CBSE Class 8 · Artificial Intelligence · 2026-27

Data and Fairness: Worksheet

Skills: Decomposition, Abstraction. Circle the correct answer. Show your working in the space beside each question.

What is a representative dataset?
- (a) One that includes all the kinds of people or cases the AI will meet
- (b) One that is very small
- (c) One with only one kind of example
- (d) One that nobody can see
A dataset has 900 cat photos and 100 dog photos. What is the likely problem?
- (a) The model may lean towards predicting “cat” and do worse on dogs
- (b) The model will be perfect
- (c) Dogs cannot be photographed
- (d) There is no problem with imbalance
A model is 95% accurate for one group of users and 70% for another. What should the team conclude?
- (a) The model is unfair to the second group and needs investigation
- (b) An overall average is all that matters
- (c) The second group is using it wrong
- (d) Nothing; 70% is the same as 95%
Many diseased leaves in a dataset were labelled “healthy” by mistake. What happens to the model?
- (a) It learns wrong patterns and may call sick plants healthy
- (b) It fixes the labels itself
- (c) Nothing changes
- (d) It becomes more accurate
What does “consent” mean when collecting data?
- (a) People agree to their data being collected and used
- (b) Data is collected secretly
- (c) Data is copied from anywhere
- (d) Data is sold without asking
A survey about how families use the internet is done only through an online form. Who is missing?
- (a) Families without internet access
- (b) Families with fast internet
- (c) Nobody
- (d) Only teachers

Answer key: Data and Fairness

(a) One that includes all the kinds of people or cases the AI will meet – Otherwise it may work poorly for missing groups.
(a) The model may lean towards predicting “cat” and do worse on dogs – Collecting more dog photos balances the data.
(a) The model is unfair to the second group and needs investigation – The team should check the data for that group and improve it.
(a) It learns wrong patterns and may call sick plants healthy – Checking labels during data exploration prevents this.
(a) People agree to their data being collected and used – Fair data is collected with consent.
(a) Families without internet access – So the results do not represent all families.