Skip to content

AI 101

Machine Learning 101: The "Detective"

What is Machine Learning (ML)?

If Generative AI is the "Creative Artist," Machine Learning is the "Smart Detective."

It doesn't write poems or draw pictures. Instead, it looks at massive amounts of evidence (data) to find patterns and answer questions. It uses these patterns to predict what will happen next or identify what something is.

The Difference:

  • Generative AI: "Write me a new story about a cat."
  • Machine Learning: "Here is a photo. Is this a cat or a dog?"

How Does It Work? (The "School" Analogy)

Computers don't "know" anything until we teach them. This process is called Training.

Explore

  1. Training Data (The Textbooks) Imagine you want to teach a computer to spot a "Healthy Leaf" vs. a "Diseased Leaf." You show it 1,000 photos of healthy leaves and 1,000 photos of sick leaves. This is its "study material."

  2. The Model (The Student) The computer studies the photos. It notices patterns humans might miss—like "sick leaves always have tiny yellow spots with brown edges." It builds a rulebook in its brain.

  3. Testing (The Exam) You give the computer a new photo it has never seen before. If it correctly guesses "Diseased," it passes! If it fails, you give it more data to study.

The Two Main Jobs of ML

1. Classification (The Sorter)

Putting things into buckets.

  • Spam Filters: Looks at an email and decides: [Inbox] or [Junk]?
  • Medical Scans: Looks at an X-ray and decides: [Healthy Bone] or [Fracture]?
  • Computer Vision: Looks at a video feed and decides: [Person], [Car], or [Tree]?

2. Prediction (The Fortune Teller)

Guessing a number or a future event based on the past.

  • Netflix Recommendations: "You liked Batman, so you will probably like Spider-Man."
  • Weather: "It rained on the last 5 days with this air pressure, so it will likely rain tomorrow."
  • Google Maps: "Traffic is usually bad here at 8:00 AM, so your trip will take 15 minutes longer."

Common Traps: Why ML Can Be "Unfair"

Machine Learning predicts the future by studying the past. But what if the past was unfair?

  • The Resume Problem:
    Imagine a company uses AI to hire new bosses. It trains the AI on the resumes of the last 10 years of successful bosses.
  • The Trap: If most past bosses were men, the AI might learn the pattern: "Men make better bosses."
  • The Result: The AI starts rejecting female candidates, even if they are qualified.

Student Tip: When building your challenge project, ask yourself: "Does my data have bad habits from the past that I don't want to repeat?"

Try It Yourself (No Coding Needed)

  • Teachable Machine (Google): Train a computer to recognize your face or voice in 2 minutes.
  • Link: teachablemachine.withgoogle.com
  • Akinator: A classic example of a "Decision Tree." It asks you questions to narrow down possibilities until it guesses your character.
  • Link: akinator.com

The "Time Capsule" Problem: Why AI Expires

One of the biggest weaknesses of Machine Learning is that it is frozen in time.

When you finish training a model, it takes a snapshot of the world as it is right now. It knows everything that happened before today, but nothing about tomorrow. If the world changes, the AI doesn't automatically update—it becomes a dinosaur.

1. The "Knowledge Cutoff"

Imagine you studied a history textbook written in 2019. If someone asked you, "Who won the 2022 World Cup?", you wouldn't know.

AI is the same.

  • Example: A chatbot trained on internet text from 2021 might not know what "Generative AI" is, or it might not understand new slang words like "Rizz" or "no cap," because those words weren't popular in its training data.

2. Model Drift (When the World Changes)

"Drift" happens when the data you trained on no longer matches the real world.

  • The Travel Example: Imagine a travel AI trained on flight data from 2019. It learns that "Summer is the busiest time for travel."
  • The Shift: Then, the 2020 Pandemic hits. Suddenly, nobody is flying in the summer.
  • The Failure: Because the model is "frozen" in 2019 habits, it keeps predicting massive crowds that don't exist. It fails because the world changed, but the model didn't.

3. The Solution: Retraining (Going Back to School)

To fix this, engineers must periodically Retrain the model.

This means feeding the AI fresh data (e.g., flight patterns from 2024) so it learns the new rules of the world.

Student Tip: For your challenge, check the date of your data.

