As AI redefines how students learn, higher education must redesign how learning is assessed—and what constitutes meaningful evidence of it.
{{ vm.tagsGroup }}
10 Jul 2026
6 Min Read
Ishaanaah Ravi (Alumni Writer), Nellie Chan (Editor)
As AI redefines how students learn, higher education must redesign how learning is assessed—and what constitutes meaningful evidence of it.
For years, higher education has relied on a relatively stable model for demonstrating learning. Students are assessed through familiar markers of academic performance, including the coherence of an argument, the clarity of expression, and adherence to assessment criteria. Within this framework, written work has long served as credible evidence of learning—something that can be read, measured, and rewarded with reasonable confidence.
The rise of generative artificial intelligence (AI), however, has disrupted this stability. Writing that was once interpreted as a reflection of a student’s understanding can no longer be taken at face value when the process behind it may involve varying degrees of human and machine input, blurring the boundaries of authorship. In response, concerns over AI-related academic misconduct have grown, prompting institutions to strengthen efforts to detect and deter its misuse.
Yet this response has also introduced an unintended tension: authentic writing is increasingly subject to scrutiny, particularly when it is well structured, clearly expressed, or stylistically sophisticated—precisely because it resembles assisted writing. Even punctuation has entered the debate. If the em dashes caught your attention, you are not alone. They have come to be read as an informal signal of AI assistance, despite providing no evidential value in themselves.
What, then, constitutes meaningful evidence of learning when written work that once demonstrated it can now be replicated by machines—or interpreted as implicating their use?
This tension is felt most directly in the classroom. On one side of the desk are students. For those who view assignments primarily as tasks to be completed, AI offers an easier route through them. Yet for those who see them as opportunities to engage with knowledge, AI makes that path more complicated. Although their perspectives differ, the effect is the same: instead of focusing solely on demonstrating learning, students are increasingly required to demonstrate authorship.
This places students in a new kind of double bind: write too polished and risk suspicion; write too plain and risk criticism. Bound by this contradiction, students who are genuinely engaged must now navigate a shifting set of expectations about how their work should be perceived, how it should be produced, and how they should prove that it is, in fact, their own. Ironically, some students have even begun using AI to make their writing appear less like AI.
Yet not all students bear this burden equally. Those who can afford paid subscriptions may have access to more capable AI models, giving them a further advantage over peers who cannot. Nor do the systems used to detect AI perform equally across different groups of writers. Research has shown that some GPT detectors are more likely to misclassify essays by non-native English writers as AI-generated, placing students writing in English as an additional language at a further disadvantage.
Once students begin writing for interpretation rather than expression, teachers, on the other side of the desk, face a parallel dilemma. They are no longer assessing only what students understand, but also trying to ascertain how that understanding was produced. Time that might have been devoted to supporting learning is now spent scrutinising polished essays for signs of AI, without any definitive way to verify those suspicions. Some institutions now use AI-based detection systems as part of assessment, raising serious questions about who—or what—is ultimately trusted to judge authorship.
Taken together, these demands reveal a paradox in modern higher education: the more institutions attempt to protect academic integrity, the more they risk undermining the trust and fairness on which it is built. Stricter policies can weigh on conscientious students, detection systems can cast legitimate writing as suspect, and the pressure to prove authorship can draw attention away from learning itself. What emerges as a response to misconduct points beyond individual behaviour towards a reconsideration of the role assessment is meant to play.
That reconsideration begins with recognising that this is not merely a technological problem, but a structural one. Pedagogy and assessment are guided by distinct logics: one prioritises cultivating understanding, while the other prioritises evaluating it through measurable outputs that can be compared and categorised. However, the logic of measurement has come to dominate, with marks and grades becoming the central currency through which learning is recognised and rewarded—not because they reflect learning most meaningfully, but because they render it standardisable, scalable, and legible to institutions.
The outcome is a model that privileges products over processes, a model that AI has since complicated. Within contemporary pedagogy, it is positioned as a tool that can support the learning process, enabling students to broaden their thinking and deepen their understanding. Yet once that process is translated into an assessed product, it is viewed through a different lens. Rather than greeting AI with the same openness, assessment meets it with surveillance and scrutiny. The disconnect becomes clear: students are encouraged to become digitally fluent in one context, yet the very fluency they demonstrate is treated with suspicion in another.
The challenge, then, lies less in adapting education to AI than in ensuring educational structures are capable of adapting themselves. For a generation growing up amid personalised recommendations, real-time information streams, and AI-generated content, technology is no longer an influence on learning; it is part of the environment in which learning takes place. When institutions respond to AI primarily as a disruption, they risk reinforcing distrust rather than developing discernment.
The consequences extend beyond the classroom, resulting in students who are prepared to prove themselves within assessment systems but less prepared to participate actively in the world beyond them. The competitive edge will belong neither to those who reject AI nor to those who rely on it uncritically, but to those who recognise that while AI can assist in learning, it cannot assume the responsibility of understanding.
What this calls for, fundamentally, is a reevaluation of what constitutes evidence of learning. If written work—or indeed any assessed work—can now be prompted and generated in ways that are difficult to distinguish from unaided authorship, then the final product alone cannot be made to carry the full weight of proof. Meaningful evidence must make visible not only what students produce, but also the process behind it: how they interrogate ideas, examine information, and challenge assumptions in the pursuit of understanding.
None of this means abandoning written work as a form of assessment. It should instead be situated within a broader body of evidence. Accordingly, assessments must be designed to allow students to demonstrate what they can do with technology through reflection, collaboration, and application, rather than to prove that they can do without it. Ultimately, the future of higher education depends not on preserving the methods used to measure learning before AI, but on preserving the purpose of learning itself in an AI-enabled world: cultivating understanding and the capacity to think with, not merely around, new tools.
Ishaanaah Ravi is a Bachelor of Education (Honours) alumna from Taylor’s University. She enjoys breaking down educational topics into bite-sized insights and finds joy in simple pleasures, like volunteering, watching comedy shows, and building Lego sets.
