How ChatGPT can help disrupt assessment overload

Recent developments in generative AI – not least the notorious ChatGPT – offer opportunities to stimulate long-overdue reforms to assessment practices. But these reforms need careful thought and principled implementation within an already hard-pressed sector. To make space for these developments, now is an opportune moment to tackle the issue of assessment overload, too many assessments and limited opportunities for students to revise work iteratively.

Modularised higher education has led to an assessment arms race in which academics inadvertently compete for student attention using grades as both carrot and stick. Students often end up jumping through hoops rather than achieving sustainable learning, and an inevitable consequence is that if something is not assessed then students won’t do the work.

• So, you want to use ChatGPT in the classroom this semester?
• AI text detectors: a stairway to heaven or hell?
• How generative AI like ChatGPT is pushing assessment reform

Meanwhile, end-of-semester bottlenecks, with students juggling multiple deadlines, lead to rushed work or inadvertently encourage students to take shortcuts through copying, plagiarism or outsourcing to humans or bots.

Reducing the assessment burden could support trust in students as individuals wanting to produce worthwhile, original work. Indeed, students can be co-opted as partners in designing their own assessment tasks, so they can produce something meaningful to them.

A strategic reduction in quantity of assessment would also facilitate a refocusing of assessment priorities on deep understanding more than just performance and carries potential to enhance feedback processes.

If we were to tackle assessment overload in these ways, it opens up various possibilities. Most significantly there is potential to revitalise feedback so that it becomes a core part of a learning cycle rather than an adjunct at its end. End-of-semester, product-oriented feedback, which comes after grades have already been awarded, fails to encourage the iterative loops and spirals typical of productive learning.

So, what are the implications of generative AI for feedback? Given the capabilities of ChatGPT to generate speedy feedback on work in progress, students could be encouraged to document and evidence the process of their learning. Staged assessments with built-in cycles of feedback are valuable in their own right – while countering the risk of one-shot products generated mainly or entirely by ChatGPT. Staged assessment sequences might involve, for example: a one-minute elevator pitch as stage one; a short, annotated bibliography as stage two; drafts shared with peers or chatbots, then revised further, as stage three onwards.

Staged, process-oriented assessments could involve: students trying out various prompts for ChatGPT; evaluating and building on its outputs; learning to detect AI hallucinations and inaccuracies; relating content to other modules or key readings; or personalisation to real-life events in regional or national contexts. These kinds of processes might enable students to evidence how they have added value by curating and adding to AI-generated content. Acknowledgement of AI inputs to student work is expected, but perhaps not in so much detail that audit trails become overwhelming for lecturers.

Assessing process as well as product could be valuable but time-consuming for lecturers, hence the need to reduce the overall assessment burden. With our ever-decreasing attention spans and the prevalence of short messaging platforms, clear and concise written communication is at a premium. Can a standard 3,000-word essay be profitably shortened or rethought entirely?

AI advances also prompt consideration of assessment types. Do universities tend to privilege written forms of assessment at the expense of oral ones? After all, the workplace needs staff who are effective in both oral and written communication.

Oral forms of digital assessment can be engaging, personalised and authentic in preparing for the world of work. Because of their spoken, interactive nature, they also engender increased student accountability and often lead to worthwhile learning outcomes.

Digital oral assessment includes forms such as video presentations, podcasts or vlogs. These are often more appealing to students than conventional written assignments. Whereas speaking in public can be anxiety-inducing, with these recorded forms of presentation students can practise and refine work in the comfort of their own environment.

Again, digitally enabled oral presentations should be concise. If the three-minute thesis can become a worldwide phenomenon, undergraduates can be encouraged to produce meaningful contributions of short duration. Less can sometimes be more, because concise communication does not equate with dumbing down. In fact, it is often more challenging to prepare a short presentation than a longer one.

A lifelong learning imperative for all of us is honing the capacity to work productively with generative AI. Lecturers and students need to learn together how to use AI constructively, responsibly and ethically, including how to critique and build on its contributions. Its potential to enable new forms of feedback is promising but needs careful preparation, support and staff development.

Educational specialists have been recommending assessment reform for decades but inertia, workloads and conservatism are perennial barriers. Assessment in the generative AI era is likely to be more challenging and time-consuming for staff. Something has to give, and reducing the quantity and length of assessments would be a good starting point.

David Carless is professor of education at the University of Hong Kong.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.