Does GenAI assessment challenge fit wicked problem theory?

Drawing on interviews with 20 teachers responsible for assessment design at "a large Australian University," Corbin, Bearman, Boud, and Dawson apply Rittel and Webber's (1973) framework for wicked problems to the GenAI-and-assessment challenge, using a deductive "theoretical thematic analysis" that mapped interview responses onto the framework's characteristics by hand. They report that "the GenAI-assessment challenge exhibits all ten characteristics of wicked problems": it "resists definitive formulation, offers only better or worse rather than correct solutions, cannot be tested without consequence, and places significant responsibility on decision-makers." Two characteristics are illustrated at length. First, there is no agreed definition: one teacher frames GenAI as a workforce-readiness issue ("we can't just say to students, you cannot use it"), another frames it as "cheap learning" and "a form of educational fraud," and a third insists it is layered — "a problem of cheating, but it's also a problem of engagement, and it's also a problem of workload." Second, there is no stopping rule: unlike a chess problem or a leaking pipe, "there are no clear criteria for knowing when you have reached the solution," so a teacher who built dual AI and non-AI submission streams did so as a practical trade-off, not a resolution.

The paper's reasoning is that tame problems "have clear definitions and measurable solutions" with true-or-false answers, while wicked problems "lack definitive formulations" and yield only solutions that are "better or worse," demanding "judgment, compromise, and adaptation." It argues that most institutional and scholarly responses — stronger policy, AI-detection software, redesigned assessment — implicitly treat the challenge as tame, assuming "a problem-solution alignment that just needs to be uncovered." Misdiagnosing a wicked problem as tame, the authors write, "typically generates frustration, policy churn, and blame rather than progress," because a technology like AI detection, however imperfect, gets asked to deliver a definitive resolution a wicked problem structurally cannot provide. Their response is to argue that teachers need institutional permission "to compromise, diverge, and iterate," rather than a correct method to adopt.

This reframes what Do university AI policies actually protect what credentials mean? documents as a gap. That audit of 30 universities' public guidance found permission categories (what AI use is allowed) stated more clearly than the evidence standards and safeguards that would show a credential still certifies what it claims — a pattern consistent with institutions treating the challenge as tame, as if naming the allowed boundary were the whole task. This paper supplies the diagnosis for why that gap persists: if the underlying problem is wicked, no permission category, however precisely drawn, can function as a complete solution, since there is no stopping rule by which "evidence standard, solved" could ever be declared. The permission this paper calls for is also a different kind from the audit's — not a rule naming what AI use is allowed, but authorization for each teacher to resolve the trade-off differently, and to stop short of a clean answer, without that being treated as a failure of policy or will.

The excerpt is a single deductive coding exercise, applying one pre-chosen framework to interview data from one institution in one country, with no comparison against teachers who were not primed to think in wicked-problem terms and no account of how the ten characteristics were operationalized or checked across coders. It does not establish that the GenAI-assessment challenge is wicked by any test independent of the framework used to look for it — only that the framework, applied by the researchers, fit what teachers described. The implication the authors draw — that policy churn and teacher frustration will continue as long as institutions treat the problem as solvable rather than wicked — follows only as strongly as that fit holds, and should be read as a reframing argument rather than a measured finding.