The arrival of powerful AI systems in higher education has triggered a familiar cycle of excitement and anxiety. Students can now generate essays, solve problem sets, write code, and even summarise entire bodies of literature in minutes. Predictably, concerns about plagiarism, assessment integrity, declining efforts, and even the value of education have followed.
But this framing misses the deeper point. AI does not fundamentally threaten higher education. Rather, it exposes a more uncomfortable truth: Much of what we have been measuring and rewarding in education was never central to it.
At its core, higher education has never been about producing answers or imparting skills necessary for jobs. It has been about cultivating judgement — about learning how to reason, how to justify claims, how to recognise the limits of one’s knowledge, and how to decide what can be trusted. If AI appears to disrupt education, it is only because we have increasingly conflated education with its proxies: Outputs, surface-level coherence, and measurable performance.
Consider a simple example from computer science. AI systems can now generate moderately complex code with ease. But this does not render the study of algorithms and programming redundant. The central issue was never merely whether a program works on some inputs. It has always been about understanding why it works, the assumptions under which it is valid, how it might fail, and whether one can produce a convincing witness — an argument or proof — of its correctness. A program without clearly specified preconditions, postconditions, and invariants is not just incomplete; it is untrustworthy. AI can produce code, but it cannot, in any substantive sense, certify its correctness. That requires disciplined reasoning. As Edsger W Dijkstra famously observed, “program testing can be used to show the presence of bugs, but never to show their absence.”
The same distinction applies across disciplines. A student can produce an essay on the causes of a historical event, but can they distinguish between competing explanations, evaluate sources, and defend their interpretation? A model can report that it is “95 per cent accurate,” but does the student know what that number means, how it was measured, and whether it is even relevant in context? In science, a claim may be supported by data, but has confounding been addressed? Is the conclusion causal or merely correlational?
These are not skills in the narrow sense. They are habits of mind — forms of intellectual discipline and epistemic rigour. They cannot be outsourced.
AI systems, however, are extremely good at producing the artefacts that we have come to treat as evidence of these abilities. They can generate essays that look coherent, code that runs, and analyses that appear sophisticated. In doing so, they destabilise the proxies we rely on. When outputs become cheap and abundant, they cease to be reliable indicators of understanding.
This is why the current “assessment crisis” is so often misdiagnosed. The problem is not that students can cheat more easily. It is that our methods of assessment have been overly dependent on outputs that can now be generated without a corresponding understanding. Take-home assignments and essays without personal interactions, and even coding exercises, were always imperfect measures of learning. AI has simply made its limitations impossible to ignore.
A similar issue arises in research and knowledge production. AI tools can summarise large bodies of literature and generate plausible syntheses. But they can also produce incorrect or unverifiable claims, fabricate citations, and present shallow or misleading conclusions with great fluency. The challenge here is not merely one of academic misconduct. It is a question of epistemic trust. If we cannot reliably distinguish between well-founded knowledge and plausible-sounding fabrication, the integrity of scholarly communication is at stake.
The appropriate response is not to retreat from AI, nor to double down on surveillance and control. It is to re-centre education on what it was always meant to be.
This has several implications.
First, we must shift emphasis from outputs to reasoning. It is no longer sufficient to ask for answers; we must ask for justification. Oral examinations, iterative problem-solving, and open-ended discussions that probe understanding become more important than ever.
Second, we must take verification seriously. Students should be trained to question claims, interrogate metrics, and identify assumptions. In an environment where information is abundant but not always reliable, the ability to decide what to trust is foundational.
Third, we must recognise that intellectual maturity involves an awareness of uncertainty. A student who can say, “This argument holds under these assumptions, but I am unsure whether they apply,” demonstrates deeper understanding than one who confidently presents an answer generated by a machine.
Finally, institutional leadership must resist the temptation to frame AI adoption as a technological upgrade. The real challenge is not whether to use AI tools, but how to align them with educational purposes. Introducing AI assistants without rethinking pedagogy and evaluation risks reinforcing precisely the proxies that are now failing.
It is sometimes suggested that universities face competition from online platforms and AI-driven learning systems. This is misleading. Platforms excel at delivering modular skills and certifications. Universities, at their best, are concerned with something else: The formation of judgement. The real danger is not external competition, but internal drift — towards treating education as a sequence of tasks to be completed rather than a process of intellectual development.
AI, in this sense, is not a disruption of education but a diagnostic. It reveals where we have substituted measurable outputs for meaningful learning, and where we have mistaken fluency for understanding.
The question, then, is not what AI can do. It is what we are willing to accept as knowledge without verification.
In a world where answers are cheap, judgement becomes a scarce resource. Higher education must decide whether it is in the business of producing the former, or cultivating the latter.
The writer is Professor of Computer Science (also associated with the Centre for Digitalisation, AI, and Society), Ashoka University. Views expressed are personal
