The competition in generative artificial intelligence (GenAI) ignited by the arrival of the large language model ChatGPT in late November 2022 does not seem to lose momentum, forcing various social systems to overcome the existential distress they might feel about the uncertainty of what the AI-future might bring. Higher education institutions are no exceptions. Universities around the globe are trying to align themselves to the new realities imposed by GenAI in the classrooms. This entailed changes in assessment methods and the adoption of various guidelines and regulations to set the ethical framework for the responsible use of AI.
While some believe AI could transform education by offering personalised learning experiences and customised teaching methods, others argue it might widen socio-economic disparities due to unequal access to technology.
Examining the initial reactions from universities reveals several key challenges and effective responses to the ethical dilemmas posed by AI.
Accountability and responsibility
Definition: Mechanisms should be implemented to ensure responsibility and accountability for AI systems and their outcomes before and after deployment. Generative AI must be designed to allow for auditing, enabling the evaluation of algorithms. Clear roles should be defined in data management and design processes to build trust in AI technologies. Minimising and reporting negative impacts is essential to identify, assess, and document potential issues. At the same time, redress mechanisms must be available to provide legal and practical recourse in cases of unjust outcomes.
Best practice: In the academic context, AI cannot be considered an author or co-author, and AI-assisted technologies should not be cited as authors. GenAI platforms cannot initiate an original piece of research without the direction of human authors. For this reason, the ethical guidelines of Wiley, for example, state that
[AI tools] cannot be accountable for a published work or for research design, which is a generally held requirement of authorship, nor do they have legal standing or the ability to hold or assign copyright.
–Wiley 2023
These views are shared by the most prominent academic publishing houses, including Springer Nature, Elsevier, and Taylor & Francis, who all maintain that human authors remain responsible for their work’s originality, validity, and integrity.
Universities globally uphold a similar stance in their ethical guidelines, requiring students to follow established scientific referencing and authorship rules. Academic honesty and mutual trust are fundamental, and students are prohibited from submitting AI-generated content as their own. Students bear full responsibility and accountability for the content of their papers.
Human agency and oversight
Definition: Human agency and oversight in AI systems require individuals to be involved in every decision-making stage, allowing users to make autonomous choices regarding AI tools.
Best practice: A key element in managing this is including a “stop” function or similar regulatory mechanisms to maintain control over the operation of generative AI tools. Current university guidelines recommend a cautious approach, often opting for soft sanctions. This is primarily due to the unreliability of AI detection platforms, which struggle to distinguish between human and AI-generated work. Additionally, these tools raise concerns about privacy and data security. As a result, universities are careful when implementing such measures.
In summary, the best practice that can be identified under this ethical dilemma is to avoid overreaction and to secure human oversight through a blend of preventive measures (e.g., a shift in assessment methods) and soft sanctions. Given that AI detectors are unreliable and can cause a series of data privacy issues, the sanctioning of unauthorised AI use should happen on a “soft basis” as part of a dialogue with the student concerned (Imperial College London calls this an “authenticity interview”).

Transparency and explainability
Definition: Explainability means providing clear information about how AI tools work, focusing on how the user’s input and output interact with each other. On the other hand, transparency means ensuring clear communication about various aspects of AI usage. Universities should spread information about AI and encourage basic AI literacy to empower people and close the digital gaps that have grown precisely because of the widespread use of such tools.
Best practice: The implementation of this principle seems to revolve around the decentralisation of university work, including the respect for teachers’ autonomy. Teachers should be given the authority to determine whether and to what extent AI tools may be used in their courses. However, they must communicate these decisions to students and university administrators via the course syllabus.
Overall, the best practice answer to the dilemma of transparency is the internal decentralisation of university work and the application of a “bottom-up” approach that respects the autonomy of university professors.
Inclusiveness and diversity
Definition: In higher education, inclusiveness aims to ensure that an institution provides all its students with the same opportunities to access the benefits of AI technologies, irrespective of their background. It also considers the particular needs of various vulnerable groups potentially marginalised based on age, gender, culture, religion, language, or disability.
Best practice: Proactive central support and the pledge to leave no one behind is pivotal. AI-related inclusivity for students is best fostered if the university does not leave its professors solely to their resources to develop AI-related initiatives. Universities should engage in elaborating/providing concrete materials (e.g., subscriptions to AI platforms, templates, video tutorials, training, open-access answers to FAQs…etc.) to ensure equal access to AI knowledge and specific ideas as a source of inspiration for teachers.
Conclusions
1. AI-related ethical dilemmas form complex challenges connected in multiple ways. University sources show that the primary aim is to explore and harness the potential benefits of generative AI, but not with an uncritical mindset. They focus on opportunities while simultaneously trying to address the emerging challenges in the field.
The main ethical rule is that students must complete university assignments based on the knowledge and skills they acquired during their university education unless their instructors determine otherwise. Moral and legal responsibility in this regard always rests with human individuals. AI agents possess neither the legal standing nor the physical basis for moral agency, which makes them incapable of assuming such responsibilities.
–Attila Dabis – Csaba Csáki 2024
2. This “top-down” requirement is often complemented by the “bottom-up” approach of providing instructors with proper manoeuvring space to decide how they would like to make AI use permissible in their courses.
3. Good practice in human oversight can thus be achieved through a combination of preventive measures and soft, dialogue-based procedures. This latter category includes teachers providing written communications in their course description and engaging in a dialogue with their students to provide clear and transparent instructions on using generative AI tools within their courses. Preventing the unauthorised use of AI tools can be reached more effectively by changing course assessment methods by default rather than engaging in post-assessment reviews due to the unreliability of AI detection tools.
In summary, the overarching challenge of generative AI in higher education is not the mere use of these tools but rather their unethical, uncritical, and irresponsible overuse.
🔬🧫🧪🔍🤓👩🔬🦠🔭📚
Journal reference
Dabis, A., Csáki, Cs. (2024). AI and ethics: Investigating the first policy responses of higher education institutions to the challenge of generative AI. Humanities and Social Sciences Communications 11, 1006 (2024). https://doi.org/10.1057/s41599-024-03526-z

