AI in university assessment: do grades themselves have to go?

Since I started teaching at the University of Amsterdam, roughly 20 years ago, assessment has become more and more schematic. Rules and regulations have proliferated. So have forms, check-lists and online registration portals.

At the same time, learning goals and assessment have been coupled more tightly. We shouldn’t try to impart skills or knowledge on students unless we also test whether they have acquired them. In consequence, learning goals that cannot be assessed well – say, intellectual maturity – have been sidelined. “Digressions” in lectures about issues we care about as teachers but that don’t show up in multiple choice exams are out. The same goes for exercises or activities that we believe benefit student growth even if they don’t reappear in some test. More and more, what and how we teach is shaped by what and how we (can) test.

This combination – pedagogical work that follows an assessment logic plus more and more rules – marginalizes open-ended forms of engaging and sharing our knowledge and experience with students. Nevertheless, the time and sweat that lecturers expend on assessment has continued to rise. Even before ChatGPT, the balance between learning and assessment was already off.

Enter AI

There is much to be said about AI in education, but for assessment, the message is simple. We cannot be sure that an assignment completed outside the university hasn’t been done with the help of AI. (Any detection tool that works today won’t work tomorrow.) And AI is capable. If you’re a good prompter and upload the relevant materials, AI can do your statistics assignments or write your essays in one go. If necessary, prompt a second or third time, and you’re done for sure. With assignments completed at home, we don’t know what we’re testing: whether someone is a master prompter, or whether they put in the work themselves. Unless we want to test how well students can use AI, take-home assignments are done as a form of assessment.

In my department, that pushes assessment back into “controlled settings” – on-site exams of various shapes (including oral ones). Two problems follow: first, these assessments further push up workload. On-site exams come with loads of extra requirements, more difficult arrangements for resits, etc. Oral exams in particular eat into our time and are relatively inflexible. Already under pressure, lecturers have even less time to keep up with academic literature or developments in their scholarly field. The quality of their teaching suffers.

Second, many of the things we might want to teach are impossible to assess in controlled settings. Papers were meant to let students do much of the work themselves and report back final results. They require the ability to perform complex tasks on their own. We also let students write papers to allow them to explore topics that interest them independently. Chewing on a topic is an integral part of the paper-writing process. No oral exam can mirror that.

As AI pushes us towards testing in controlled settings, our teaching follows – because assessment and learning goals need to match. The tail (assessment) wags the dog (learning goals) even more, and university degrees increasingly resemble learning styles that students know from school. Intellectual autonomy suffers.

What happened to Bildung?

When I ask colleagues what the key thing is the want to teach students, the most frequent answer is “critical thinking”, rooted in solid evidence. That might also be my first response. And even though students themselves might not use this label, I believe that is what many of them want, too. When we manage to create moments of real engagement in our classrooms, genuinely deliberating about issues that they and we care about – that is when I see them most alive and most satisfied. Then they seem to know why they are here.

Excess assessment and a focus on box ticking and accumulating grades has always stood in the way of that ideal. AI supercharges that dynamic. It has come to a point where assessment, if we follow its logic and the limits AI imposes on it, can become the enemy of the learning and personal growth both teachers and students value most. Already, extrinsic motivation (the need to study quickly, to accumulate study points, to get high grades, not to bump up against any rules like attendance requirements, etc.) can bury students’ intrinsic motivation – their innate hunger for understanding things. How and what they study is driven by the incentives we dangle in front of them and by the competitive pressure they experience, not by their eagerness to learn.

It is conceivable – if unfortunate – that the labour market asks for people who have been trained in such an instrumental mindset. I don’t think it is our mission to produce them. I also don’t think it serves students well. We are not a factory churning out office workers but central institutions for the development and transfer of knowledge, in all its complexity. If anything, as digital technology flattens society and the quality of public discourse and deliberation, it is our duty to stick to our mission and to push back.

What way forward for assessment in an AI-heavy age?

If there is something to the idea that grading can be the enemy of learning, the consequence should be obvious: stop grading. That may sound radical. But I think it is worth at least entertaining that line of thought.

As a teacher, I want to give students feedback on their written work – tell them what I think was strong, what wasn’t, and what could have been done differently. How do I make sure they haven’t used AI, one way or another, when they put together the text I find in my Inbox? There is only one way: by taking away the incentive to let a machine work for them – in other words, by abolishing extrinsic motivation. If students are free to choose whether or not they submit written text to me, and I don’t grade it but give them substantive feedback, there is no point for them to submit something written with AI. The only reason to send me their text is because they want my feedback. So: no obligation, no grade.

That sounds drastic. But what is the alternative? Not providing feedback on things written at home? That would be a loss. We want and need students to be able to execute larger projects independently. Should we grade written take-home assignments after all? Before we know it, we grade AI prose and reward capable prompters, not sharp thinkers. Neither of those options is better than voluntary exercises.

With voluntary, ungraded exercises we'd actually have more time to go into the substance because less would be wasted on box-ticking, perfunctory rubrics and the like. And we’d be free to be critical even vis-à-vis students whose work we otherwise appreciate. If we don't have to attach grades to it, the substantive feedback we would give students would get better, not worse.

Part of me wants to get rid of grades altogether. Substantive feedback is for students and their learning. Grades, in contrast, are ultimately for future employers – and they are not the ones we should be serving as university teachers. We pay too high a price if much of our pedagogical work is driven by having to pre-sort students for the job market. Let employers figure out themselves who they think is good for the job they have to offer. (The rumour has it more and more of them use AI for that purpose, anyway…)

There are intermediate solutions, however: how about concentrating assessment, with grades and all, during two moments per academic year – whole weeks in which we dedicate all our resources to genuinely sophisticated testing? We’d invest major resources in making sure that there, we actually assess what we want to assess, not only rote memorization of factoids.

Of course, the question is how we can suddenly test students after months of free learning with little external pressure. Fair point. But that is exactly the logic I worry about: that the need to offer grades at the very end – the ones mentioned on the diploma – works its way backward through everything else we do. The diploma grade is a poor measure of whether we have given students the chance to grow their minds and personalities in a way that they want, we cherish, and society needs. As AI supercharges that dynamic and possibly leads it ad absurdum, it is worth asking whether the time hasn’t come to reserve course fundamentally and make teaching about teaching again, not about scoring points for a piece of paper.