The Wicked Problem of AI and Assessment

Source: Assessment & Evaluation in Higher Education (Corbin, Bearman, Boud, Dawson) · 2025-09-03

ABSTRACT Generative artificial intelligence (GenAI) has created significant assessment challenges in higher education. Universities and teach­ ers have responded to these in various ways, including by way of new policies, revised assessment formats, and technological safe­ guards. These responses are typically predicated on an assumption that AI in assessment is a problem that can be solved if only the right approach can be found. However, in this paper we argue that GenAI may not be that kind of problem at all. Drawing on inter­ views with 20 teachers responsible for assessment design within a large Australian University, this paper applies Rittel and Webber framework of wicked problems to analyse educators’ experiences with assessment design in the context of GenAI. Our findings demonstrate that the GenAI-assessment challenge exhibits all ten characteristics of wicked problems. For instance, it resists definitive formulation, offers only better or worse rather than correct solu­ tions, cannot be tested without consequence, and places signifi­ cant responsibility on decision-makers. In the light of this redefinition of the AI and Assessment problem, we argue that edu­ cators require certain institutional permissions – including permis­ sion to compromise, diverge, and iterate – to appropriately navigate the assessment challenges they face.

Generative artificial intelligence (GenAI) challenges established assessment practices in higher education (Luo 2024; Jensen et al. 2024). Educators are grappling with how to preserve assessment validity and students appear concerned about being falsely accused of inappropriate AI use (Wu et al. 2024). Simultaneously, the new technology appears to offer powerful aids, potentially including improving assessment feedback or perhaps assisting students to better understand assessment expectations (Corbin, Tai, and Flenady 2025a; Sporrong, McGrath, and Cerratto Pargman 2025).

In this paper we argue that there is reason to think the GenAI-assessment challenge exhibits the hallmarks of what Rittel and Webber (1973) term a ‘wicked problem’. Wicked problems, as opposed to ‘tame’ problems, do not have ‘correct’ or ‘incorrect’ solutions (Rittel and Webber 1973). This does not mean there are no ways forward, nor does it mean that all ways forward are equally valuable. However, it does mean that responses must look very different. For one, they require a shift from seeking definitive answers to engaging in ongoing, adaptive work shaped by competing priorities and evolving conditions. Perhaps more importantly, it also means that the heavy burdens placed on those positioned to navigate the challenges, most obviously university teachers, are both substantial and – at least to a certain extent – unavoidable.

Wicked problems, as originally conceptualised by Rittel and Webber (1973), describe challenges that defy simple solutions. For problems of this kind there is no ‘silver bullet’, as Keep and Mayhew (2014, 765) put it, that will solve the problem. They are complex, contested, and continually evolving. Unlike their counterpart ‘tame’ problems, which have clear definitions and measurable solutions, wicked problems lack definitive formulations, and their solutions are not true or false but rather better or worse, requiring judgment, compromise, and adaptation. This distinction is key because it disrupts the assumption that there is a ‘correct’ policy, assessment method, or institutional response waiting to be discovered.

Many influential institutional and scholarly responses to the GenAI-assessment problem in the higher education literature assume that the challenges it poses are tame rather than wicked. Recall that a tame problem is one that can be clearly defined and solved with the right intervention, for example through technological interventions. The clearest case of such an intervention is likely AI detection tech­ nology. The problem of AI and assessment, from at least one standpoint, is that for any given piece of writing teachers cannot tell what a student wrote themselves and what they outsourced to AI. If that is the problem, AI detection software could be seen as the correct solution, as, at least in theory, it can see beyond what the human teacher can and point them to areas of the text written by GenAI. Although he is sceptical regarding the likely success of these tools, as Ardito (2025) writes, ‘Teachers naturally seek methods to control and guide AI usage in the classroom, and a working system to enable AI usage detection would be the most immediate, intuitive, and easy-to-implement solution to ensure academic integrity’. Underlying this view is an assumption that there is a problem – solution alignment that just needs to be uncovered. In this study, we look to teachers’ accounts of assessment design to see if this is indeed the case.

GenAI. We suggest that such an analysis matters precisely because misdiagnosing the problem type leads to counterproductive responses. Treating wicked problems as tame typically generates frustration, policy churn, and blame rather than progress. If the AI challenge is tame, then conventional solutions (such as stronger policies, better detection tools, or redesigned assessments) should eventually provide definitive reso­ lution. But if it is wicked, then these approaches, while potentially valuable as partial responses, may be insufficient when positioned as complete solutions, potentially creating unrealistic expectations and cycles of policy frustration rather than progress.

We read and re-read our interview data and discussed it collectively. The over­ whelming impression was how teachers were wrestling with the situations they found themselves in, and there were no easy assessment design answers, irrespective of how experienced the teachers were. This then led to a decision to analyse our interview data deductively, using a theoretical thematic analysis (Braun and Clarke 2006) guided by Rittel and Webber (1973) framework for characterising wicked problems. Interview responses were systematically mapped to these characteristics. This process was conducted manually by the researchers.

The first defining feature of wicked problems is that they cannot be clearly or con­ clusively defined. Unlike technical problems where stakeholders can in theory agree on what needs fixing, wicked problems mean different things to different people and these varying definitions pull solutions in contradictory directions. Without agreement on what the problem is, a singular, cohesive response becomes impossible.

This lack of formulation was apparent in the markedly different ways participants framed the problem of designing assessment to accommodate the rise of GenAI. For some, the challenge was fundamentally about workforce readiness: ‘If they’re using it in the workforce already [...] we can’t just say to students, you cannot use it’ (T5). This framing positioned resisting GenAI as professionally irresponsible with the corollary that the solution lies in integration into teaching. Another participant framed the issue very differently, casting GenAI use as a form of educational fraud: ‘It’s cheap learning, because students end up finishing university knowing zero, having learned zero from day one to the end’ (T20). This framing positioned a response of restriction and policing, directly contradicting the integration approach.

Some teachers explicitly acknowledged that the problem contained layers of problems. For example, one stated that: ‘We talk about GenAI as a problem of cheating, but it’s also a problem of engagement, and it’s also a problem of workload’ (T3). Relatedly, when addressing similar tensions, another stated that ‘It’s hard, basically. I don’t know. I’m at a loss. I keep on trying to have conversations with people, and people seem to be at a loss too’ (T12). Other teachers also used this expression, finding themselves ‘at a loss’. Consider for example the frustration of the teacher who stated: ‘I’ve spent so much fucking time on developing this stuff.

The second defining characteristic of a wicked problem is that it has no stopping rule – that is, there are no clear criteria for knowing when you have reached ‘the solution’ (Rittel and Webber 1973 p. 160). Unlike solving a chess problem where checkmate signals definitive completion, or fixing a leaking pipe where success can be measured by the absence of water, wicked problems offer no such clarity. As Rittel and Webber present it, ‘The planner terminates work on a wicked problem, not for reasons inherent in the “logic” of the problem. He stops for considerations that are external to the problem: he runs out of time, or money, or patience. He finally says, “That’s good enough”’.

For participants, these trade-offs manifested practically in assessment design. One teacher created dual submission streams – one with and another without AI – in an attempt to balance competing learning objectives.