Artificial Intelligence · Grade 7 · Chapter 5
How AI Models Work
Inside the Neural Network: How machine learning turns patterns into predictions.
Chapter Goals
What We Will Master Today
- Traditional vs ML: Contrast hardcoded rules with pattern recognition.
- Feature Extraction: Convert real-world data into numerical inputs.
- Neural Network Structure: Understand Input, Hidden, and Output layers.
- Training & Weights: Learn how models tune parameters to reduce error.
The Magic Photo Classifier
How Does Your Phone Know It's a Cat?
No programmer wrote 10,000 if statements describing every cat ear, whisker, or tail angle. So how does AI instantly identify cats in photos?
Discuss before revealing: Could you write if/else rules that work for EVERY cat in the world?
Core Concept 1
Traditional Programming vs Machine Learning
Traditional Code
Inputs + Hardcoded Rules → Output
Human programmer writes explicit rules for every edge case.
Machine Learning (AI)
Inputs + Target Answers → Learned Rules (Model)
AI discovers patterns automatically from training data!
Core Concept 2
Data Feature Extraction
Computers can't "see" colors or "hear" sounds — they only understand numbers. Data must be converted into numerical features.
Image Pixels
A photo is converted into a grid of numbers representing Red, Green, and Blue brightness (0 to 255).
Audio Waveforms
Microphone sound is broken down into frequency numbers (pitch and volume over time).
Visualizing Data
How AI Sees a 3x3 Pixel Image
Imagine a tiny 3x3 image grid converted into numbers for the AI:
[255, 0, 0]
[0, 255, 0]
[0, 0, 255]
[128, 128, 0]
[255, 255, 255]
[0, 0, 0]
Core Concept 3
What is an Artificial Neural Network?
An AI model inspired by biological brains, made of interconnected mathematical nodes (neurons) grouped in layers.
2
Hidden Layers (Pattern Finding)3
Output Layer (Prediction)
Architecture Diagram
Structure of a 3-Layer Neural Net
Core Concept 4
What are Weights and Biases?
Connections between neurons have Weights (importance strength) and Biases (threshold modifiers).
High Weight
If feature = "Pointy Ears", weight is high for "Cat" prediction.
Low / Negative Weight
If feature = "Pointy Ears", weight is low for "Fish" prediction.
Physical Analogy
How AI Learns: Shooting a Basketball
1
First Shot: Misses left (High Error)2
Calculate Error: Adjust wrist angle & force3
Repeat 10,000 times until swish!
AI training works the exact same way: calculate prediction error and tweak weights backward!
Training Cycle
The Machine Learning Loop
Forward Pass
Input data moves forward through layers to make a guess.
Loss Calculation
Compare guess with true answer to calculate error score.
Backpropagation
Adjust connection weights backward to improve next guess!
Critical Thinking
Garbage In = Garbage Out!
An AI is only as good as the data used to train it.
Scenario: You train an AI dog classifier using 5,000 photos of dogs in snow. What happens when you test it with a dog on a green lawn?
Discussion: The AI might learn that "Snow = Dog" instead of learning dog features!
Smartboard Voting
Is This AI (Machine Learning) or Traditional Code?
1. ATM machine counting cash bills
2. Spam filter learning suspicious email text
3. Spotify generating personalized weekly playlists
4. Microwave timer counting down 60 seconds
AI Domains
Computer Vision vs Natural Language Processing
Computer Vision (CV)
Processes visual pixels to recognize objects, medical X-rays, and self-driving roads.
Natural Language Processing (NLP)
Processes text tokens to translate languages, answer questions, and summarize documents.
Spot the Flaw
What is "Overfitting"?
A student memorizes the exact answers to last year's math test, but doesn't understand the concepts. When given new test questions, they fail!
In AI, Overfitting happens when a model memorizes training photos instead of learning general patterns.
Quick Check · Quiz 1
How does Machine Learning differ from Traditional Programming?
AML uses faster microchips
BML learns patterns from data instead of hardcoded rules
CTraditional code doesn't use math
DML only runs on smartphones
Click to reveal answer
Quick Check · Quiz 2
What is the primary role of "Weights" in a neural network?
ATo measure the file size of the AI
BTo adjust connection strength between neurons
CTo increase computer battery life
DTo turn the monitor on and off
Click to reveal answer
Quick Check · Quiz 3
Which neural network layer is responsible for extracting hidden features?
AInput Layer
BHidden Layer
COutput Layer
DPower Supply
Click to reveal answer
Quick Check · Quiz 4
What is the consequence of training an AI model on incomplete or biased data?
AThe AI becomes conscious
BThe computer melts
CThe AI makes biased or inaccurate predictions
DNothing changes
Click to reveal answer
Guided Practice
Design an AI Weather Predictor Model
Input Layer (3 Features): Temperature, Humidity %, Wind Speed
Hidden Layer: Combines features to calculate storm likelihood
Output Layer: "Rain" (85% confidence) vs "Clear" (15% confidence)
Real-World AI
Where Neural Networks Work Today
- Healthcare: Detecting lung tumors in X-rays faster than human eyes.
- Autonomous Vehicles: Identifying pedestrians, lanes, and traffic signs in real-time.
- Voice Assistants: Converting speech waveforms into text queries.
- Fraud Detection: Flagging suspicious credit card swipes instantly.
Myth vs Fact
AI Myth Buster
Myth
"AI is conscious and thinks like a human brain."
Fact
"AI is high-speed mathematical probability matching data patterns."
Safety Rule
Human-in-the-Loop (HITL)
Critical decisions (medical diagnoses, legal judgements) must ALWAYS be verified by human experts, not left solely to AI predictions.
Summary & Takeaways
What We Learned in Chapter 5
- Machine learning discovers rules automatically from training data examples.
- Neural networks process features through Input, Hidden, and Output layers.
- Weights tune connection strength during training to minimize error.
- Biased training data produces unfair or inaccurate AI predictions.
Pipeline Recap
The AI Pipeline in 4 Steps
Exit Ticket
Neural Net Reflection
In 2 sentences, explain:
What is the job of the Hidden Layer inside a neural network?
Next Chapter: Python Conditionals (if/elif/else)!