Paper 1 AI and ML Answers

These answers correspond to Paper 1 AI and ML Drills.

Answer 1: AI Definition

Artificial Intelligence is the ability of a computer system to perform complex tasks without constant human guidance and to improve its performance when more relevant data is collected and used.

Mark points:

• states that AI performs complex tasks without constant human guidance;
• states that AI can improve performance when more relevant data is collected and used.

Common weak answer:

• saying only “AI means computers are smart”, because this is vague and does not identify either defining feature.

Answer 2: AI Task Examples

Any four valid examples:

• face recognition;
• voice recognition;
• image classification;
• spam filtering;
• game playing;
• text generation;
• image generation.

Mark points:

• one mark for each valid task example, up to four.

Answer 3: ML Versus Traditional Programming

Traditional programming would require the programmer to write explicit rules, such as rules about suspicious words, links, or sender addresses.

Machine learning would use labelled examples of spam and non-spam emails to learn patterns, then use the learned model to classify new emails.

Mark points:

• traditional programming uses explicit hand-written rules;
• ML learns patterns from data or examples;
• answer applies the difference to the spam/not-spam scenario.

Common weak answer:

• saying ML is “automatic” without explaining that it learns from data.

Answer 4: Supervised or Unsupervised

Scenario	Learning type and reason
A set of past images is labelled as cat or dog; a model should classify new images.	Supervised learning, because the training examples already have known labels.
A shop has customer spending records but no customer group labels; it wants to find groups of similar customers.	Unsupervised learning, because the data has no given group labels and the goal is to discover clusters.

Mark points:

• identifies image classification as supervised;
• explains labelled examples;
• identifies customer grouping as unsupervised;
• explains missing labels or clustering goal.

Answer 5: k-NN

k-NN predicts a label by comparing the new item with stored labelled training examples. It calculates distances to the examples, selects the $k$ nearest examples, then predicts the majority label among those neighbours.

Mark points:

• compares the new item with labelled examples;
• calculates or uses distance/similarity;
• selects the $k$ nearest neighbours;
• uses the neighbours’ labels, usually by majority vote.

Common weak answer:

• saying it “chooses a random nearby point.”

Answer 6: k-Means

k-means is used for clustering unlabelled data into $k$ groups.

After initial centres are chosen, it repeatedly:

1. assigns each point to its nearest centre;
2. updates each centre to the mean position of the points assigned to that cluster.

Mark points:

• states that k-means is for clustering;
• states that data is unlabelled or unsupervised;
• assigns points to nearest centres;
• updates centres using the mean of assigned points.

Answer 7: Meaning of k

In k-nearest neighbours, $k$ is the number of neighbours used to classify a new item.

In k-means, $k$ is the number of clusters to form.

Mark points:

• correct meaning for k-NN;
• correct meaning for k-means.

Answer 8: ML Workflow

The standard forward sequence is:

gather data
prepare data
choose model
train model
evaluate model
tune parameters
make predictions

Mark points:

• one mark for gather data then prepare data before modelling;
• one mark for choose model, then train model, then evaluate model;
• one mark for tune parameters before final make predictions.

Real ML work may revisit earlier stages after evaluation, but this question asks for the standard forward sequence. Award full marks only when all seven supplied steps appear once and the dependency order is correct.

Answer 9: Problem Formulation

Possible input features:

• number of previous late submissions;
• attendance percentage.

Label:

• whether the next homework was late.

Learning type:

• supervised learning, because past records include the known outcome label.

Mark points:

• identifies a valid input feature;
• identifies a second valid input feature;
• identifies the label;
• identifies supervised learning with reason.

Answer 10: Model Evaluation

Correct predictions:

• Item 1: correct;
• Item 2: incorrect;
• Item 3: correct;
• Item 4: correct.

Accuracy:

$$\frac{3}{4}=0.75=75\%$$

Mark points:

• counts 3 correct predictions out of 4;
• gives accuracy as $0.75$ or $75\%$.