Generative AI: Until a few years ago, a well-argued take-home assignment gave a teacher some basis for judging how well a student had understood a subject. Generative artificial intelligence has weakened that connection. An essay, case analysis or programming solution can now be produced with substantial machine assistance, often without leaving much evidence of how much intellectual work the student actually did.
Universities therefore face a problem that cannot be settled through rules on plagiarism alone. They have to decide what a degree is expected to certify when AI can perform many of the tasks traditionally used to assess students.
India’s education policy provides some room for such a rethink. The National Education Policy 2020 gives higher education institutions considerable freedom over curriculum, pedagogy and assessment, and calls for assessment to test the application of knowledge. CBSE has gone further at the school level, issuing a computational thinking and artificial intelligence curriculum framework, along with student and teacher resources, for Classes III to VIII for 2026-27.
READ | India’s education trap begins with unequal access
Higher education will have to address a different set of questions. Universities need to decide what students must know before using AI, how teachers should assess work produced with its assistance and which forms of examination still provide reliable evidence of independent understanding.
AI literacy must extend beyond technical disciplines
The content of AI literacy will vary across courses. A management student may encounter AI in forecasting or organisational decisions. A lawyer will have to understand what an algorithmic recommendation means for liability, evidence or intellectual property. Medical education will have its own requirements.
Students in the humanities and social sciences will confront questions about bias, privacy, employment and inequality created or amplified by automated systems. Their need is less for technical proficiency than for an ability to examine the assumptions behind those systems and their consequences.
Universities should accordingly define AI literacy by discipline. Prompting is one part of it. A student also needs enough subject knowledge to recognise an implausible answer, check the origin of a claim and understand when confidential information should not be entered into an AI system.
This puts verification at the centre of AI literacy. Generating a plausible response has become easy. Judging the response still requires knowledge.
Faculty capability will shape classroom use
Changes in assessment will be difficult to sustain if faculty members are unfamiliar with the tools their students are using.
This is partly a problem inherited from the existing examination system. Many university courses continue to place considerable weight on the reproduction of taught material. Generative AI can perform that task remarkably well. An examination that mainly rewards reproduction consequently reveals less about a student than it did earlier.
Teachers will need to redesign assignments around the learning outcome they want to measure. That requires familiarity with the capabilities of AI systems as well as their weaknesses.
UNESCO’s AI Competency Framework for Teachers offers a useful reference. Published in 2024 and updated online in January 2026, it sets out 15 competencies across five dimensions and arranges them at different levels of proficiency. The framework covers the use of AI in teaching as well as the ethical and professional questions that accompany it.
India could adapt this approach for higher education through a national faculty competency framework. The requirements would differ across disciplines, but all faculty members would need a working understanding of AI-generated content, academic integrity, data protection and the redesign of assignments.
READ | India’s skills vs degrees debate misses the real problem
IIT Delhi has already confronted some of these issues. Its guidelines on generative AI place responsibility for checking AI-assisted work on the person submitting it, require disclosure of substantial AI use and caution users against putting sensitive information into these systems. Academic units are also advised to reconsider assessments that AI can reproduce easily. The committee advises against blanket restrictions and leaves room for course-specific policies.
Those recommendations recognise a practical constraint. Students already have access to powerful AI tools. A university policy that assumes otherwise will have little effect on what happens outside the examination hall.
Assessment must distinguish output from understanding
Take-home assignments are becoming a less reliable measure of individual understanding when substantial parts of the task can be handed over to an AI system.
Universities will still need examinations in which students work without such assistance. Supervised written papers, classroom problem-solving and oral examinations can establish whether a student possesses the knowledge required to reason through a problem independently. The appropriate form will depend on the subject.
Independent knowledge also affects how well a student uses AI. Someone who does not understand a subject may struggle to spot a fabricated citation or a faulty assumption in an otherwise persuasive answer. The ability to challenge AI cannot be developed entirely by using AI.
Other assessments can permit its use and test a different capability. A student might be asked to examine an AI-generated answer, check its sources and explain which parts should be rejected. An AI-assisted business plan could be followed by an oral defence of its assumptions. Such exercises provide evidence of judgement that the submitted document alone cannot provide.
Imperial College London’s Business School has taken this approach to the design of assessments. Its IDEA Lab developed a generative AI “stress test” that exposes existing assessment tasks to AI tools and examines whether the work being assessed can now be produced too easily by a model.
The University of Melbourne provides a different example. In a redesigned sports medicine assessment, students compare their own clinical judgement with a generative AI second opinion. The assessment examines how they evaluate the machine-generated response rather than treating access to AI itself as misconduct.
Indian universities need not copy these exercises. They illustrate a useful principle. Once AI can produce the assigned output, the assessment has to reveal something more about the student.
READ | NEP 2020: School reform moves from access to outcomes
Cognitive offloading can weaken learning
There is a separate problem with allowing AI to undertake too much of the intellectual work involved in an assignment.
Writing, calculation and the construction of an argument are means through which students learn. The effort involved is part of the educational process. Repeatedly transferring that work to a machine can improve the submitted product while leaving the underlying capability unchanged.
The OECD’s Digital Education Outlook 2026 reviews emerging evidence on this question. It finds that access to general-purpose generative AI can improve students’ immediate performance without producing equivalent gains in learning. Some studies reviewed by the OECD found that the advantage seen in AI-assisted work disappeared, and sometimes reversed, when students were later examined without access to the tool. The report also discusses the risk of cognitive offloading and what researchers describe as “metacognitive laziness”.
The OECD evidence does not support excluding AI from education. It finds better learning outcomes when AI is used for a defined pedagogical purpose. The distinction lies in how much of the thinking the student still has to do.
A sensible course design would therefore establish subject knowledge before assigning work that relies heavily on AI. Students can use the technology later to test an argument or examine alternatives, provided they remain responsible for checking what it produces.
Otherwise, universities could end up awarding high marks for work that demonstrates the capability of the software more clearly than that of the student.
Universities need workable rules and equal access
Course-level changes will work better within an institutional policy that tells students and faculty what is permissible. Such a policy should cover disclosure of AI use, academic integrity, privacy, intellectual property, assessment design and access to paid tools.
Uniform rules for every subject would be difficult to justify. The acceptable use of AI in a coding assignment may differ from its use in an examination in history. Departments need room to set rules according to what a course is trying to teach and measure.
Universities must also consider unequal access. The more expensive AI models, institutional licences and faculty training will be easier for well-funded campuses to provide. Smaller colleges may have less access to each of them. IIT Delhi’s own guidelines recognise part of this problem by recommending institute-wide licences for advanced AI tools so that access does not depend on an individual user’s ability to pay.
That concern is particularly relevant to India, where institutional resources vary sharply across the higher education system. If access to better AI tools improves academic performance, differences in access could reinforce existing educational inequalities. Policy Circle has previously examined the gap between institutions that can afford AI infrastructure and those that cannot.
Generative AI has exposed an old weakness in higher education. The system has often found it easier to examine the reproduction of information than to measure whether a student can use knowledge independently. That arrangement becomes harder to sustain when software can reproduce the information in seconds.
A university degree is expected to certify the capabilities of the person who receives it. Universities will have to preserve that connection even as AI becomes a routine part of study and work. If they cannot tell what a graduate understands without the machine, the value of the certification itself will eventually come into question.
Krishna Uppuluri is a development professional, educator, and research scholar whose work explores the intersection of public policy, governance, rural development, and institution building.