Edu Inspirations. How to assess students fairly in the age of AI?
I think the phenomenon we might tentatively call AI fatigue is becoming increasingly common, adding yet another layer to our collective exhaustion with today's digital world.
But reading AI-generated texts is not the only problem I face in my teaching life. There are others, and I'm curious whether you're struggling with them too.
First, when I receive an assignment that a student has completed at home, I sometimes suspect it was generated using AI. But I have no idea how much of it reflects the student's own work. And that worries me, because the last thing I want is to assess someone unfairly.
Second, I see that students approach AI chatbots very differently. Some have access to paid versions, others don't, and some choose not to use them at all because of their personal beliefs. Developing fair assessment criteria under these circumstances is genuinely difficult.
Third, I worry about my students because I care about them. What concerns me most are situations where students submit an assignment, but during a conversation it becomes clear that they don't actually understand what they've written. I find myself thinking that they've wasted their time, and I worry that they don't realise how the habit of generating effortless answers may train their brains to avoid cognitive effort. They may not think about it today, but by doing so, they could be depriving themselves of the opportunity to develop their full potential.
And research suggests that AI use in student assignments is already widespread. According to the 2026 HEPI/Kortext report, 94% of surveyed undergraduate students reported using AI when preparing assessed work (Stephenson & Armstrong, 2026).
So what can we do about all this?
Why don't I believe AI policies and frameworks like AIAS will solve the problem?
According to a 2025 UNESCO survey, 61% of the higher education institutions surveyed had either developed their own AI policies or were working on guidelines. One of the approaches institutions have adopted is the AI Assessment Scale (AIAS), developed by Mike Perkins and his team.
Unfortunately, I don't believe these measures solve the underlying problem. I do think policies matter, and frameworks like AIAS can help make the rules more transparent. But I also believe there is one law of nature that is far more powerful than any of them.
I'm talking about the principle of least effort, discussed, among others, by Nobel laureate Daniel Kahneman in his book Thinking, Fast and Slow (2011).
The idea is fairly simple: if we can achieve something with less effort, we will often choose the easier option. Incidentally, this tendency has probably helped our species survive more than once, which may explain why we're so susceptible to it.
If an assignment can be generated effortlessly, and we as educators don't currently have tools that can reliably determine whether a particular piece of work was produced by an LLM (Weber-Wulff et al., 2023), then quite a few people will take advantage of that opportunity.
That's why, personally, instead of concentrating on rules that can easily be broken, I prefer to design tasks in which using a chatbot where I don't want students to use one simply isn't worth their while.
And I'll let you in on a little secret: I don't want to waste my energy playing detective or judge, trying to establish whether someone has broken the rules. I'd much rather be a teacher, because I genuinely love what I do.
How can we make assessment more trustworthy?
A while ago, while thinking about how to make assessment more trustworthy in the age of AI, I suddenly remembered a cartoon I loved back in the 1990s: Captain Planet. Do you remember it?
The story followed five young heroes who each possessed a ring with a different power: earth, fire, wind, water, and heart. Each power was useful on its own, but whenever they faced a particularly powerful enemy, they combined their powers to summon Captain Planet.
For me, fair assessment in the age of AI works in much the same way. We need to combine different approaches to increase the likelihood that our assessments genuinely reflect what students know and can do. I've accepted that I'll probably never have complete certainty, but I also know that we have several interesting options at our disposal.
So instead of looking for one perfect assessment method, perhaps we should start combining different approaches?
And we don't necessarily need to revolutionise our entire assessment system. Sometimes, small changes can make a surprisingly big difference.
Here are a few ideas I find particularly useful.
I start with prevention: what I call CHATOEDUCATION
These days, we spend quite a lot of time teaching students how to use AI. We introduce them to tools, teach them how to write prompts, and show them how to generate presentations, reports, and analyses.
But we talk much less about the potentially high long-term costs of overusing AI.
That's why I try to talk to my students about what happens when we develop the habit of handing over our own ways of thinking to chatbots.
In my view, we don't talk nearly enough about the fact that outsourcing our unique way of expressing ourselves is a tremendous loss, almost like giving away a small piece of our individuality.
I also believe that prevention, which I've given my own name, Chatoeducation, is one of the most important yet underestimated aspects of this whole challenge.
Whenever I have the time, I do various activities with my students in this area. We also discuss why certain effortful activities are valuable to us and why quick, easy answers are so tempting.
Such conversations and activities may ultimately prove far more valuable than yet another list of prohibitions.
P.S. If you're interested, send me a message. I'd be happy to share some of the activities I use with my students.
