More supervision of university exams to combat AI use

The University of Tasmania will place greater emphasis on in-person and supervised assessments to ensure students are not relying entirely on artificial intelligence to pass their degrees.

Under a new initiative targeting excessive AI use in higher education, university courses in Tasmania will now have to prove to external auditors that students have completed certain learning outcomes.

As part of the standard co-developed by UTAS and human-certification company Proudly Human, the university would need to prove that it has an appropriate number of in-person, versus unsupervised, assessments.

Developers have said it aims to address the eroding trust in university degrees, which is occurring off the back of excessive AI use by students to complete assignments.

The University of Sydney introduced a similar model two years ago.

'Old-fashioned' in-person assessments

Co-developer and UTAS pro vice-chancellor Luke Bereznicki warned there were consequences if the university did not think carefully about whether assessments were held in-person or online.

"What that can mean is that a student can attain an award and we don't actually know if the student themselves has done the work or whether they've used AI tools to do the work,"

he said.

Professor Bereznicki said under the standard, the university would be focused on ensuring students were completing assessments under 'secured conditions' such as supervision, so that they couldn't rely on AI to complete assignments.

"It's the student themselves, so the identity of the person has been verified to be the enrolled student. It's the authorship, which means it's the student that produced the work and was observed doing the work.

"Those are the three key things that confirm secure conditions. So identity, authorship, and observation."

Co-developer and Proudly Human founder Alan Finkel said UTAS would need to lay out the key learning outcomes students had to achieve in each course, and then ensure that they were being assessed appropriately, meaning a balance of supervised and unsupervised assignments.

"Secure assessment conditions are a little bit old fashioned, but it's the human assessment. Writing an essay in class or giving an oral defence or sitting a supervised exam or for law students who could be doing practice sport,"

he said.

"Our role will then be to come in once or twice a year … as an auditor."

Mr Finkel said it was "100 per cent" the university's responsibility to determine what the assessments involved, but that in-person assessments were preferred.

"[If there are online assessments], they have to set up procedures that they can defend as ensuring that the student does the work authentically themselves. And then it's up to the university to convince us as an auditor that that is working," he said.

 "We will sit down with [the university] and say, 'tell us the process you're using and let's imagine scenarios where a student might try to do this and do that'. 

"'How would your system ensure that they can't use AI to do the assessment under those various scenarios?'"

Students worry value of going to uni declining

The president of the Tasmanian University Student Association, Jack Oates Pryor, said some students worried that the perceived quality of their university degrees had dropped, in the wake of AI being used.

"Certainly, a general fear about what university degrees mean when other students are abusing AI," he said.

"I've talked to some who are concerned about how industry will perceive the quality of their degree when graduating.

"I think that goes back to the importance of students working with the university to ensure really fit-for-purpose approaches towards learning, teaching and research, as well as stronger transparency from the university as to how it is approaching AI."

Following in USYD's footsteps

It is not the first time an Australian university has returned to in-person assessments to combat excessive AI use.

In 2024, the University of Sydney introduced a 'two-lane approach' to how it conducted assessments, to try and ensure that students were graduating with the necessary skills required in their courses.

Assessments were sorted into two categories, or lanes.

One was 'secure assessments', where students were monitored while completing the assessment, such as an in-person exam.

The other was 'unsecured assessments', where the student could not be observed, such as a take-home essay.

Courses would strive to have a balance of both, with guidance provided around unsecured assessments and how appropriate they were.

Standards targeting AI necessary

Deakin University education professor Thomas Corbin said universities increasingly needed to introduce Standards around AI and assessments.

"One [reason] is that there is a lot of uncertainty about what assessment can and should be doing after AI is in the mix,"

he said.

The other, he said, was the legal obligation universities had to demonstrate that students "really can do or do actually know whatever their degree certifies that they are supposed to".

Professor Corbin explained that universities were more frequently resorting back to traditional methods of testing students, such as in-person exams and oral assessments.

"They're invigilated and interactive in a way that we hope allows us to see a student as a student themselves, as separate from the AI," he said.

He said teachers in higher education were constantly challenged by the need to rethink how they were assessing students, in order to keep up with the pace of AI.

"A big challenge with AI is not necessarily figuring out ways to make assessment secure, but figuring out what should be assessed, what is important to be learning and what we value now in this age of AI,"

he said.

"I think a lot of people who thought we had good approaches and could just rely on exams are now reeling from the rise in wearable AI in a way that's forcing us to rethink assessment again as the technology evolves."

Professor Corbin explained teachers were constantly juggling the pressure of ensuring assessments were interesting and meaningful for students, while being robust enough for the degree to be taken seriously by the general public and industry.