How do YouTube, TikTok, and similar platforms guess which videos we’ll enjoy? How do social networks recommend posts and help us connect with friends? How do online stores know which products to show each visitor?

In this activity, students explore how recommender systems work behind the scenes. The topic connects directly to their own lives: younger children pick their favorite cartoons, while older students reflect on the series they enjoy most.

By establishing simple rules to determine a gnome’s profession based on its appearance, students gain a hands-on introduction to machine learning and classification models. In the first part of the activity, students identify the rule themselves; in the second part, they draw gnomes’ houses according to their own rule, and the computer’s task is to infer the rule the students had in mind.

By gradually introducing different kinds of animals and their distinguishing features (such as has wings or gives milk), the computer learns to differentiate between mammals, birds, insects, and other groups.

Students often become highly engaged as they try to think of animals that “break” the current rules and force the introduction of new features. The a-ha moment typically comes at the end, when they realize that the same principle can be used to teach a computer to recognize poisonous mushrooms, predict which drug will be effective for a patient, classify emails as spam or not, or identify objects in images.

We can classify quadrilaterals by observing their properties — such as parallel sides or right angles. From these observations, we build a classification tree that organizes shapes into groups based on their features.

Classification trees are among the oldest machine learning methods, yet we still commonly use them as illustrative examples in teaching.

If you are running this lesson in an unplugged setting, it works best if students have already encountered similar activities, such as What are the gnomes doing or Animal tree. When working on a computer, however, you may prefer to skip the manual construction of the tree and focus on exploring the results instead.

Ordering a set of 15 cards with images of cars — from very old models to the newest — can be a fun and surprisingly challenging task for students (and even for adults!). Students receive the cards and try to arrange them in chronological order. After checking the correct sequence, the class reflects on how to score their arrangements (scoring is the main topic of the lesson!), compares group performance, and examines how a computer tackles the same task.

Despite common stereotypes, cars proved just as exciting for girls as they were for boys.

Other images — like inventions, books, or items sortable by a property (e.g., towns by distance) — can be used, making the activity adaptable across subjects. Alternatively, items can be ordered based on a chosen property (for example, towns in a certain country by distance from the school).

Given a set of items — countries, images, books, movies, or songs — how can we group them?

In this simple kinesthetic activity, we arrange students based on their imagined grades in math and PE, gradually forming larger and larger groups to reveal clusters of similar students. This introduces a popular hierarchical clustering algorithm, which we then apply in the following activities with real data to make the learning more meaningful and goal-oriented.

Dendrograms are visual diagrams that show how items are grouped hierarchically into clusters. In this lesson, students learn how to read and interpret these diagrams. It is recommended to first explore the clustering process itself (Clusters in the Classroom), so that the concepts behind the dendrograms become clear and meaningful.

Learning about climatographs and climate zones becomes more engaging when students explore the data themselves. In this activity, students investigate climate patterns across different continents, while the teacher guides the process and encourages them to consider the distinct characteristics of each climate zone.

Behind the scenes, students are also encountering the idea of clustering. The activity can conclude by demonstrating how a computer can analyze the same climate data and identify the climate zones on its own, showing that the patterns the students discovered can also be recognized algorithmically.

Students play Tic-Tac-Toe against an “intelligent paper” that follows a set of fixed rules (for example, “if the opponent does this, do that”) to ensure the white player can at least tie and sometimes win. This sparks a discussion about intelligence: students usually conclude that the paper itself isn’t intelligent—the person who wrote the rules is. They then agree that a computer would be considered intelligent if it could learn the rules by itself, which naturally leads into the Candy Computer activity, where the system discovers strategies through learning from mistakes.

The activity was developed by Paul Curzon at Queen Mary University of London and is published as part of CS Unplugged.

As a follow-up to Intelligent Paper, students get to build a computer out of candies that learns to play a simple game with six pawns (Hexapawn). At first, the “computer” is completely clueless — making random moves and sometimes missing an easy win — but as it learns from its mistakes, it gradually becomes unbeatable.

This hands-on activity takes some work to set up, but it’s always a hit with students. While they play, it naturally sparks discussions about how learning happens. Where is intelligence? Is each candy — or each cup of candies — becoming more intelligent with each game? These questions can lead to even bigger debates, like: can a machine truly be intelligent?

The activity comes from CS4FN: The Sweet Learning Computer, part of the CS4FN project started by Paul Curzon and Peter McOwan at Queen Mary University of London.

By looking at enough photographs of mosques, churches, and Buddhist and Hindu temples, we learn to tell them apart. Can computers learn this skill, too?

Distinguishing between rennaisance and impressionism is easy. How about adding surrealism and cubism: can you tell them apart? What about students?

The activity involves a competition between would-be art historians in the classroom, followed by computer attempting the same task: shown a set of examples of each period and movement, can it learn to distinguish between them?

Édouard Manet (1832) and Claude Monet (1840) were French painters and contemporaries during the Modernism period. A layperson can easily confuse them. In this activity, students - and later the computer - will try their hand at distinguishing between their works.

While lessons like Gnomes and Animal Tree focus on building the tree, this lesson is aimed at older students and emphasizes data analysis. Working individually or in pairs, students use Orange to independently explore classification trees and examine the characteristics of different animal groups through informative visualizations.

Remember the brilliant visualizations in Hans Rosling’s TED Talk? We can use similar data and visualizations— perhaps less flashy, but conceptually deeper — to explore, update, or even challenge our understanding of different countries around the world.

Are your students complaining about “impossible” timetables — too many challenging lessons in one day? Overloaded Fridays?

This activity shows them that creating a good schedule is no easy task. Students tackle a scheduling problem where only one classroom is available, but multiple groups request different time slots. How many of these reservations can be accommodated at most?

How can we use data on players’ physical attributes and shooting accuracy in the NBA to determine their playing positions? Can we guess which current player doesn’t fit the typical profile?

The activity is intended for upper-level students who are comfortable with linear combinations, roots, exponential functions, and logarithms. Additionally, they are interested in “guessing” physical laws from data without prior knowledge of physics.