Source: The Conversation (Au and NZ)
Since the introduction of ChatGPT in late 2022, there have been rising concerns about university assessments. Are we testing students’ learning? Or simply their ability to use generative AI?
Universities and teachers have been working hard to reform assessments to ensure degrees still genuinely demonstrate students’ knowledge and skills. But it’s an ongoing challenge for the sector.
How are universities responding, and are we making progress?
AI doesn’t always equal cheating
In July, the US software company Turnitin published a report showing nearly 54% of Australian uni students’ submissions, processed through its system from October 2025 to April 2026, showed some form of AI use. About 10% of these submissions contained more than 80% material generated by AI.
But the presence of AI in assessments doesn’t automatically mean all students who use it are cheating.
Students may use AI selectively in their work. For example, by asking it to help develop an idea, edit a sentence, improve a draft, or provide feedback on an argument.
Many students also use AI to support their assessments well before they start writing. As part of research I conducted with colleagues in 2024, we interviewed 24 students to find out how they use AI to assist with reading academic texts. We found more than half of students used AI to summarise texts, clarify difficult readings and check their interpretations.
Rules aren’t always clear
But research shows the line between what is acceptable and unacceptable use of AI is not always clear to students or teachers. For example, does improving a draft with AI mean only correcting grammar, or does it mean restructuring the argument?
One of the biggest challenges is the line between appropriate and inappropriate AI use can differ between assessments and might also change for each student even within the same assessment.
How are unis responding?
Australian universities have a regulatory requirement to ensure assessments can confirm students have achieved the knowledge and skills their courses require. Universities’ approaches in how to go about this have differed in recent years.
An influential response is the University of Sydney’s “two-lane approach”, with variations now used at many Australian universities.
The two-lane approach separates assessments into two categories. The first lane is “secure” and requires students to demonstrate their skills and knowledge without any use of AI. These assessments are typically in-person exams and interactive oral assessments, where assessors can physically see students completing a task without opening ChatGPT to help them.
The second lane is “open” and allows students to freely use AI, for example, in online quizzes and essays. Because it is so hard to judge if AI has been used inappropriately in such formats, lane two assessments are typically framed as being for feedback and learning, rather than as demonstrations or assurances of student learning.
Unis are still struggling to keep up
But increased reliance on in-person assessments is not a perfect long-term solution. These assessment formats:
-
are often expensive to run
-
limit the kinds of things teachers can assess
-
may also unfairly bias against certain student demographics.
In-person assessments also don’t necessarily mean AI is beyond students’ reach. Wearable forms of AI, such as AI enabled smart glasses, can resemble ordinary eyewear and provide exam assistance to students without showing obvious signs to invigilators.
Students have already been caught using them in exams in New Zealand, Taiwan and South Korea, and monitoring the tech remains a challenge for unis.
Are unis on the right track?
Our research describes AI and assessment as a “wicked problem” – it doesn’t have a simple, one-fix solution.
Part of the complexity of this challenge is that universities don’t always know if students are using AI to help their learning or to undermine a teacher’s ability to judge if learning actually took place.
Universities need credible evidence of what graduates can do. But treating assessment solely as a security problem to deter AI use risks measuring the success of the control measures over the success of a student’s education.
Perhaps instead of simply asking whether a new form of assessment works in an AI world, we should ask which problem it solves, and which other problems it might create.
For instance, if the aim is to keep AI out of assessments, then more supervision and tighter controls may look like progress. But if the goal is to prepare students to use AI well for the workforce, those same restrictions might prevent the learning we want. This suggests that what a “solution” looks like in one context may look like anything but in another.
If universities are to make progress, they will need to understand which approaches to assessment reform work – and which do not – in particular disciplines and contexts, rather than a singular “perfect” response across an entire institution.
Thomas Corbin does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
Original source: https://analysis1.mil-osi.com/2026/11/10/ai-has-turned-university-assessments-into-a-wicked-problem-and-we-dont-have-all-the-answers/