S2 ICT AI practice 4: ten common misconceptions corrected
- 2026-11-20
S2 ICT AI practice 4: ten common misconceptions corrected
Ten items are classroom errors such as AI equalling machine learning, 95 per cent accuracy being enough, tuning on the test set and lower training error being better.
S2 ICT AI practice 4: ten items S2 ICT AI practice 4: ten retry items
01 Paper details (check answers afterwards)
| Item | Detail |
|---|---|
| Questions | 10 |
| Suggested time | 25 |
| Difficulty mix | Foundation 4 · Intermediate 2 · Challenge 2 · Error redo 2 |
| Chapter | Chapter: S2 ICT: Introduction to AI and machine learning |
| Level | Judge each statement first, then write the correct practice. |
02 Questions (10 items)
- FoundationError: artificial intelligence and machine learning are the same thing. What is correct?
- FoundationError: 95 per cent accuracy is enough to judge a model. What should be done?
- FoundationError: the test set can be used for tuning. What is correct?
- FoundationError: lower training error is always better. What is correct?
- IntermediateError: more data always makes a model more accurate. What is correct?
- IntermediateError: standardising all data before splitting into train and test is fine. What is correct?
- ChallengeError: high precision always comes with high recall. What is correct?
- ChallengeError: an AUC of 0.5 is a decent model. What is correct?
- Error redoError: a bigger learning rate always trains faster and better. What is correct?
- Error redoError: unsupervised learning needs no labels, so it needs little data. What is correct?
03 Answers and one-line notes
| # | Answer | One-line note |
|---|---|---|
| 1 | Machine learning is only one branch of artificial intelligence. | Artificial intelligence also covers rule systems, search and planning, while machine learning is just the data-driven part. |
| 2 | Check precision, recall and the confusion matrix as well. | When classes are imbalanced a high accuracy can completely hide failure on the minority class. |
| 3 | Tune on a validation set and use the test set once at the very end. | Once the test set influences decisions it stops being objective and the score no longer reflects real performance. |
| 4 | Very low training error often means overfitting, so watch the validation error. | The model memorises noise along with the signal and fails as soon as new data arrives. |
| 5 | Data must be relevant, clean and representative to help. | Duplicated or skewed data only amplifies bias, and wrong labels actively teach the model errors. |
| 6 | Split first, then standardise with statistics from the training set only. | Computing the mean and standard deviation on all data leaks test information into the model. |
| 7 | The two trade off, so raising one usually lowers the other. | Raising the decision threshold lifts precision while lowering recall, so the balance depends on the use case. |
| 8 | An AUC of 0.5 is random, and a value near 1 is good. | AUC measures how often a positive outranks a negative, and 0.5 is a coin flip. |
| 9 | Too large a rate oscillates or diverges while too small a rate converges very slowly. | In practice start small and increase gradually, watching whether the loss curve falls smoothly. |
| 10 | Unsupervised learning still needs plenty of representative data. | Without labels only the manual annotation is saved, while clustering and dimensionality reduction still need enough samples to find reliable structure. |
04 What to do after marking
- After marking, redo only the wrong items and write down which step went wrong (formula, unit, or misread number).
- Repeat the same item two days later; move to the next paper only when both attempts are correct.
- Parents only need the answer and note columns; there is no need to explain every item.
Want more items on this chapter? The chapter page lists the other three papers in this set.Tutor Registration →
Questions and answers are written originally by the TutorZone team and are not taken from any school, publisher or examination authority paper. Every value is checked by computer.



