New research published this week by Wonkhe asks a question that initially sounds like it is about AI but turns out to be about something older and, in some ways, more uncomfortable. The research, based on a national survey of over a thousand students and a series of focus groups involving student representatives from across disciplines and levels, finds that the single most important factor in determining how students use AI is not institutional policy, the availability of detection tools, or peer behaviour. It is whether there will be a future moment when students will need to demonstrate that they actually understood what they produced.
When that moment is clear, students describe using AI to learn. They prompt it carefully, question its outputs, and compare their own reasoning with it. When no such moment occurs, the same students describe using it on autopilot. The accountability moment, the research suggests, influences not whether AI is used but how. Assessment design, rather than AI use itself, appears as the main factor most strongly linked to whether learning happens.
That is, on its face, a striking finding. But perhaps the more uncomfortable observation in the research is what it implies about the period before AI arrived.
A problem that predates the tool
The research reports that 47 per cent of students worry their grades do not reflect what they actually know, and that 38 per cent admit submitting work they could not fully explain. Fourteen per cent say they would feel anxious if asked to explain the reasoning behind their last submission without notes. These figures suggest something broader than AI use alone. They point, rather, to the relationship between production and understanding within the current assessment model.
The research is quite straightforward on this point. The gap between submitting work and understanding it is not new, nor is it mainly caused by AI. It has existed for as long as higher education has assessed learning by asking students to produce written artefacts. What AI has done is make it significantly easier to produce the artefact without any of the learning. However, the assessment system, as it is designed, could not reliably tell the difference between a student who understood and one who produced something plausible. The research indicates that this was true in 2015, and even earlier. AI has changed the scale and visibility of the gap, but not its origin.
This reframing appears important because the sector’s main response to AI has been to ask how to detect it. The research indicates that this is, at best, an ineffective answer to the wrong question. Approaches focused on detection tend to identify students who are contemplating what is and isn’t allowed, while largely failing to address those with the highest risk tolerance. The students bearing the psychological burden of institutional incoherence are arguably those whose principled engagement with AI use most deserves support, not surveillance.
When did we remove the accountability function?
The question this research prompts for me and it is genuinely a question rather than a conclusion, concerns how and when the accountability function of assessment was removed. The shift away from examination-heavy assessment, which gained momentum from the 1990s onwards, was, in most institutions, well-intentioned. Concerns about the fairness of single high-stakes performances, a desire to evaluate a broader range of student abilities, and a realisation that the traditional examination rewarded a particular kind of performance under pressure were all reasonable motivations. The move towards coursework, portfolios, projects, and continuous assessment reflected genuine pedagogical thinking.
What is less clear, in hindsight, is whether that transition involved a deliberate decision regarding the role of the accountability function that examinations had served. Not the examination itself, as that specific format has its own issues, but the function it fulfilled: the moment when a student needed to be present, to think, and to demonstrate understanding in real time without relying on the artefact. Whether that function was intentionally replaced or simply allowed to fade away as assessment design evolved seems, in retrospect, a question worth considering.
The research shows that students themselves seem to understand this distinction quite well. When asked about assessment design in focus groups, they suggested not returning to traditional exams but rather redesigning accountability: low-stakes iterative attempts, conversational assessment, and normalising peer explanation as part of learning rather than just for formal settings. They were not arguing against being asked to demonstrate understanding; they were advocating for being assessed in ways that felt developmental rather than punitive.
What this asks of senior leaders
For those of us who have spent considerable time working with academic standards, curriculum design, and quality enhancement processes, the research raises a question that is worth honestly contemplating. The assessment frameworks that most UK institutions currently use were developed and refined over decades through programme approval processes, annual monitoring cycles, and periodic review. They were created by individuals with genuine expertise and a real commitment to student learning, for conditions that may no longer entirely apply.
It is possible that the accountability function was not simply removed but rather distributed: spread across formative assessment, feedback loops, tutorial conversations, and other elements of the learning architecture that, taken together, were meant to perform the same role as the examination did alone. Whether these elements are functioning in that way, and whether students perceive them as fulfilling that role, seems worth examining rather than assuming.
The research does not demand that institutions change their approach to assessment design. Instead, it asks a more specific and possibly more challenging question: whether institutions can justify their decisions, explain where accountability now resides within the assessment structure, and genuinely evaluate whether it is fulfilling its intended role.
Whether that question is currently being asked within programme teams, academic boards, or in conversations between senior leaders and their academic communities, and what the answers are, seems to me to be one of the more important issues the sector could focus on this year.