Higher Education’s GenAI Dilemma: Innovate Without Losing Integrity

Generative AI has moved from experiment to everyday academic infrastructure. Students use it to summarise readings, structure essays and test ideas; instructors use it to design courses, generate examples and reduce administrative work. However, many universities still treat the issue mainly as a compliance question: where to draw the line between legitimate assistance and misconduct.

A study published in the journal AI & Society argues that this framing is too narrow. "Social Representations of GenAI and Paradoxical Tensions in Its Adoption in Higher Education," by Audris Umel and Christoph Lattemann of Constructor University in Bremen, examines how students and instructors understand GenAI and how those shared understandings shape behaviour and institutional pressure.

Based on 12 semi-structured interviews with seven students and five instructors from Azerbaijan, Colombia, Ghana, Mexico, the Philippines and Russia, the study finds that GenAI is simultaneously seen as an innovative but demanding tool, a form of revolutionary progress and a normative challenge to academic values. The key lesson is not that universities must choose between adoption and restraint. It is that they must govern both at once.

The Efficiency Gain Comes with a New Institutional Burden

The attraction of GenAI is obvious: it saves time. Participants described using it to retrieve information, organise ideas, improve presentations and support writing. Instructors also valued it for brainstorming, structuring courses and overcoming creative blocks. But the study shows that efficiency is only one side of the equation. Effective use requires prompting skills, source verification, tool selection and awareness of bias, hallucinations and data risks. Students and instructors do not possess these capabilities equally, while university policies often lag behind actual use.

This produces an efficiency–complexity paradox. A tool that reduces workload in one area creates new work elsewhere. Time saved on drafting may be offset by time spent checking accuracy. Faster course preparation may require more careful assessment design. Students may complete tasks quickly, while instructors shoulder the responsibility of deciding whether the work still reflects meaningful learning.

GenAI is not simply another software licence to procure. It changes workloads, teaching roles and expectations. Instructors become policy interpreters, technical guides and ethical gatekeepers, often without formal support. Students may also become more proficient with new tools than the educators expected to regulate them.

Blanket restrictions may preserve short-term control, but they can leave students without the AI literacy increasingly expected in professional life. Efficiency gains therefore need matching investment in training, support and redesigned academic processes.

Control Without Empowerment Will Push AI Use Underground

The study also identifies a tension between control and empowerment. Students and instructors wanted clearer guidance on acceptable use, privacy, verification and academic integrity. At the same time, they valued GenAI because it gave users greater autonomy: students could explore unfamiliar subjects, while educators could create more adaptive learning experiences.

The danger lies at both extremes. Weak governance can encourage overreliance and uncritical acceptance of outputs. Excessive control can produce surveillance, distrust and covert use. When students believe institutions will punish any engagement with GenAI, they are less likely to disclose how they used it or seek guidance.

This is why the study's idea of "ambidexterity" matters. Universities need to manage contradictory demands together: standardisation and flexibility, openness and vigilance, steady integration and urgent adaptation. These tensions will not disappear once a policy is written.

A credible framework should set university-wide principles while allowing disciplines to define context-specific practice. A coding course, design studio and history seminar will not need identical rules. The common core should cover transparency, privacy, verification and accountability; departments should decide which forms of AI assistance advance or weaken learning.

The findings also suggest that students should help shape governance. They are often the earliest adopters and possess direct knowledge of how GenAI is used in practice. Excluding them risks producing rules that appear coherent but fail in classrooms.

Instructors, meanwhile, may need to become "orchestrators" of AI-enabled learning rather than only enforcers. That means designing tasks in which students explain reasoning, disclose tool use, compare outputs, verify sources and defend decisions. Responsible use is more likely to emerge through guided practice than prohibition.

Academic Integrity Must Be Rebuilt Around Process, Not Detection

GenAI is most disruptive where universities rely on written products as evidence of learning. If a machine can generate a plausible essay, code solution or presentation, the link between submitted work and demonstrated understanding becomes weaker. The study describes this as a tension between ethical safeguarding and ethical adaptation. Participants wanted honesty, accountability and critical engagement to remain intact, but they also recognised that rules designed before generative systems existed cannot simply be stretched indefinitely.

Academic integrity should not be confused with preserving every pre-GenAI assessment format. Its purpose is to protect genuine learning and fair evaluation. If an assignment can be completed convincingly by outsourcing most reasoning to a model, the assessment may no longer measure what the institution intends.

Universities thus need to move from detection-led governance toward process-led assessment. AI detectors can produce false accusations, but even perfect detection would not solve the educational problem. The more durable response is to make thinking visible. It could involve oral defence, staged submissions, annotated source trails, reflective accounts of AI use, in-class application and comparison between human- and model-generated outputs. Students should show not only what they produced, but how they reached it, what they accepted or rejected from GenAI and why.

Critical thinking becomes the bridge between adoption and integrity. Students need to test reliability, question assumptions and distinguish useful assistance from intellectual substitution. University leaders and instructors must also reconsider what counts as originality, authorship and legitimate support in an AI-rich environment.

GenAI forces universities to clarify their purpose. If learning is defined mainly as producing information, machines will look increasingly competitive. If it is defined as judgment, interpretation, argument, ethical reasoning and disciplinary understanding, the human role remains central.

The Biggest Risk Is Institutional Delay, Not Technological Speed

The research is based on a small convenience sample and should not be treated as representative of all universities or developing countries. Participants were connected to a digital-transformation course and may have been more technologically engaged than average. The study also excludes university leaders, regulators and employers, and it examines perceptions rather than measurable learning outcomes.

Even so, its conceptual contribution is strong. It shows that universities struggle with GenAI because legitimate demands conflict. Institutions must encourage innovation while preserving standards, support experimentation while managing risk, and respond quickly without hard-coding rules around tools that may soon become obsolete.

In developing and emerging economies, universities may face weaker infrastructure, limited training budgets and dependence on commercial platforms developed elsewhere. Unequal access to premium AI tools could widen educational disparities, while weak data governance may create privacy and sovereignty risks.

These challenges connect directly to SDG 4 on quality education and SDG 10 on reducing inequalities. GenAI can expand access to explanation, feedback and content generation, but only where students have connectivity, language support, digital literacy and human mentorship. Without those conditions, the technology may amplify inequality.

Future research should use larger and more diverse samples, compare disciplines and countries, and include administrators, policymakers and employers. Longitudinal studies could track how perceptions change as policies mature, while experimental work could test which assessment and governance models improve learning, trust and integrity.