Final Assessment

Teaching AI for Classes 9–10 — Final Assessment

Application-heavy questions drawn from across all twelve modules, with a safety, ethics and privacy subset. Pass mark 75%. Attempt again after focused revision if needed.

1. A student says a washing-machine timer is "AI because it decides when to stop." What is the accurate correction?

It follows a fixed rule with no learning — that is automation, not AI.Correct — anything automatic is AI.Yes, because it uses electricity.Only if it is a front-loading machine.

2. Which task belongs to the computer-vision domain rather than NLP or data AI?

Detecting objects in an image.Analysing the sentiment of a review.Predicting demand from sales numbers.Translating a sentence between languages.

3. Why is problem scoping done before data collection in the AI Project Cycle?

A clearly defined problem determines what data is actually needed.Data collection is never necessary.Scoping is only paperwork with no effect.The order does not matter at all.

4. A teacher wants AI help with report comments and has a marksheet with names. What must they do first? (safety/privacy)

Remove all identifiers (use Student A, B, C) or use synthetic data.Paste the full marksheet with names to save time.Add more personal details for accuracy.Upload student photos as well.

5. Why can a large dataset still produce a biased model?

Because a large but unrepresentative sample is still biased.Large datasets are always perfectly fair.Model bias has nothing to do with data.Only small datasets can be biased.

6. When is the median a better summary than the mean?

When outliers or skew would pull the mean away from the typical value.Never — the mean is always better.Only for categorical data.The median cannot be calculated.

7. How can a chart mislead even when its underlying data is accurate?

A truncated axis or wrong chart type distorts an honest number.Charts can never mislead.Only the data values matter, never the axes.Misleading requires changing the data.

8. How are AI, machine learning and deep learning related?

Deep learning is part of ML, which is part of AI; not all AI is ML.They are three completely separate fields.All AI is deep learning.ML is broader than AI.

9. What distinguishes supervised from unsupervised learning?

Supervised uses labelled data; unsupervised finds structure in unlabelled data.Supervised needs no data at all.They are the same thing.Unsupervised always uses labels.

10. In a confusion matrix, what is a false negative?

The model predicted negative but the true answer was positive — a miss.The model predicted positive and was right.The model predicted positive but was wrong.The model made no prediction.

11. How is precision calculated?

TP / (TP + FP).TP / (TP + FN).(TP + TN) / total.FP / (FP + TN).

12. Why can accuracy be a misleading metric on imbalanced data?

A model can score high accuracy while missing every rare but critical case.Accuracy is always the best metric.Imbalanced data cannot be evaluated.Accuracy and recall are identical.

13. Why does generative AI sometimes produce confident but false output (hallucinations)? (safety)

It predicts plausible content, not verified facts, so it can be fluently wrong.It always retrieves checked facts from a database.Hallucinations are impossible.Fluent output is always true.

14. An AI-generated resource cites a reference to support a claim. What must the teacher do? (safety)

Verify the source exists and genuinely supports the claim.Trust it because the AI provided it.Ignore sources entirely.Assume all AI citations are real.

15. How is a colour image represented to a computer?

A grid of pixels, each with three numbers (RGB) for its colour.A single word describing the picture.One number for the whole image.A sound wave.

16. What key information does a bag-of-words text representation lose?

Word order and context — only word counts remain.The number of words in the text.It loses nothing at all.The language of the text.

17. Why can an NLP tool be unfair in a multilingual classroom?

They often work better for high-resource languages and fail others.They work equally well in every language.Language never affects an NLP tool.Only English exists in NLP.

18. Why must you always read and run code suggested by an AI assistant before using it? (safety)

AI-suggested code can be wrong and must be read, run and verified.AI code is always correct and safe.You should paste it without checking to save time.Running code is never necessary.

19. Which uses of AI on students are forbidden under the course's child-safety rules? (ethics/safety)

Profiling children, emotion inference, biometric categorisation and covert surveillance.Using synthetic data for a demonstration.Teaching students to verify AI output.Reviewing AI content before class.

20. What is the purpose of an AI impact assessment before building a school project? (ethics)

To weigh benefits against harms and decide, with safeguards, before building.To make the project look longer.It is only needed after harm occurs.Impact assessments are never required.