I try to understand how the final work was created
Another possible approach is to ask students to document how their work developed. For example, we might ask them to share drafts or selected prompts.
But there's something important here that I think we sometimes overlook.
Documenting the creative process is not nearly as straightforward as it sounds.
Anyone who creates, writes, designs, or conducts research knows that the creative process rarely follows a perfectly organised, linear sequence.
Sometimes we start at the end, go back to the beginning, abandon several ideas, and then come up with our best solution during a walk or just one day before the deadline.
That's why I don't want to require students to construct a beautiful but artificial story showing exactly how they arrived at their final result, step by step.
Especially since that kind of report can also be generated by AI.
I find the idea of a small developmental portfolio, consisting of intermediate pieces of work, much more convincing.
At the same time, I recognise that assessing the process is not easy, so I use this approach only when the nature of the assignment makes it appropriate.
I raise the bar
There's another approach I've been thinking about more and more lately: raising expectations.
If students have access to a powerful tool that can improve their writing, organise the structure of their work, and perform all sorts of advanced tasks, should we really assess their work according to the same standards as before?
Sometimes I quietly think to myself:
Dear student, if you have such a powerful tool at your disposal, perhaps I, as your teacher, should now expect work of the highest international standard?
Sometimes, I also assess students' work based on the quality of the prompts they have developed themselves.
I check what's actually stayed in their heads
I also return to some good old-fashioned methods: pen-and-paper tasks, live questioning, and in-class tests.
During my classes, I try to let students use whatever resources they find helpful: search engines, lecture materials, AI chatbots, and so on.
But I design the final assessment so that they can only rely on what they've actually learned and retained (Roediger & Karpicke, 2006).
Sometimes, I also ask students to defend their work, although this creates practical challenges. For example, how do you organise individual oral defences when you have 150 students in a lecture hall?
There are a few possible solutions, such as written questions or randomly selected oral checks, but none of them is likely to solve every problem.
I design tasks that draw on students' personal experiences
Large language models have extensive general knowledge, but they don't have easy access to our personal experiences.
That's why I try to design assignments in which reflecting on one's own experiences is an important part of the task.
Anchoring an assignment in personal experience doesn't guarantee independent work, but it increases the chances that students will need to do something beyond simply generating text.
It's also worth remembering that LLMs often lack local context unless we provide it.
And they tend to struggle with something very simple: knowing what I said during a class but never included in the lecture notes I later shared with my students.
I change the conditions and see what happens
Imagine that a group of students has prepared an excellent financial analysis of a company.
The report looks good, the calculations are correct, and the conclusions seem reasonable.
During a discussion with the team, we change just one element and ask them to transfer their knowledge to a new situation.
For example, we provide new data or introduce an additional constraint, then check whether students can adapt their solution.
If they genuinely understand the method they have used, they are much more likely to handle this kind of challenge.
A tough nut to crack
To conclude, I just want to say that assessment in the age of AI is a genuinely serious and complex challenge.
That's why I hope we can be a little more understanding towards ourselves when we don't yet know exactly how to approach it.
I've also been thinking that perhaps we should involve students themselves in this process.
Together, we could develop assessment criteria that are clear and fair from their perspective, while also making it difficult to submit an entirely AI-generated assignment without engaging in meaningful thinking and learning.
Do you have own approach to assessment in the age of AI?
Send me a message! I'm genuinely curious, and I always enjoy collecting new ideas.
And who knows? If enough of us join forces, perhaps we'll manage to summon Captain Planet and find some effective ways to tackle the challenges I've described. ;-)
Stay in touch!
Article by Alina Guzik
Subscribe to my newsletter so you don't miss my future articles.
References
Stephenson, R., Armstrong, C. (2026). Student Generative Artificial Intelligence Survey 2026. HEPI Report 199.
Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux. Polskie wydanie: Pułapki myślenia. O myśleniu szybkim i wolnym.
Furze, L., Perkins, M., Roe, J., MacVaugh, J. (2024). The AI Assessment Scale (AIAS) in Action: A Pilot Implementation of GenAI-Supported Assessment. Australasian Journal of Educational Technology, 40(4), 38–55.
Weber-Wulff, D. i in. (2023). Testing of Detection Tools for AI-Generated Text. International Journal for Educational Integrity, 19, 26.
Chi, M. T. H., Wylie, R. (2014). The ICAP Framework: Linking Cognitive Engagement to Active Learning Outcomes. Educational Psychologist, 49(4), 219–243.
Roediger, H. L., Karpicke, J. D. (2006). Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention. Psychological Science, 17(3), 249–255.
National Research Council (2000). How People Learn: Brain, Mind, Experience, and School: Expanded Edition. National Academies Press.
