RLVSNederlands

AI is creating a sea of sameness

Grading student assignments feels like reviewing twenty variations of the exact same solution.

When you start teaching, you know that with every passing day, you get closer to the day you catch a student cheating. I dare say you even develop a sixth sense for it. With enough experience, you can spot very quickly whether a particular solution was shared among students. At least, you could until recently. AI is increasingly causing us to receive nearly identical solutions across the board.

This became very concrete a few weeks ago when I assigned my annual recurring question: how much does a specific cloud setup cost based on conditions x, y, and z? The assignment I use for this is five years old. I intentionally keep it unchanged every year. That way, I can run some analyses over the years.

The context

AI is a disruptive technology that is already having a massive impact on the IT sector. No need to waste more words on that; nobody needs convincing there.

The dilemma

But how education should deal with it is far less clear to me. Specifically for IT skills like programming, the question quickly arises: what knowledge is still necessary, and at what point can AI take over? On one extreme end of the spectrum is the view that you can only program well with AI if you are already a good programmer; on the other extreme, programming knowledge has become obsolete. I’ll leave it to the experts to decide where it lands, though the truth likely lies somewhere in the middle.

Either way, it is getting more and more attention in the degree program, and that brings new dilemmas to light. One observation, for example, is that AI threatens to quickly lead to higher pass rates for those who have the money to independently buy the most expensive and best models. This becomes clear daily in the lab, where students working with free tools sit shoulder to shoulder with students who sometimes hold multiple subscriptions to the likes of OpenAI and Anthropic.

Towards the next academic year, the decision will likely be made to provide students with an AI subscription so everyone competes on a level playing field. There are several choices here, and there is something to be said for each.

The three options:

  • Option 1: We have all students purchase a Claude subscription. That way, we give them tools on par with what is used in the industry. It’s an extra cost for students, but IT degrees are already very cheap compared to other fields of study anyway.

  • Option 2: You don’t learn to drive in a Porsche either, and education has a moral duty to give European alternatives room to grow and build a user base, even if statistics show those models are roughly a year behind the American alternatives. By having students buy a subscription to Mistral, we send a clear signal and push for a stronger European AI footprint.

  • Option 3: A university of applied sciences is primarily a place where research and development must have a home. Instead of subscriptions for students (which amounts to about EUR 75,000 annually), it would be better to invest in in-house AI servers that students can use and that also benefit research. The experience gained from hosting models ourselves can then be shared with the industry.