Most people meet AI as a headline or a chatbot, which makes it feel like magic. Good teaching replaces the magic with something better: a system you can poke, break and fix.
Let them see it work first
Start with a running model, such as an object detector on a webcam. Learners ask better questions after they have seen it succeed and fail than after a definition.
Make the data theirs
Have learners collect their own small dataset, ten photos of two objects. Training on their own data shows why more data and cleaner data matter, in a way no slide can.
Teach failure as a feature
Show where the model is wrong and ask why: lighting, angle, a confusing background. Understanding limits makes learners safer users of AI.
Measure, do not guess
Introduce a simple scorecard: right, wrong, missed, false alarm. Numbers turn opinions about a model into evidence.
End with a build
Put the model on something that acts: a motor, a light, an alert. When a learner sees their model move a real object, AI stops being intimidating and becomes a tool.