1. i. I think a very clear example would be the use of AI in our workplace (I work as a software developer). When we use AI to create or maintain products, we have a very high performance (we get a lot of work done in a short amount of time, and the clients are very happy with that), but the learning is low. This is especially noticable for new collagues, who have very little knowledge of what the system or product actually works like. ii. A lot of people think that ideal learning should feel effortless and easy. So if something gets done quickly and with high quality (aka high performance), we feel like we’ve learned a lot, while in reality we might not have learned anything (because learning and performing is not the same).

^je kan alleen performance toetsen.

2. i. “‘Gewenste moeilijkheden’ die inspanning vergen resulteren in meer leeropbrengsten dan minder inspannende methoden zoals highlighten of het bekijken van filmpjes.” -> this is a direct reference to the “desirable difficulties”, discussed in the article by Yan et al. “Studenten die zichzelf overhoren worden geconfronteerd met wat ze nog niet goed beheersen, en overschatten zichzelf minder snel.” -> this is a direct reference to the “testing effect”, also discussed in Yan et al.

ii. I think there is very little to learn about the use of AI. A fundamental understanding of the technology (how it works), would be useful for most people, because it allows you to understand when AI can be, and when it is not, a suitable tool. Because there are technological limitations / inherent characteristics of LLMs that should be taken into consideration (eg. hallucinations, probabilities, source-use etc.) But other than that, no hands-on experience is needed in my example. Especially, since the “best practises” for “prompt engineering” change every few months, because of the rapid pace of AI-development. And in the end, its not hard. You type a message, the agent replies. The “competent use” of AI is determined entirely by the ability of the user to distinguish good and bad responses from eachother, and their ability to understand when AI is producing good or bad work. Because that steering is exactly what makes some people’s AI-use better than others. Because they understand what the AI is doing and have the ability to say “do it in a different way” and guide it in the correct direction. And this ability to do that, is driven entirely by domain knowledge. In other words: competent AI-use is only possible if the user is competent in the first place. And how do you get that competency? Exactly: learning it yourself. (Also this is my opinion and I in no way claim to be objective. This is based purely on my experience of AI-use in the workplace.)