The crisis is deeper than cheating
Generative artificial intelligence has not invented the crisis of university assessment. It has exposed it. For decades, higher education has relied heavily on essays, take-home assignments, reports and exams designed around one basic assumption: the text submitted by a student is a reliable proxy for that student’s understanding. That assumption is no longer stable.
This is not a simple story of dishonest students and innocent institutions. Students now live in a world where AI systems can summarize readings, suggest arguments, polish prose, generate code, translate, produce outlines and imitate academic style. Some students use these tools responsibly, as support for searching, drafting or revising. Others use them in ways that hollow out the learning process. The problem is that universities often lack fair, transparent and pedagogically meaningful ways to distinguish between assistance, collaboration, substitution and deception.
The answer cannot be surveillance as pedagogy. Keystroke tracking, compulsory writing inside monitored platforms, intrusive detection systems and a general presumption of suspicion would create poorer universities. They would damage trust, increase anxiety, add administrative burden, and risk unfair accusations against students whose writing differs from dominant academic norms. AI detectors are not a sound foundation for academic justice. A university should not become a text-policing institution.
Assessment must make learning visible
The first reform must therefore concern assessment. Universities should ask less often whether a student can submit a polished final product and more often whether the student can explain, defend, apply and revise what they claim to know.
This requires a mixed ecology of assessment. Extended writing should remain central, because the ability to develop an argument over time is still a core academic competence. But the final take-home essay should no longer carry such disproportionate weight. It can be accompanied by research logs, annotated bibliographies, drafts, source decisions, oral defenses and clear declarations of AI use. Students should learn that responsible use of AI includes accountability for every claim, source and conclusion.
More assessment also needs to happen in conditions where the student’s thinking is directly observable. In-class writing can ask students to reconstruct concepts, interpret passages, compare arguments or solve problems. Oral exams and short viva-style conversations can reveal whether a student has actually understood the course material. In science, engineering, health and applied disciplines, laboratories, fieldwork, demonstrations, prototypes and reproducible workflows can make process as important as output.
The goal is not to invent a perfect AI-proof assessment. There is no such thing. The goal is to create a resilient assessment system in which cheating is less attractive, genuine learning is easier to identify, and students encounter multiple opportunities to show understanding in different forms.
AI literacy is now part of academic literacy
Universities should not treat AI only as a threat to integrity. They must teach it as part of contemporary literacy. Every student, whatever their discipline, needs a working understanding of how generative models operate, why they produce fluent but unreliable outputs, how bias and data quality shape results, what privacy risks arise when sensitive information is entered into commercial tools, and how intellectual property, citation and authorship are affected.
AI literacy is not prompt training. It is judgment. Students must learn when AI can help and when it weakens their thinking. They must learn how to verify outputs against reliable sources, how to compare machine-generated explanations with disciplinary standards, and how to recognize that fluency is not evidence. A student who can critically interrogate an AI answer has learned something valuable. A student who merely delegates thinking has not.
This also means that every course needs clear rules. In some tasks, AI use should be prohibited because independent performance is being assessed. In others, AI can be allowed as a tool, provided students disclose how they used it. In advanced settings, students can be asked to critique AI outputs, improve them, identify errors or compare them with primary sources. Transparency should replace ambiguity.
More teachers, not more platform dependency
A serious response to AI will require more educational labour. Oral exams, draft feedback, smaller seminars, laboratory supervision, meaningful assessment design and student mentoring take time. If universities try to respond to AI by cutting staff while buying more automated systems, they will accelerate the very crisis they are trying to solve.
The reform of assessment must therefore be connected to the public funding of teaching. Universities need more instructors, properly supported teaching assistants, professional development for academic staff, recognition of teaching quality in academic careers and institutional time for redesigning courses. Generative AI does not reduce the need for teachers. It increases the need for human guidance, interpretation and academic responsibility.
Access is equally important. If powerful AI tools are available only to students who can pay for them, universities will reproduce inequality under the language of innovation. Institutions should provide safe, transparent and preferably open digital infrastructures for learning. Open educational resources, open standards, auditable tools and locally governed AI systems matter because knowledge institutions should not become dependent on a handful of commercial platforms. Public universities in particular have a duty to defend knowledge as a public good.
A new contract of trust
The rise of generative AI forces universities to ask what higher education is for. If the aim is simply to produce fluent text, machines will keep getting better. If the aim is judgment, understanding, intellectual honesty, disciplined inquiry and democratic responsibility, then universities are more necessary than ever.
The right response is not nostalgia for a pre-digital classroom. It is a more demanding, fairer and more open university. One with explicit AI policies at course level. One that assesses process as well as product. One that values oral fluency, written reasoning, practical demonstration and collaborative inquiry. One that avoids both naive trust and authoritarian surveillance. One that invests in teachers, not only in software.
Generative AI does not abolish the university. It abolishes the comfort of a university that confuses submitted work with learning. The task now is radical reform: assessment that reveals understanding, curricula that teach AI critically, infrastructure that protects equality, and academic cultures that rebuild trust through better design. Students should not be trained merely to avoid AI or to hide behind it. They should be educated to think, judge and create in a world where AI is everywhere.
Article sources
David A. Bell, Beating AI in the Classroom: This article provides a concrete higher education case, centred on Princeton and generative AI, and explains why traditional honor-based assessment, AI detection and take-home writing are no longer sufficient without in-class writing, oral assessment and more teaching labour: https://davidabell.substack.com/p/beating-ai-in-the-classroom,
UNESCO, Guidance for Generative AI in Education and Research: UNESCO’s guidance directly supports the article’s argument for human-centred governance, privacy protection, institutional readiness and pedagogical redesign in the use of generative AI in education and research: https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research,
UNESCO, AI Competency Frameworks for Teachers and Students: These frameworks show why AI literacy should become part of formal education, covering human-centred thinking, ethics, technical understanding, responsible use and creative participation in AI-enabled societies: https://www.unesco.org/en/articles/ai-competency-framework-teachers and https://www.unesco.org/en/articles/ai-competency-framework-students,
Russell Group, Principles on the Use of Generative AI Tools in Education: The Russell Group principles are directly relevant because they call for AI literacy, staff support, adaptation of teaching and assessment, academic rigour, equal access and cross-sector collaboration: https://www.russellgroup.ac.uk/policy/policy-briefings/principles-use-generative-ai-tools-education,
Jisc, Trends in Assessment in Higher Education: The report documents current assessment reform trends in higher education, including programme-level assessment, redesign in response to generative AI, oral and process-focused assessments, portfolios and student co-design: https://discovery.ucl.ac.uk/id/eprint/10223514/1/trends-in-assessment-report.pdf.