OPINION: In the age of GenAI, assessment must reflect the dynamic learning process

This is an opinion piece. The views expressed are those of the author.

Imagine two chemistry students using GenAI to produce polished explanations of a reaction mechanism. One of them submits the first plausible answer GenAI spits out, while the other manipulates a molecular model, evaluates the claim under different conditions, finds a problem, revises the GenAI prompt, and tests the explanation on a new molecule. The two submissions may look alike, but the epistemic activity differs profoundly – written and oral assessments may not catch these differences.

GenAI has exposed an old problem

Shortcut or not, GenAI has exposed a double misalignment in education. First, many of us still operationalise learning through an old information-processing perspective: students acquire knowledge, store it, and retrieve/produce it when examined. Yet learning arises through embodied, situated, and socially and technologically mediated dynamic activity. We actively navigate and make sense of the world through our bodies, movements and interaction. Sense-making is situated in, relational to, and constitutive of our socio-material, economic, and technological contexts. Second, we infer these dynamic learning processes from what is usually a static product (e.g., text), as GenAI increasingly participates in producing it.

As a cognitive and learning sciences researcher, I deeply appreciate the powerful affordances text and similar media offer. Writing and drawing, for instance, externalise, stabilise, or even (positively) destabilise ideas. They support imagination and dynamic reasoning beyond the page. Writing well is thus a central competence in many domains. But like every other assessment form, a written product is a proxy, partial evidence of complex learning processes. The scalability and efficiency of text-based assessments offer genuine practical advantages. Problems arise when these logistical advantages allow one form of evidence to dominate.

Competence in context

Competence is enacted in relation to what disciplines entail. Students learn to coordinate disciplinary ideas, bodily action, data, tools, representations, technologies, and other people. A medical student using AI-supported diagnosis should notice when symptoms conflict with a recommendation, seek missing evidence and justify overriding it. A law student drafting a brief should trace claims to relevant precedents and adapt the argument when a crucial fact changes.

An assessed performance may thus emerge from a learner-technology-world system (just like an essay emerges through interactions with language, pen and paper) and still allow us to assess the learner's epistemic agency (e.g., selecting resources, verifying claims, recognising uncertainty).

More questions and no simple answer

More embodied and dynamic forms of assessment can certainly help us elicit relationally enacted competence. Can we, as educators and researchers, invest more in developing and making such approaches practicable at scale? Are institutions ready for such a shift?

Assessment at university scale will always sample proxies for practical reasons. A big task is to sample more sensibly, and I do not believe there is one, simple solution. Different programmes and courses need different combinations of assessment evidence. So, to develop a richer ecology of evidence aligned with the aimed competences, we must work together. And many of us, consciously or unconsciously, already have been.

GenAI has created an opening for philosophy and practice of education to reconsider what learning and competence mean. Perhaps it is time to shift attention from "Who produced every sentence?" to "How does this student understand, judge, create and responsibly accomplish through the epistemic relations that ‘constitute’ contemporary disciplinary practice?" Education, after all, is bigger than what can be assessed at any given time.