One-size-fits-all AI
Grading student assignments: does it still make sense?
When you start teaching, you know that the students sitting in front of you are looking for two things: knowledge and a degree. When the urge for the latter becomes much greater than the urge for the former, it sometimes happens that ‘alternative study methods’ are sought out. In the past, we’d occasionally find a cheat sheet during an exam, or a student’s gaze would wander to a peer’s sheet. It is of all times.
As you gain more experience as a lecturer, you develop a sixth sense for this. You quickly smell a rat when solution methods are closely aligned. That instinct has been sounding the alarm constantly for two years now, but I learned to suppress it. Now, it’s almost standard that the submitted solutions are virtually identical. Students no longer need to look at their neighbor’s paper for that; their favorite chatbot helps them put the statistically most average solution on paper time and time again.
This gut feeling became very concrete a few weeks ago when I assigned my annual recurring task around FinOps in the Cloud Infrastructure course. In that exercise, I ask them how much it costs to set up a specific cloud setup based on a number of constraints. The goal is to translate a set of vague descriptions into a concrete list of required cloud components and to estimate a monthly cost for it.
The assignment I use for this is five years old. I deliberately keep it unchanged, amidst a constantly changing set of assignments. That way, I can run a few analyses over the years. For instance, I’m curious about the cost evolution within cloud providers on the one hand, and the ratio between the different providers on the other.
Converging solutions
As AI gains more and more traction, I notice that the solutions for this assignment are also becoming less diverse. In 2022, there was a spread of about 400 euros in the answers; by 2026, that is much less. In this case, that’s good news, because the expected solution is indeed around 350 euros.
I have no idea if I should be happy about this. The fun has drained out of the grading process. I am now reading variations of the exact same solution over and over again. If students used the AI tools to challenge themselves, that’s great. If they went looking for the answer with a quick prompt, that is a disaster for their development as critical thinkers.
Unfortunately, both strategies yield the exact same answer to my assignment. The future will have to show whether this generation of students succeeds in being critical and tackling problems in a structured way.