  • Bad: Using crime statistics from 2010 to predict safety in 2026 (Neighbourhoods change!).
  • Good: Using the most recent council data available.
graph TD
    %% Define the outer layers as subgraphs (containers)
    subgraph AI ["Artificial Intelligence"]
        direction TB

        subgraph ML ["Machine Learning"]
            direction TB

            subgraph DL ["Deep Learning"]
                direction TB

                %% The innermost circle
                GenAI(("Generative AI"))
            end
        end
    end

    %% Styling to match the blue gradient from the image

    %% Outer Layer (Lightest Blue)
    style AI fill:#9ecae1,stroke:#ffffff,stroke-width:2px,color:#ffffff,rx:10,ry:10

    %% Second Layer (Medium-Light Blue)
    style ML fill:#6baed6,stroke:#ffffff,stroke-width:2px,color:#ffffff,rx:10,ry:10

    %% Third Layer (Medium-Dark Blue)
    style DL fill:#3182bd,stroke:#ffffff,stroke-width:2px,color:#ffffff,rx:10,ry:10

    %% Center Node (Darkest Blue)
    style GenAI fill:#08519c,stroke:#ffffff,stroke-width:2px,color:#ffffff

Generative AI 101: A Student’s Guide

What is Generative AI?

Traditional AI (Machine Learning) is like a detective: it looks at data and finds patterns (e.g., "This photo is a cat").

Generative AI is like an artist or writer: it takes what it has learned from billions of examples and creates something new.

It doesn't just "copy and paste" from the internet. It predicts the next best word or pixel to create original text, images, or code.

The Golden Rule of Prompting

Generative AI is only as good as the instructions you give it. These instructions are called "Prompts."

If you ask a vague question, you get a vague answer.

❌ Bad Prompt: "Write a blog about pollution."

Result: Generic, boring text that sounds like a textbook.

To get great results, use this formula: Context + Task + Constraints.

Component What it means Example
Context Who is the AI? Who is the audience? "You are a friendly expert on recycling. I am a Year 8 student."
Task What exactly do you want it to do? "Write a 30-second script for a TikTok video about why plastic bottles are bad."
Constraints What are the rules? (Length, tone, format) "Use funny, energetic language. Do not use complex words. Include 3 emojis."

✅ Good Prompt: "You are a friendly recycling expert. Write a 30-second TikTok script for Year 8 students about why plastic bottles are bad. Use funny, energetic language, avoid complex words, and include 3 emojis."

3 Ways to Use GenAI in Your Project

Don't let the AI do the work FOR you. Use it to help you work BETTER.

  1. The "Brainstorm Buddy" Stuck on ideas? Ask: "I need to solve the problem of litter in my school playground. Give me 10 creative ideas for an app or tool that could help."
  2. The "Editor" Wrote a pitch but it sounds boring? Paste your text and ask: "Read this pitch for my school project. Tell me 3 ways I can make it sound more exciting and persuasive."
  3. The "Coder" Need help with Python? Ask: "I am writing a Python script to detect blue objects in a photo. Explain which library I should use and show me a simple example code snippet."

⚠️ The Danger Zone: Risks & Ethics

Before you use AI, you must understand the risks.

  • 1. Hallucinations (AI Lies)
  • Generative AI is a "people pleaser." It wants to give you an answer so badly that it sometimes makes things up.
  • Rule: Never trust a fact from AI without checking it on Google.
  • 2. Bias (Unfair Stereotypes)
  • AI learns from the internet, and the internet has stereotypes. If you ask for an image of a "Doctor," it might only show men. If you ask for a "Criminal," it might unfairly show certain groups of people.
  • Rule: Always look at your results and ask: "Is this fair to everyone?"
  • 3. Plagiarism (Cheating)
  • If you copy-paste an entire essay from ChatGPT, you learn nothing, and teachers can often tell.
  • Rule: Use AI to generate ideas, not final products. Always declare your use of AI in your Technical Document.

Quick Tool Guide

Check with your teacher which tools are allowed on your school network.

  • Text: ChatGPT, Google Gemini, Microsoft Copilot.
  • Images: Adobe Firefly (often free for schools), Canva Magic Media, Bing Image Creator.
  • Code: Replit Ghostwriter, GitHub Copilot.

Useful Demos and Tools

Machine Learning 101

Artificial Intelligence 101

  • AI for Oceans (Code.org): An interactive game that teaches how AI classifies data and how "bias" happens in data.
    https://code.org/ai
  • Elements of AI: A free online course that explains what AI is without complex math.
    https://elementsofai.com

Generative AI 101