Introduction to Artificial Intelligence – Training a Simple Model
Artificial Intelligence can sound mysterious, but at its heart it’s built on ideas students already know: patterns, data, graphs, and feedback. In this introduction, we strip AI back to first principles by training a very simple model—no coding background required.
What do we mean by “training” an AI model?
Training is simply the process of showing a computer examples, letting it make predictions, and then correcting it when it gets things wrong. Over time, those corrections improve its accuracy.
Think of it like teaching a student to estimate the height of a tree from its shadow:
You give several examples (shadow length → tree height)
The student guesses
You say how far off they were
They adjust their method next time
That loop—predict → check → adjust—is the core of machine learning.
A simple example: predicting exam scores
Imagine we want to predict a student’s exam score based on hours of revision.
| Hours of revision | Exam score (%) |
|---|---|
| 1 | 40 |
| 2 | 50 |
| 3 | 60 |
| 4 | 68 |
| 5 | 75 |
We might start with a simple rule:
Score = (hours × 10) + 30
This is our model. It won’t be perfect, but it gives us a starting point.
How the model learns
Make a prediction
For 3 hours:
Predicted score = (3 × 10) + 30 = 60%Compare with the real result
Actual score = 60%
Error = 0 (perfect!)Adjust if needed
If predictions are consistently too high or too low, we tweak the numbers.
After many examples, the model settles on values that minimise the overall error. That process—reducing error step by step—is called training.
Key ideas students should remember
Data: The examples we train the model on
Model: A rule or equation that makes predictions
Prediction: The model’s output
Error (loss): How wrong the prediction is
Training: Repeating predictions and corrections to reduce error
At GCSE and A-Level, this links directly to:
Graphs and lines of best fit
Averages and spread
Iterative improvement
Cause-and-effect reasoning
AI is not magic—it’s applied maths and logic at scale.
Why this matters in school science and maths
Understanding how a simple model learns helps students:
Demystify headlines about “AI taking over”
See real-world applications of algebra and graphs
Develop critical thinking about data quality and bias
Build confidence with modern technology
At Hemel Private Tuition, we often teach AI ideas using familiar experiments and datasets, so students focus on understanding—not buzzwords.
The Next Step: Training a Model to Classify Images (Cats vs Dogs)
Once students understand training a simple numerical model, the natural progression is to ask:
Can a computer learn from pictures instead of numbers?
Yes—and this is where image classification comes in.
What does “classifying images” mean?
In our example, there are just two classes:
🐱 Cat
🐶 Dog
The model’s job is simple:
Given a new image, decide whether it is more likely to be a cat or a dog.
Step 1: Images must become numbers
Computers don’t “see” images the way we do.
An image is actually a grid of pixels, and each pixel has numbers attached to it.
For a colour image:
Each pixel has Red, Green, and Blue (RGB) values
Each value is usually between 0 and 255
So an image becomes a very large table of numbers.
This links nicely to:
Matrices in maths
Grids and coordinates
Data representation in computer science
Step 2: The model looks for patterns, not animals
The model is not told what a cat or dog is.
Instead, during training it starts to notice patterns such as:
Fur texture
Edges and shapes
Ear positions
Contrast between background and subject
Early layers might detect:
Straight and curved lines
Light vs dark regions
Later layers combine these into more complex features.
At school level, you can explain this as:
The computer learns which patterns usually appear in cat photos and which appear in dog photos.
Step 3: Training works just like before (predict → check → adjust)
The training loop is exactly the same idea as before:
Show the model an image
Model predicts: “cat” or “dog”
Check the label (we already know the correct answer)
Calculate the error
Adjust the model slightly
This is repeated thousands of times using many images.
The only difference from the simple maths model:
The model has many more adjustable values
The maths happens automatically behind the scenes
Step 4: Testing with new images
Once trained, the model is tested using images it has never seen before.
This is crucial:
A model that memorises training images is useless
We want generalisation, not memory
Example output:
Cat: 92%
Dog: 8%
The model chooses the highest probability.
Key concepts students should remember
| Term | Meaning |
|---|---|
| Training data | Images the model learns from |
| Labels | The correct answers (cat / dog) |
| Features | Patterns the model detects |
| Model | The system making predictions |
| Accuracy | % of correct classifications |
| Overfitting | When a model memorises instead of learning |
Why cats vs dogs is such a good teaching example
Clear, familiar categories
Easy to understand success and failure
Shows limits of AI (misclassified images are often interesting)
Links maths, computing, biology (vision), and ethics
Students quickly realise:
AI doesn’t understand animals — it recognises patterns in data.
Classroom-friendly ways to explore this (no coding required)
Sort printed images by hand → discuss “features”
Use an online image classifier demo
Compare correct vs incorrect classifications
Change the dataset and see accuracy change
Discuss bias (e.g. only fluffy cats, only small dogs)
This fits beautifully into:
GCSE Computer Science
A-Level Computer Science
STEM enrichment sessions
Cross-curricular maths & science lessons
The big idea
Training an image classifier is just a scaled-up version of what students already know:
Patterns → predictions → feedback → improvement
Once that clicks, AI stops being intimidating—and starts being understandable.
No comments:
Post a Comment