Unit 4 — AI & Canva: Creating with and about AI. The aim is confident AND critical use of AI — not passive acceptance of whatever it produces.
Lesson 11 — AI as a Creative Partner: Human vs Machine
What you'll achieve: an AI-assisted creative piece — an illustrated story, poster or campaign — with a written human-vs-AI comparison.
CopilotCanva
Prompt
The instructions or question typed into an AI tool.
Generative AI
AI that creates new content, such as text or images, rather than just analysing existing content.
Critical evaluation
Carefully judging the quality, accuracy or fairness of something rather than accepting it automatically.
I Do
Generative AI tools like Copilot and Canva's Magic Write don't "think" like people — they predict likely words or pixels from patterns in huge amounts of existing text and images. Watch your teacher type a prompt into Copilot, read the result, then improve the prompt and generate again. AI does not "know" facts — it produces plausible-sounding output that must always be checked.
We Do
As a class, agree on one prompt topic; generate text in Copilot and a version in Canva's Magic Write.
Compare both AI outputs, then compare against a human-written example across Accuracy, Creativity, and "Would you actually use this?"
You Do
Choose one output type: a short illustrated story, a poster, or a short social media campaign.
Use Copilot to generate written content — try at least 2 prompts and pick your favourite.
Use Canva (Magic Write and/or text-to-image) to generate a matching visual.
Combine text and image into one finished Canva design.
Write 5–6 sentences: what did the AI do well, what did it get wrong or make bland, and would a human have done it differently?
Lesson 12 — Can We Trust AI? Decision-Making and an Infographic
What you'll achieve: a Canva infographic, "5 things AI can and can't do well".
Canva
Decision-making AI
AI used to make or support real decisions, for example medical diagnosis or self-driving cars.
Training data
The information an AI system learns from before it is used.
Bias (in AI)
When training data causes an AI to make unfair or inaccurate decisions for some groups.
I Do
"Would you trust a robot doctor to diagnose you? Why or why not?" Three short case studies — medical diagnosis, self-driving cars, content moderation — introduce what could go right and wrong with decision-making AI. AI mistakes don't mean AI is "broken" — they reflect the limits and biases of its training data and design.
We Do
As a class, agree on one "can do well" point and one "can't do well" point, and build both into a shared Canva infographic — choosing a template, adding a heading, and placing text and an icon for each.
You Do
Open Canva and choose an infographic template.
Add a clear title, for example "AI: What It Can and Can't Do".
Write 5 short points — at least 2 things AI does well and 2 it struggles with — each with an icon.
Use consistent colours and fonts (try Canva's Styles panel).
Reference at least one of today's case studies in one of your points.
Lesson 13 — Teach the Machine: Training Your Own AI Model
What you'll achieve: a working gesture-recognition model, plus a written reflection added to your HTML portfolio.
Teachable MachineHTML
Training data
The examples an AI model learns from before it can make predictions.
Model
The "trained brain" that Teachable Machine produces after learning from your examples.
Webcam input
Using a camera in real time as the source of data for a model.
I Do
Watch a model trained on just 2–3 examples per gesture fail on a new angle — training data quality and quantity directly affects performance. Watch your teacher create an Image Project, add classes, record samples from different angles, click Train Model, then test live. More training only makes a model better at its own specific classes, not "smarter" generally.
We Do
Together, build a shared model: 2 classes (e.g. "Hand Up" and "Hand Down"), recording samples from a few different angles and distances, then train and test live, discussing any misclassifications.
You Do
Go to teachablemachine.withgoogle.com and start a new Image Project.
Create 3 classes for 3 different hand gestures.
Record 25–30 samples per class, varying angle, distance and background.
Train and test your model live, including from an angle you didn't train on.
Add a new "My AI Reflection" section (<h2> and <p>, 4–5 sentences) to your HTML portfolio describing what you built, how training data affected accuracy, and one limitation.
BLC