the proposed learning outcomes improve ethical reasoning or professional action. The framework is also closely tied to European trustworthy-AI guidance. Its principles travel well, but their interpretation will change across legal, cultural, and institutional settings.
Practical Insights
The three-level model can be turned into a classroom sequence built around one realistic case. Suppose a university wants to use an AI system to identify students considered at risk of dropping out.
- Appreciation: Students identify the ethical issues and affected groups. They might notice questions involving privacy, consent, bias, transparency, human agency, and the consequences of a false prediction.
- Analysis: Students examine the data being collected, consider who could be disadvantaged, compare possible responses, and connect their reasoning to ethical principles or professional codes.
- Application: Students propose a workable design. This might include limits on data collection, a procedure for contesting decisions, documentation requirements, bias testing, or a rule preventing automated decisions without human review.
The same structure can be adapted to medicine, business, education, media, engineering, law, and the social sciences. The technical depth should match the discipline. Every student may need to recognize how an AI system could harm or exclude people. Students responsible for building or deploying systems need the additional ability to implement safeguards.
Assessment should follow the same progression. Asking students to define fairness measures awareness. Giving them a case and asking them to defend a response measures analysis. Requiring them to redesign a workflow, test a dataset, prepare an impact assessment, or document an audit moves closer to application.
This complements the emphasis on judgment in critical AI literacy, where students are expected to question how systems work, whose interests they serve, and what consequences follow from their use.
Universities also need to stop treating responsible-AI education as the private project of a few motivated instructors. Faculty need shared cases, adaptable rubrics, time for course redesign, and access to colleagues in computing, ethics, law, education, and the disciplines where AI is being applied. Research should now test these materials in actual courses, examine how competencies develop over time, and use student work or performance tasks alongside self-report measures.
Reference
Aler Tubella, A., Mora-Cantallops, M., & Nieves, J. C. (2024). How to teach responsible AI in higher education: Challenges and opportunities. Ethics and Information Technology, 26, Article 3. https://doi.org/10.1007/s10676-023-09733-7
