Artificial intelligence and machine learning

Categories of AI

• Narrow AI (weak AI): designed for one specific task or a limited range of tasks, such as recommending videos, recognising faces, filtering spam or playing chess. It cannot transfer its ability to unrelated tasks. All AI in use today is narrow AI.
• General AI (artificial general intelligence): a hypothetical system that could understand, learn and apply intelligence across any task at a human level.
• Super AI: a hypothetical system that would exceed human intelligence in every area.

Benefits of AI: automation of repetitive work, speed and accuracy on large data, availability at all hours, help in areas such as medical imaging. Drawbacks and concerns: job losses, bias learned from data, errors that are hard to explain, privacy, and (for general and super AI) questions about control and safety.

Machine learning

Machine learning (ML) is a branch of AI in which a system learns patterns from data to make predictions or decisions, improving with more data, instead of being explicitly programmed with rules.

Supervised learning uses labelled training data: each example has the correct answer (label).

1. Training data with inputs and labels is fed to the algorithm.

2. The model learns the link between inputs and labels.

3. It is tested on new labelled data to measure accuracy.

4. It then predicts labels for new, unlabelled inputs.

Used for classification (spam or not spam; which digit is in an image) and regression (predicting a house price). Accurate and measurable, but needs large amounts of labelled data, which is costly to produce.

Unsupervised learning uses unlabelled data. The algorithm finds structure on its own, such as clustering similar items (grouping customers by buying habits) or spotting anomalies (unusual transactions). No labelling needed and it can find unexpected patterns, but results are harder to evaluate and the meaning of the groups must be interpreted by people.

How ML systems are drawn

Supervised: labelled data → training → model → test (compare predictions with labels) → predictions on new data.

Unsupervised: unlabelled data → algorithm finds patterns → clusters or groups → humans interpret.

In both, the quality of the output depends on the quality and fairness of the data (biased training data gives a biased model).

Worked example

A bank wants to flag fraudulent card payments. Describe how supervised and unsupervised learning could each be used.

1. Supervised: train a classifier on past payments labelled 'fraud' or 'genuine'; it then labels new payments.
2. Unsupervised: cluster normal spending behaviour from unlabelled payments; payments far from any cluster are flagged as anomalies.
3. Supervised needs many labelled fraud cases; unsupervised can catch new kinds of fraud but gives more false alarms.

Answer: Supervised learns from labelled examples; unsupervised finds unusual patterns without labels.

Check yourself

1. A voice assistant that sets timers and answers simple questions is an example of which category of AI?
   • General AI
   • Narrow AI
   • Super AI
   • It is not AI
   Show answer

   Narrow AI. It handles a limited range of tasks.

2. Which is the key feature of supervised learning?
   • The data has no labels
   • The system is given no data
   • A human makes every decision
   • The training data is labelled with the correct answers
   Show answer

   The training data is labelled with the correct answers. Labels supervise the learning.

3. A shop groups its customers by buying habits without any predefined categories. Which approach is this?
   • Supervised learning (classification)
   • Unsupervised learning (clustering)
   • General AI
   • A rule-based expert system
   Show answer

   Unsupervised learning (clustering). No labels; the algorithm finds groups.

4. Predicting the price of a house from its size and location is an example of
   • clustering using unsupervised learning
   • super AI
   • regression using supervised learning
   • encryption
   Show answer

   regression using supervised learning. A numeric prediction learned from labelled examples.

5. Which category of AI exists today?
   • Narrow AI
   • General AI
   • Super AI
   • Every category exists today
   Show answer

   Narrow AI. General and super AI are hypothetical.

6. Explain how a machine learning model can become biased.
   Show answer

   If the training data is unrepresentative or reflects past prejudice, the model learns those patterns and makes unfair or inaccurate predictions for some groups.. Link biased data to biased output.

7. Give one drawback of supervised learning.
   Show answer

   It needs a large amount of labelled data, which is slow and expensive to produce, and it can only learn the categories it was given.. Labelling costs time and money.

Practise with instant feedback Print this lesson

Sources

• Pearson Edexcel International Advanced Subsidiary/Advanced Level in Computer Science specification (2026)

Last updated 8 November 2026. Spotted a mistake? Tell us.