AI and Assessment: A Wicked Problem
I remember a colleague saying, “I’m not excited about ‘thinking’ being optional now,” a few months after OpenAI released ChatGPT. I was still teaching high school at the time, and we were standing on the softball field during a fire drill, alarms blaring in the distance.
The anxiety was real. Suddenly, students could produce work in a few seconds without thinking, blurring the line between thoughtful practice and mindless performance. Coming so soon after the pandemic, the exhaustion in his voice was palpable.
If some instructors held early hopes that these tools would just go away or that detection tools would regulate them, three years on, we should recognize that neither is realistic. The tools are here to stay, and automated detection remains unreliable. And, the tools continue to rapidly evolve both in their capabilities and our understanding of how to use them effectively. They may not always live up to the hype of the most ardent techno-optimists, but they are providing real value and changing how we work. Most of our students will likely spend their entire careers in workplaces infused with AI.
Still, in our classes, assessing real learning remains a challenge with no clear solutions. Over the past three years, I have watched thoughtful instructors redesign assignments, experiment with formats, tighten and loosen policies, and adjust the weight of in-person exams up and then down again. While the work has been serious, the results have been fleeting.
To move forward, we must revise our mindsets as much as our assessments. In “The wicked problem of AI and assessment”, Thomas Corbin et al. (2025) frame the challenge as a wicked problem. Unlike “tame” problems, which have clear definitions and measurable solutions, wicked problems are “complex, contested, and continually evolving.” Their solutions are not "true or false" but rather "better or worse." As Corbin argues, GenAI impact on assessment is not a problem to be solved but a condition to be navigated.
Viewing the problem through this lens provides the following important permissions:
- Permission to Compromise: Understanding that there is no perfect solution allows us to be more honest about tradeoffs in designing assessments and more willing to accept and learn from failures when we experiment with new approaches.
- Permission to Diverge: Successful approaches to this problem are going to be different in different contexts. We should see divergence in practices as a thoughtful response, not an institutional failure.
- Permission to Iterate: Assessment design has always been iterative, but the need to evolve is even more pronounced now. We should expect to make changes and build time into our workflows to support that rather than seeing the need to continuously update assessments as a failure.
These permissions are not a lowering of standards. They are an acknowledgment of a current reality that requires adaptation.
My colleague’s fear that thinking would become optional was understandable but also incomplete. The potential power of AI tools makes critical thinking, thoughtful judgement, and deep disciplinary knowledge more important than ever. Our role as educators is to explore these shifts and do our best to prepare students for the changing world they are entering through course and assessment design. Reframing assessment as a wicked problem does not remove the difficulty. Instead it gives us the freedom to be adaptable, iterative, and experimental in our approaches.
If you would like to explore these ideas further and collaborate with colleagues about how to approach AI and Assessment in a deliberate way, please consider registering for the AI and Assessment: Reframing the Challenge workshop.
Note on AI use in this article: Generative AI tools were used during the final revision stages of this article to help clarify language and shorten sections.
Thomas Corbin, Jack Walton, Peter Bannister & Jean-Philippe Deranty. (2026) On the essay in a time of GenAI. Educational Philosophy and Theory 58:3, pages 198-210.
