Rethinking Assessment in the Age of Generative AI

Almost all students now use generative AI (genAI) to support their coursework (Freeman, 2025), often without guidance. While this may seem like a levelling force, given that AI has been found to disproportionately benefit the productivity of less experienced and lower-skilled workers (Brynjolfsson et al., 2025), it is not. In practice, weaker students are particularly at risk for three reasons. First, by relying too heavily on genAI, they avoid the cognitive effort required to structure arguments, critically weigh evidence, or revise their writing. This directly sabotages their learning, as cognitive engagement is key to learning (Gerlich, 2025; Kosmyna et al., 2025). Second, the weaker students often struggle particularly to judge the quality of the outputs they receive, and as a result, their grades often remain low. Finally, and worse still, their underlying difficulties become harder for educators to detect. The illusion of fluency masks a lack of understanding. In other words, while grade distributions may appear to converge, with fewer failing essays, the actual learning is diverging.

This creates a genuine challenge for educators. The easy response - banning genAI from assessments altogether - is appealing in its simplicity but flawed in practice. Some advocate a return to in-person essay exams as a way of ensuring students’ work is entirely their own. But this, too, has limits. GenAI makes it easier than ever to prepare detailed essay plans for a wide range of potential prompts. In this context, a timed, handwritten response may ultimately test nothing more than memory and recall. The appearance of rigour is not the same as actual learning.

Moreover, this approach risks reinforcing an outdated model of education that ignores how professionals in nearly every field are already incorporating AI into their workflows. 

If we prohibit its use, we may prevent students from developing skills that are not just relevant, but increasingly essential: how to use genAI critically, selectively, and ethically.

We may end up widening the gap between academic training and workplace expectations.

On the other hand, allowing genAI into assessments is not without complications. If students are encouraged - or even allowed - to use it, we are implicitly assessing their ability to prompt, evaluate, and edit AI-generated content. That may well be a valuable skill. But we should only assess it if we also teach it. Otherwise, we risk rewarding prior digital literacy or socio-economic advantage, and penalising those who are still learning how to use the tools effectively or are concerned about the ethical implications of their use.

We are thus caught between two unsatisfactory options: banning genAI and missing an opportunity to develop essential skills, or allowing it and shifting the burden of AI literacy onto students who may not be adequately prepared. Neither route is ideal - and both underscore a deeper issue. Changing how we assess is difficult. It’s not just a question of redesigning an assignment. It’s about adjusting what we value, teach, and reward.

Finally, changing the assessment is risky. Trialling a new method without knowing how students will respond can result in unusually high or low marks. While high grades may raise concerns about standards, low ones can prompt formal complaints. In a time-pressed, risk-averse institutional environment, this makes experimentation hard to justify, especially when curriculum changes require long lead times and layers of approval. Often, we try to redesign a plane while it is in flight.

Still, change is necessary. The presence of genAI in students’ lives is not temporary. It is not a passing trend to be contained. It is a shift in the knowledge landscape that demands thoughtful, measured responses. It doesn’t mean rushing into fully AI-integrated assessments. But it does mean investing in staff development, creating space for trial and error, and, most importantly, engaging students in the process. They, too, are navigating this shift, and without explicit support, we risk leaving behind those who need the most help to learn how to think critically with, rather than despite, the tools at their fingertips.

References

Brynjolfsson, E., Li, D., Raymond, L. 2025. Generative AI at Work, The Quarterly Journal of Economics, Volume 140, Issue 2, May 2025, Pages 889–942, https://doi.org/10.1093/qje/qjae044

Freeman, J. 2025. Student Generative AI Survey 2025. HEPI Policy Note 61 https://www.hepi.ac.uk/2025/02/26/student-generative-ai-survey-2025/

Gerlich, M., 2025. AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), p.6.

Kosmyna, N., Hauptmann, E., Yuan, Y.T., Situ, J., Liao, X.H., Beresnitzky, A.V., Braunstein, I. and Maes, P., 2025. Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv preprint arXiv:2506.08872.