Why should students incur huge debts to undertake a leisurely three-year (or more) degree when a machine can be prompted?
UNSW vice chancellor Attila Brungs is quite correct to argue that universities should teach students to use artificial intelligence and to focus on deep thinking and problem-solving.
That does, however, raise the somewhat embarrassing problem that Australian universities already claim to teach critical thinking, problem-solving, communication, creativity and professional judgment.
Of course, AI should be used to eliminate unnecessary “grunt work”. The difficulty is in differentiating between clerical labour and intellectual apprenticeship.
Many tasks that look like grunt work are part of how students learn. Reading difficult material, checking a claim, reconstructing an argument, discovering that a plausible answer is wrong, and revising a judgment are not “make-work”. They are the very training of an educated mind.
Universities should improve the quality of student thinking. They should make it more disciplined, better informed, more sceptical of weak evidence, and more capable of being defended under pressure. If universities cannot demonstrate that change, the claim to teach deep thinking is mostly branding.
AI can search, summarise, recombine, classify, draft, and generate alternatives. At best, it can be a useful learning assistant. It does not know which problems are worth solving, which assumptions matter, which sources should be trusted, or which conclusions should be defended under uncertainty.
There is an old Socratic point here. Learning to prompt well is the old educational problem in new language – a good question is often worth more than a good answer. A weak student asks AI for a conclusion. Stronger students ask it to test assumptions, identify missing evidence, compare explanations, generate objections, or apply a principle to a new case.
“The challenge now is how to design assessment tasks in which AI use is expected, visible, and insufficient on its own.”
Prompting well requires creativity, context, clarity, judgment, and the ability to recognise when an answer is not good enough – capacities that depend on prior learning.
Assessment is the single largest AI challenge that universities face.
The current response to AI is being built around “permissioning”, compliance, and policing. A good assessment requires students to demonstrate what they know, what they can do, and how their thinking has evolved. A declaration that AI was or was not used tells us very little.
This is where the so-called “authentic assessment” is especially exposed. The movement away from supervised pen-and-paper examinations was sold as a move towards “real-world” relevance. Yet in practice, authentic assessment often meant unsupervised essays, reports, reflections, and case studies. Too often, those tasks confused relevance with evidence of learning.
AI has made that weakness impossible to ignore.
The answer is full-scale adoption
The widespread requirement of detailed rubrics compounds the problem. It used to be that how a student went about answering the assessment was itself the assessment task. Now the smart students have worked out that the rubric is the AI prompt to an A-grade or high distinction.
The answer is full-scale adoption and not permissioning. AI tools are cheap, useful, accessible, and increasingly embedded in ordinary software. Universities cannot build an assessment system around pretending students will not use them. Nor should they. Students will use AI in the workplace. They should learn to use it well at university.
The challenge now is how to design assessment tasks in which AI use is expected, visible, and insufficient on its own. Universities need educational entrepreneurs who can devise assessment tasks for an AI-rich environment. That is a much harder task than adding an AI statement to a course guide.
Right now, universities face a double exposure; the claim that they already teach deep thinking and problem-solving cannot survive scrutiny, and the assessment technology that feeds directly into credentialing has been disrupted.
The university business model is in trouble on both the demand and supply sides. Why should students incur huge debts to undertake a leisurely three-year (or more) degree when a machine can be prompted? Why should employers pay the graduate wage premium for work that a machine can produce on demand?
Then there are the organisational challenges of incorporating AI into workflows and products. I have previously argued that AI adoption requires significant and difficult organisational changes. Universities also face this problem – on steroids. Vice chancellors have far less authority than their remuneration packages suggest. They can announce as much mandatory AI use as they like. Universities are decentralised professional bureaucracies. Faculties, schools, disciplines, accreditation bodies, academic boards, compliance systems, student expectations, unions, and legacy assessment rules all create veto points.
The challenge for universities is to rediscover and redefine their value proposition. To be fair, this is a problem that all knowledge industries face, as I recently argued.
The good news is that AI should allow universities to raise standards. If students have powerful tools, more can be expected of them. The danger is that universities respond with policy language, compliance forms, and another layer of assessment bureaucracy.
