Class 10 · Unit project
Unit 2 project: Teach a machine, explain its mind
Build a model that decides something real. Then explain how it decides to someone who has never heard of AI.
The task
You have built rules by hand, trained a model on leaves, and watched a machine write its own rules for apples. Now your group builds a model of its own choosing — and then does the harder part: explains it to someone who has never heard of AI.
A model nobody can explain is a model nobody can argue with. By the end of this period, anyone who reads your work should be able to say what your model looks at, how it decides, and when it gets it wrong.
Work in your project groups of 3–4. One period. Everything goes in model-explainer.odt.
Choose a model (3 min)
Pick one:
- A. Will the pass road be open?
datasets/road-days.csvhas 60 winter mornings: overnight snow, the temperature at 6 a.m., rainfall, and whether the road opened. Build a decision tree in Orange (or by hand in Calc). Test it onroad-days-test.csv— 10 mornings it has never seen. - B. An image classifier in Teachable Machine (online), from photos your group takes of something real: three kinds of leaves, three kinds of waste, ripe and unripe apples. At least 15 photos per class, and 5 more per class kept back for testing. No people in any photo.
Build and test (17 min)
- Build the model. For A, limit the tree depth to 2 or 3. For B, train it.
- Test it only on the examples you kept back. Count how many it gets right, out of how many.
- Find one case it gets wrong, and work out why.
Explain it (15 min)
Fill the five parts of the explainer, two or three lines each:
1. What it looks at. The features (A) or the classes and photos (B). What does it not look at, that a person would?
2. How it decides. For A, write the tree as sentences: "If … then …". For B, say what you think it notices in the photos — and how you could check.
3. How well it works. "On … examples it had never seen, it was right … times."
4. One mistake, explained. Which case, what it said, what was true, and why you think it went wrong.
5. What kind of model is it? Rule-based or learning-based; supervised, unsupervised or reinforcement; classification, regression, clustering or association. One line each, with a reason.
The Unit 1 line. Finish this sentence: "If people relied on this model, the person it could let down is … because …"
Your own line (5 min, alone)
Each person writes three sentences explaining your model to a Class 7 student — no words they wouldn't know. Sign with your roll number. This part is marked for you alone.
How it is marked
Each row is marked 0–3. Rows 1–5 are shared by the group; row 6 is yours alone. Marks are returned privately and never compared with anyone else's.
Total: 18. A model that is right less often but honestly tested and explained scores higher than an impressive one nobody tested properly.
1. It works
- 0No working model
- 1A model, not tested
- 2Tested, but on its training examples
- 3Tested only on examples it never saw, with the count
2. How it decides
- 0Missing
- 1"It uses AI"
- 2The rules or the features, stated
- 3Stated as plain sentences anyone could check
3. One mistake
- 0None found
- 1A mistake, no reason
- 2A mistake and a reason
- 3A mistake, a reason, and what would fix it
4. What kind of model
- 0Missing
- 1Some terms, no reasons
- 2All three named correctly
- 3All three named, each with a reason
5. The Unit 1 line
- 0Missing
- 1"Nobody"
- 2A real person it could let down
- 3A real person, how, and why that matters
6. For a Class 7 student
Just you- 0Missing
- 1Uses words they wouldn't know
- 2Clear, mostly
- 3Clear, correct, and something they'd remember