Yes, these days everyone is talking about artificial intelligence. However, there’s another thing that accompanies us wherever we find ourselves: recommendation systems. Disguised as one thing or another, each with its own intention behind it, these systems recommend us products (advertisers call this »personalised notifications«, whereas for us recipients, they are simply »commercials«), social media posts, YouTube videos (naturally, again, with the aim of presenting us ads, i.e. personalised notifications).

Is it possibe to explain how such systems work in a simple way?

This is exactly what we set out to do in the fourth grade of the Alojzij Šuštar Primary School in Ljubljana. We first divided the students into groups. Each group received 42 cards with posters of cartoons, and each student picked six cartoons that they liked best. Then, each member of the group needed to write down on a piece of paper the names of all the other members of their group and next to their names, how many of the same cartoons those students picked. In the next step, each student circled the names of the three students on their list that were the most similar to them. The students then drew a »sociogram«, in which arrows pointed from each student to the ones who were the most similar to them. Finally, using one colour, each student wrote next to their name the numbers representing the cartoons they themselves picked, then using another colour, the numbers representing the cartoons picked by the students they were connected to. And, voila, here are the recommendations.

We repeated the exercise at scale: the groups were given tablets, which they used to enter their choices into the webpage that we created for this activity. We read the data into the computer and got a similar sociogram as before, but this time for the whole class, and the recommendations based on the selections of all the students not just on those of individual groups.

So what have we learned? Let’s start with ourselves: the lesson is engaging. The topic is just right for the fourth grade, it pulls them in. The system works solidly, but it would be better if the original selection was of a set of seven (rather than six) cartoons; that way, we would get a bigger cross section and would still not run out of the extra cartoons for recommendations. We also saw that the activity doesn’t only take one hour, but two – and it even spilled over into the recess, since each student wanted to see their recommendations. The new Recommendations widget that has been created especially for Pumice will be adjusted a bit; the information about the selection of a particular person is actually not really interesting, so we will only display recommendations. That way, we will also be able to look at all of them within one lesson.

And what have the kids learned? »That we now know how YouTube works.« (Yes, YouTube has served as an example all along, and the activity in fact started with the students guessing how its recommendations work.) »This activity was really good and interesting.« Of course, the students also had additional wishes; mostly, to watch the cartoon that was voted the most popular. Some were also interested to know what application we used. Unfortunately, there was not enough time left to tell them about Orange. Someday, someday.

Let’s conclude with a remark made by one of the students: »I really liked it!« So did we.