Home Blog Teaching AI Without a Screen: Five Graph Paper Activities for Class 9 and 10
Uncategorized November 19, 2026 7 min read

Teaching AI Without a Screen: Five Graph Paper Activities for Class 9 and 10

Ask a Class 10 student what "an image is a matrix of numbers" means and you will usually get the sentence repeated back word for word. Ask them to show you, and things go quiet. That gap between knowing the line and understanding the idea is where a lot of marks disappear in the AI […]

Lalit Kumar Published
Featured image
Reading time 7 min
Published November 19, 2026
Uncategorized Nov 2026
Back to Blog

Ask a Class 10 student what "an image is a matrix of numbers" means and you will usually get the sentence repeated back word for word. Ask them to show you, and things go quiet. That gap between knowing the line and understanding the idea is where a lot of marks disappear in the AI paper, especially in the questions that ask students to explain or apply rather than define.

One of the most useful fixes we have seen costs almost nothing. It is a pencil, a ruler and a few sheets of squared paper. Before students open a no-code tool or write a line of Python, getting them to work an idea out by hand on a grid makes the abstract parts of the syllabus feel concrete. Here are five activities that fit neatly into a single period each, along with the chapter each one supports.

1. Draw a picture, then turn it into numbers

This is the easiest place to start, and it lands the computer vision idea better than any diagram in the textbook.

Give each student a grid of 10 by 10 squares. Ask them to draw a simple shape, a smiley face or the first letter of their name, by shading whole squares only. No half squares, no curves. Then ask them to write a 1 in every shaded square and a 0 in every empty one.

What they now have is a black-and-white image the way a computer stores it. Swap sheets with a partner and ask them to redraw the picture using only the numbers. Most students get it right first time, and the point makes itself: the computer never "sees" a face, only a grid of values.

For a stronger class, introduce shades. Instead of 0 and 1, use 0 for white, 5 for light grey and 9 for black. Now they are close to how greyscale works, and it is a short step to explaining why real images use values from 0 to 255 and why a colour image needs three numbers per pixel instead of one.

2. Plot real data and draw a line of best fit

The data science unit asks students to read and interpret data, and regression appears when the syllabus talks about models that predict a number. Both make much more sense once students have plotted their own data.

Pick something you can measure in the classroom in ten minutes. Height and arm span works well, because the two are closely related and nobody minds sharing them. Each student measures a partner, and the class pools the results on the board.

On graph paper, students draw the axes, choose a sensible scale and plot every point. Then, with a clear ruler, they draw the straight line that they think passes closest to most of the points. That line is the model. Ask them to use it to predict the arm span of a student whose height is not on the chart, then check against the real measurement.

The discussion afterwards is the valuable part. Why don't all the points sit on the line? What happens to the prediction for someone much taller than anyone in the class? Students start asking about accuracy and about data that falls outside the training set without being told those terms exist.

3. Classify a new point by looking at its neighbours

Classification is one of the trickier ideas in the modelling chapter, because the word sounds technical. On paper, it is almost a game.

Prepare a sheet with about 20 points already plotted, each marked with either a circle or a cross. You can frame it as anything: fruits plotted by weight and sweetness, or animals plotted by size and speed. Then add one new, unmarked point and ask the class to decide whether it should be a circle or a cross.

Most students will look at the points closest to it. Ask them to count the three nearest. If two are crosses and one is a circle, call it a cross. That is the nearest neighbour approach, and they have just done it without any code. Move the new point somewhere awkward, right on the border between the two groups, and let them argue. Then ask what changes if they count five neighbours instead of three.

4. Find groups nobody labelled

Clustering is classification's quieter cousin, and the difference between the two is a favourite short-answer question. Graph paper makes the difference visible.

Hand out a sheet of plotted points with no labels at all. Ask students to circle the groups they can see. Most will spot two or three clusters straight away, and they will not all agree on where the boundaries are.

Now put the sheet from activity three next to this one. In the first, the labels were given and the job was to predict a label for something new. In the second, nobody provided labels and the job was to find structure. That side-by-side comparison does more to explain supervised and unsupervised learning than a page of notes.

5. Build a confusion matrix from a pretend model

The evaluation chapter is where students tend to mix up their terms, especially precision and recall. A 2 by 2 grid drawn on squared paper helps a great deal.

Write a list of 20 made-up predictions on the board, for example a model that guesses whether a photo contains a dog, alongside the correct answer for each. Students draw a large 2 by 2 table, label the rows "predicted yes" and "predicted no" and the columns "actually yes" and "actually no", then put a tally mark in the right box for each of the 20 cases.

When they are done, they have a confusion matrix they built themselves. Calculating accuracy from it is simple. Precision and recall become easier to remember once students can point at exactly which boxes each formula uses, rather than memorising letters.

Choosing the right grid

The size of the squares matters more than you might think. For the pixel activity, larger squares of around 1 cm are easier to shade neatly and count, especially for younger classes. For scatter plots and the clustering exercise, a finer 5 mm grid gives students room for a proper scale and more data points. Many school notebooks only come in one ruling, so it helps to have loose sheets ready.

If the stationery cupboard has run out, you can print free graph paper in the exact grid size each activity needs, which also means every student starts with an identical sheet. That sounds like a small thing, but it makes comparing work and marking much quicker.

Why bother when there are tools for all of this?

There are excellent browser tools for every one of these ideas, and students should use them. The point of the paper version is not to replace them. It is to make sure that when a student clicks "train" in a no-code platform, they already have a picture in their head of what is happening underneath.

There is also a practical benefit for exams. The AI paper is written, not typed. Students who have physically plotted a line of best fit or filled in a confusion matrix tend to explain those ideas in clearer, more confident language, and examiners notice. They also find it easier to draw a quick diagram in an answer, which often earns the mark that a vague sentence would not.

Finally, these activities work in classrooms where devices are limited or the internet is unreliable, which is still true in many schools. A stack of squared paper does not need charging.

Start with one

You do not need to run all five. Pick the activity that matches the chapter you are teaching this week and try it for one period. The pixel grid is a good first choice, because it is quick, everyone can do it and it makes a dry definition suddenly obvious. Once students have seen an idea on paper, the textbook version tends to make a lot more sense.