I’ve been writing about AI and assessment for a while now, and if there’s one thing I’ve learned, it’s that nobody has the answer. Not the institutions rolling out detection software, not the departments switching everything to oral exams, and not the faculty members quietly pretending AI doesn’t exist.
Corbin, Bearman, Boud, and Dawson (2025) put a name to what many of us have been feeling. In their study “The Wicked Problem of AI and Assessment,” they argue that AI in assessment isn’t a technical problem with a correct solution. It’s a wicked problem. And wicked problems, by definition, resist resolution.
I think they’re right. Accepting that changes how we should approach the whole conversation.
Why AI and Assessment Doesn’t Have a Clean Solution
Corbin et al. interviewed 20 unit chairs at a large Australian university about their experiences redesigning assessment in response to generative AI. What they found is messy, contradictory, and deeply human.
Teachers don’t even agree on what the problem is. Some see AI as workforce preparation: “If they’re using it in the workforce already […] we can’t just say to students, you cannot use it” (p. 5). Others see it as academic fraud, worrying that students may finish “having learned zero from day one to the end” (p. 5).
When you can’t agree on the problem, you definitely can’t agree on the solution. And that’s a defining feature of wicked problems. How you frame the challenge determines what responses seem reasonable. If AI is an integrity threat, detection dominates. If it’s a professional skill, integration takes over. If it’s a workload issue, the conversation becomes about resources. Everyone’s trying to solve a different version of the same challenge, and then getting frustrated when their solution doesn’t travel well across departments.
The Trade-offs Are Inescapable
One of the most compelling sections in the paper deals with trade-offs. Teachers in the study describe designing dual assessment tracks, one with AI and one without. Increasing security reduces creativity. Oral assessments improve authenticity but become logistically impossible at scale. One teacher did the math: oral exams for 250 students would require “2500 min” of administration (p. 9). Try fitting that into an already overloaded semester.
Perkins and Roe (2025) made a similar argument in their chapter on the future of assessment. Retreating to proctored exams doesn’t solve anything. It just moves the problem. And wearable AI devices will eventually complicate even controlled environments. Corbin et al.’s data confirms this from the ground: teachers are making trade-offs, not finding solutions.
There’s also no reliable way to test whether those trade-offs worked. One participant put it this way: “If a student uses AI appropriately for brainstorming, we might never know. If they use it inappropriately, we also might never know” (p. 7). Detection tools don’t help either. A teacher ran their own writing through detection software and got flagged. The tools we’re counting on can’t do what we need them to do.
The Stakes Are High and Personal
This study takes teachers’ experiences seriously, and that’s what makes it valuable. These aren’t abstract policy debates. Teachers are making consequential decisions under uncertainty, and they feel the weight of every one.
One participant asks a question that should give us all pause: “Are we in fact sending students out into the workforce who can get through an interview, but when they start doing the job, they can’t?” (p. 11). The reputational anxiety is just as intense: “That is my biggest concern because at the end of the day, it’s the reputation of the university, not necessarily the student” (p. 8).
Choi, Jang, and Kim (2023) found that trust and ease of use are the strongest predictors of whether teachers adopt AI tools. The authors further show us what happens when neither exists. Isolation, accountability without support, and a sense that AI is moving faster than anything around them: “AI is moving too quickly for universities. We can’t keep up” (p. 10).
Three Permissions That Could Help
The most practical contribution comes in the discussion. Corbin et al. propose three “permissions” that institutions should extend to educators.
Permission to compromise. Trade-offs are unavoidable. Teachers need space to make imperfect but defensible decisions, and to say openly what’s being prioritized and what’s being sacrificed.
Permission to diverge. What works in a 15-person seminar doesn’t apply to a 250-student lecture. Context should drive design. Variation across departments is responsiveness, not inconsistency.
Permission to iterate. Redesigning an assignment and discovering it doesn’t work isn’t failure. It’s learning. Institutions need to normalize revision and stop treating change as an admission that something went wrong.
These permissions are wise and necessary. The AI Assessment Scale developed by Perkins, Roe, and Furze (2024) offers a structured framework for making intentional decisions about AI integration. But even that framework works best in a culture that supports experimentation and tolerates imperfection.
My Perspective
I advocate for intentional, pedagogically grounded AI use across education. I still do. But Corbin et al. helped me see something I sometimes understate: this is genuinely hard for teachers. The uncertainty is structural. The stakes are personal. Institutional support is often thin.
Accepting that AI and assessment is a wicked problem doesn’t mean giving up. It means letting go of the expectation that one policy, one tool, or one redesign will settle things. It means building adaptive cultures where professional judgment is valued, where iteration is normal, and where imperfection isn’t punished.
We don’t need a silver bullet. We need better navigation.
Reference
- Corbin, T., Bearman, M., Boud, D., & Dawson, P. (2025). The wicked problem of AI and assessment. Assessment & evaluation in higher education, 1-17.
https://doi.org/10.1080/02602938.2025.2553340 - Perkins, M., Roe, J., & Furze, L. (2024). The AI Assessment Scale revisited: A framework for educational assessment (Preprint). December 2024. https://arxiv.org/abs/2412.09029
- Perkins, M., & Roe, J. (2025). The end of assessment as we know it: GenAI, inequality and the future of knowing. In AI and the future of education: Disruptions, dilemmas and directions (pp. 76–80). https://durham-repository.worktribe.com/output/4472558/the-end-of-assessment-as-we-know-it-genai-inequality-and-the-future-of-knowing