Universities should be audited for whether learning outcomes are met

Fifty years ago, the genetic makeup of undergraduates was virtually identical to what it is today. The best estimates are that there would have been three or fewer differences per 100 million nucleotides. And yet that generation of authors, physicians, physicists, chemists and economists were able to win Nobel prizes without the slightest bit of help from artificial intelligence.

As we aspire to do better, we should not risk doing worse by throwing out what worked for them. They developed their intelligence by learning and internalising knowledge, by writing essays, by participating in class discussions and by writing up their practical work. Further, their performance was rigorously assessed to ensure they achieved the desired learning outcomes.

Even in a world dominated by access through internet search engines to virtually all the knowledge on planet Earth, students need to know things so that they can contribute to a conversation without having to constantly check their smartphones. And those who want to compete for the human-in-the-loop jobs that will be created will still need deep domain knowledge and critical thinking skills.

The Norwegian government gets it. This June, it announced a ban on generative AI in elementary schools and massively restricted its use in the upper school years. However, there is no ban for university students, and many of them are cheating by outsourcing both their content-gathering and their thinking to AI – while institutions are mostly looking the other way.

Indeed, many universities actively encourage AI use. They focus their teaching on “prompt engineering”, the supposed key skill in the future workplace. But students don’t need their universities to teach them how to prompt AI; young people work it out for themselves. What they need to be maximally effective at it, though, is to know what they are looking for and how to evaluate what the AI gives back. That is, they need to know stuff and they need to be able to think clearly.

Both of these skills take practice to master. The most robust solution is expensive and old-fashioned: in-person, supervised assessment. That does not mean that 100 per cent of the end-of-unit mark should come from a three-hour exam handwritten into an exercise book – but such exams could be a part of it. Other robust assessments include evidence of clinical competence for nurses and doctors, facility in laboratory experiments for engineers and scientists, performance in tutorials, oral defences of essays and theses, and in-person supervised essay-writing throughout the semester.  

The solution must ensure, under secure conditions, that every student achieves all the required subject learning outcomes. Secure conditions imply that the student’s identity is verified and that the student independently performs the assessment under supervision. If students are sick on the day of the assessment they must undertake the equivalent assessment another time.

This is referred to by some educators as “lane one” assessment. It is a series of hurdles, all of which must be cleared. The non-critical learning elements, known as “‘lane two”, can openly encourage appropriate use of AI, furthering students’ ability to productively use AI tools.

There is no more essential teaching responsibility for a university than to ensure learning outcomes. The problem is that under current conditions, the cost of doing so will not necessarily pay off for those that commit to it. Other than by reputation, the degrees they grant will be indistinguishable to a prospective employer from degrees issued by other universities that fail to ensure the learning outcomes. And given the national and global mobility of graduates, plus the huge number of universities there now are, employers cannot be expected to be familiar with the teaching reputations of every job applicant’s alma mater.

This awareness gap could be overcome by an internationally accredited annual audit process that led to an internationally recognised certification mark proving that the university in question ensured that learning outcomes were achieved – and every learning outcome of its lane-one assured assessment process was a genuine hurdle.

As in a financial audit, the starting place would be agreed standards – but how the standards were met would largely be up to the university. After all, there is no single, right way to teach or assess, any more than there is a single right way for a publicly listed company to name and structure its accounts: it all depends on its business needs.

A company can use general-purpose spreadsheets or highly sophisticated specialist programs to manage its accounts and procedures. What brings comfort to the investors and assurance to the regulator is that an independent auditor reviews those accounts and procedures and issues a report, which is presented to the board and the shareholders.

Similarly, an assured-assessment university certification mark, issued by the auditor, would proclaim to employers, parents and sceptical friends that someone’s degree came from an institution where every graduate demonstrated every learning outcome.

Such a mark would contribute to reversing the declining reputation of universities, re-establishing their role as key contributors to the future of our children. Otherwise, rampant AI use will be the beginning of the end of universities as institutions of higher learning. School-leavers will choose alternative employment pathways that are cheaper, faster and more trustworthy.

Alan Finkel is founder and executive chairman of Proudly Human. He was formerly the chancellor of Monash University, Australia’s chief scientist, founder of Stile Education and founder of Axon Instruments. He is the author of Getting to Zero (2021) and Powering Up (2023). He has a PhD in electrical engineering and was a postdoctoral research fellow in neuroscience.