BUSINESS REFLECTION: Crossed Wires: Degrees of doubt — how generative AI is fracturing higher education

I have a couple of university degrees. They have been a signal from my alma maters to the outer world. A signal that says: we, the universities that awarded these certificates, put our reputation on the line and guarantee that this fellow learnt something. We know, because we checked. He may not have been the sharpest pencil in the box, but neither was he the dullest, so when he arrives at your door looking for a job, you can be reasonably sure that he paid his fees, studied a bit, read a bit, thought a bit, and wrote and passed some tests. And to repeat: we know because we checked and have been doing this for a very long time, so please trust us.For the first time in pedagogical history, these ramparts are starting to crumble. Universities today can no longer make that claim with the certainty they used to. Soon they might not even be able to make that claim at all.Consider this. At 10pm on a Sunday, two hours before the deadline, a student submits an essay on the causes of the World War 1. It is lucid, elegantly structured, free of typos, and not too different from the student’s previous submissions during the year. The marker, who has read 60 similar documents since dawn, awards it a first. Neither party is sure what has just been certified. The student knows the text was produced by a chatbot in 40 seconds. The marker suspects something of the sort but cannot prove it. And the university, which has had a single, largely unchanged method of assessment for about two centuries, knows that trouble is at hand.So, here is the uncomfortable question facing the academy: If the essay can be written by a chatbot, the code by a copilot, and the problem solution by a reasoning model, what, precisely, does a degree certify? That a student has mastered a subject, or mastered the art of prompt engineering between seminars?The numbers are no longer in dispute. A survey by Hepi, a British higher education think-tank, published in February 2025, found that 92% of undergraduates now use generative artificial intelligence (AI) in some form, and 88% use it in their assessed work. And remember that February 2025 was in the dark ages of AI – the models are orders of magnitude more capable now, with better research, better reasoning, better problem solving and more “natural” prose. A number of more recent studies support these figures. This points to another perplexing matter for higher education regulators. In the past, guardians of higher education, well-educated themselves, have been careful and thoughtful in designing evidence-driven policies – there is little knee-jerking in these deliberations. But the very nature of this scientific approach is at odds with the speed of AI’s progress. Research from 2022 is useless, data from 2023 is near useless, data from 2024 and 2025 are dodgy and uneven, and current studies of student behaviour are constrained by the cadence of scientific publishing. The collection and distribution of evidence is too slow, and no one knows how to deal with that.What about tools to catch the cheaters, if they can even be considered that? In most cases, they are also problematic because of the number of false positives reported. For instance, an independent test of 14 tools found none exceeding 80% accuracy. No one wants to falsely accuse 20% of a student body. Or 2%, for that matter – either would be a legal and ethical nightmare. And so the University of Cape Town switched its detector off in October 2025. Waterloo, Curtin, Vanderbilt, Washington State and dozens of others have followed. Australian Catholic University generated nearly 6,000 allegations in a single year, a quarter of which collapsed on investigation, before abandoning the tool entirely.Technical workaroundsOf course, tools do get more sophisticated, but so do clever technical workarounds (there are many, some advertised openly, including those now doing an end-run around the new Claude provenance tool, announced only last month). Proscription, sanctions, bans, public shaming and threats are blunt objects that simply won’t work.In any event, the numbers of AI-written assignments are now of such epidemic proportions that regulators have noticed. Britain’s Quality Assurance Agency, the body that advises on who may award degrees, has conceded in its own guidance that some or all of the assessments contributing to the evidence base “may no longer be confidently ascribed to an individual student”. In other words, we are not sure the marks belong to the people we are giving them to. It is difficult to imagine a more serious admission, or one that has produced less consensus action.Then there are the students themselves, many seeking AI-usage guidance and not getting it. A February 2026 study showed that 86% of the students reported using AI, yet only 23% felt properly prepared to use it and most cited inadequate institutional guidance. Let’s state this brutally – near-universal student usage and disturbing institutional neglect. I took a quick look at AI-usage policies around the world (yes, I used three different AI research tools to do this), and they are scattered, non-uniform and in some cases bewildering in their near-total absence of reasoned discussion about what students are actually learning (compared with the pre-GenAI era), and how they can tell.And so what to do? Many universities are experimenting with the obvious – oral examinations (the way it was done before the 1800s and all the way back to Socrates’ “viva voce” in Athens in the fifth century), or multiple-choice exams. The former creates an unfair time burden on examiners, is expensive and does not scale. The latter is only partially applicable – it can probe only a certain kind of knowledge. Yet no one seems to have come up with much more. Which leads directly to this whispered and blasphemous possibility. Relegate the university degree to a certificate of attendance only, and shift the knowledge assessment burden to industry bodies and guilds – the medical licence exams, the legal internships, the accounting board tests, the probationary apprenticeships, the face-to-face recruitment interviews. After all, those are the places where knowledge gets put to work; perhaps they are more appropriate venues in which to determine whether the applicant knows anything. The university would then be reduced to saying: “We checked, and we know that he was here.” I am deeply uncomfortable about this suggestion; it seems like a tragic surrender. The educational institution has always been among the proudest and noblest of human endeavours. It has produced almost all history’s great minds.Surely some of them can find a way to solve this conundrum? DMSteven Boykey Sidley is a professor of practice (ex-JBS, University of Johannesburg), a partner at Bridge Capital and a columnist-at-large at Daily Maverick, where he writes the weekly Crossed Wires column. His new book, It’s Mine: How the Crypto Industry is Redefining Ownership, is published by Maverick 451 in South Africa and the Legend Times Group in the UK/EU, available now.