You Can’t Decolonise Assessment Without Touching AI
How formative GenAI work reveals the power hiding inside the concepts we teach
Universities are tying themselves in knots over generative AI. Policies proliferate, detection tools promise certainty on shaky grounds, and assessment redesign is often framed as damage control, while doing little to ensure students are prepared for the world beyond the university.
Beneath this procedural noise, however, sits a more uncomfortable – and potentially productive – question: what happens when AI does not just help students write essays, but actively reshapes how they understand the world as students and citizens?
In International Development classrooms, that question is especially urgent.
Concepts such as poverty, migration, sovereignty or the Global South are not neutral abstractions. They are historically produced, politically loaded and deeply entangled with colonial power. Yet students are routinely asked to deploy them as if they were settled facts. When generative AI enters the picture, those assumptions are not challenged – they are accelerated.
In this context, an approach to teaching and research developed by Dr Christoffer Guldberg at King’s College London is particularly instructive. Rather than treating AI as either a threat to assessment and knowledge production or a shortcut to be managed, it places generative AI at the centre of a formative and emancipatory exercise designed to expose how concepts themselves are constructed, racialised and normalised.
Text‑to‑image AI is used as a mirror of coloniality, opening up questions about creativity, knowledge and power.
The method is straightforward. Students and research participants choose a core concept from their module, lived experience or research question. They generate outputs using a text‑to‑image AI tool, and then analyse what appears: Who is visible? Who is absent? What stereotypes are reinforced? What histories are flattened into a single image or definition?
The visual nature of the exercise is crucial. Racialised and gendered dimensions of coloniality that are often otherwise hidden or implicit become immediately visible. This allows for dialogue across differences in literacy and language, and supports more inclusive participation.
Crucially, the process does not stop at analysis. It intervenes in the aesthetics of coloniality itself. By using AI as a canvas, students and participants add layers of embodied knowledge, experience and interpretation on top of what the system produces.
This results in tangible artefacts, often in the form of memes. These function as acts of critique, disrupting the authority of the AI output and opening up space to imagine alternative ways of being and knowing. They contest the homogenisation of culture and knowledge that AI systems can reproduce.
While these memes can express frustration, anger or sadness, their frequently humorous nature is particularly productive. Humour punctures rigidity. Concepts that rely on appearing natural or inevitable lose their authority when they can be laughed at, reworked or visually contradicted.
In doing so, participants experience the difference between the apparent fluency of AI-generated outputs and human creativity as something embodied, situated and unfolding in time. When students redraw or reframe AI outputs, they force concepts back into dialogue – with the technology, with each other, and with the teacher or researcher. The meme becomes both commentary and pedagogy, and a shared site of contested knowledge.
The power of this approach lies in its honesty and its flexibility. It recognises that generative AI already functions as a knowledge broker. Banning it does not restore critical thinking; it simply pushes its use out of sight. Nor does encouraging students to “use AI responsibly” achieve much if we do not examine what AI is responsible for reproducing.
Instead, the method foregrounds knowledge as process: choosing concepts, interrogating outputs, and reflecting on how language, personal histories and geopolitical assumptions shape what AI returns. Feedback is immediate, dialogic and collective. Participants move from being users of AI to critics of its aesthetic and political authority.
This does not avoid difficult questions. Digital literacy gaps remain. Ethical objections to AI must be taken seriously. Competitive tool use risks reproducing inequality, even as it can generate productive dialogue. These tensions should not be smoothed over. They become material for critical engagement. AI is neither neutral nor taboo; it is a political object to be examined.
For anyone involved in assessment design or participatory research, the lesson is clear. We cannot claim to teach critical thinking or support emancipatory practices while leaving the machinery of concept production untouched. Research and teaching must engage with AI in ways that acknowledge both its potential and its limitations.
Decolonising assessment does not begin with tighter rubrics. It begins by showing students and research participants that even the most confident outputs – essays, datasets, images or AI responses – carry histories, values and power. And that those, too, can be argued with.
Find out more at Decolonising AI and memes with Freire and Bergson and for examples of memes produced in workshops at KCL, Advance HE and the University of Westminster and UCL see here.
Dr Christoffer Guldberg is a Lecturer in International Development at King’s College London. His work focuses on decolonial pedagogy, visual methods, and critical approaches to AI and knowledge production.