S5 ICT: AI and Machine Learning, error rework
- 2026-11-20
S5 ICT: AI and Machine Learning, error rework
Paper four corrects the ten commonest misconceptions: training accuracy taken for reliability, quantity over label quality, automatic explanations, tuning on the test set, more layers always winning, correlation read as causation, no monitoring after launch, filling every gap with zero, accuracy on imbalanced data and the belief that AI output is neutral.
Error rework: ten model misconceptions to fix This paper corrects ten common misconceptions one at a time: that training accuracy of 99 per cent proves reliability, that more data outweighs label quality, that models explain themselves, that the test set may be reused for tuning, that more layers always help, that correlation proves causation, that no monitoring is needed after launch, that every missing value may be filled with zero, that accuracy suffices on imbalanced data, and that AI output is necessarily objective. Each item carries the correct view and a one-line explanation.
01 Paper details (check answers afterwards)
| Item | Detail |
|---|---|
| Questions | 10 |
| Suggested time | 25 |
| Difficulty mix | Foundation 3 · Intermediate 3 · Challenge 3 · Error redo 1 |
| Chapter | Chapter: S5 ICT: Artificial Intelligence and Machine Learning |
| Level | Error rework (10 items) |
02 Questions (10 items)
- FoundationIs it true that a model with 99 per cent training accuracy is certainly reliable in use?
- FoundationIs it true that more data always helps, whatever the label quality?
- FoundationDo machine learning models automatically explain why they decided something?
- IntermediateIs it acceptable to tune parameters repeatedly on the test set until the score looks good?
- IntermediateIs a model with more layers and parameters always better?
- IntermediateIf two variables are correlated, does one necessarily cause the other?
- ChallengeIs it fine to stop monitoring a model once it is deployed?
- ChallengeIs filling every missing value with zero the simplest reliable choice?
- ChallengeIs accuracy enough to evaluate a model on imbalanced data?
- Error redoAre the outputs of an artificial intelligence system necessarily objective and neutral?
03 Answers and one-line notes
| # | Answer | One-line note |
|---|---|---|
| 1 | Not true: its behaviour on test and real data must be checked. | High training accuracy may only mean the samples were memorised, not that the model generalises. |
| 2 | Not true: wrong labels teach the model the wrong pattern directly. | Cleaning and labelling quality usually matter more than sheer volume. |
| 3 | Not true: most models only output a result, and explanations need separate analysis. | Interpretability has to be designed in or obtained from tools such as feature importance. |
| 4 | Not acceptable: repeated tuning destroys the test set as an independent check. | Tune on a validation split and touch the test set only once at the end. |
| 5 | Not true: model size has to match the data volume and task difficulty. | An oversized model overfits small datasets and costs more to train. |
| 6 | Not true: correlation is not causation. | Causal direction needs controlled experiments or domain knowledge. |
| 7 | Not fine: shifting data distributions erode performance over time. | Keep tracking the metrics and retrain periodically on fresh data. |
| 8 | Not true: the filling method changes the outcome and must suit the column meaning. | Consider mean, median or model based imputation, and record which was used. |
| 9 | Not enough: recall and F1 must be read alongside it. | Guessing the majority class alone gives very high accuracy and no practical value. |
| 10 | Not true: bias in the training data is amplified by the model. | Audit the data sources and compare performance across groups. |
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 to compute precision and recall by hand? Book a TutorZone tutor and work through all ten items in 25 minutes.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.



