Learn
6.3.1 | WHAT IS AI?
01 | THE SYLLABUS DEFINITION
Artificial intelligence (AI) is a branch of computer science dealing with the simulation of intelligent behaviours by computers.
These behaviours can include recognising patterns, applying knowledge to a problem or producing a reasoned recommendation.
Simulation matters: a computer can produce behaviour that appears intelligent without establishing that it thinks, feels or understands in the human sense.
Use the definition precisely. “A very fast computer” or “a robot” does not by itself explain AI.
02 | WHAT COUNTS AS INTELLIGENT BEHAVIOUR?
A person can recognise a familiar object, use evidence to reach a conclusion or learn a pattern from examples. AI systems simulate aspects of such behaviour using computational methods.
| Behaviour | Illustrative computer task |
|---|---|
| Recognise a pattern | Classify an image according to learned features |
| Apply knowledge | Use stored facts and rules to recommend a fault check |
| Make a prediction | Estimate an outcome from patterns in data |
| Adapt through learning | Update a model using training examples |
Not every AI system performs all of these. Learning is one possible approach, rather than a requirement that every AI system continuously teaches itself.
03 | TWO APPROACHES IN THIS COURSE
The specification focuses on expert systems and machine learning. This lesson introduces them; later lessons explain their characteristics and operation.
EXPERT SYSTEM
Use stored knowledge and rules to infer a conclusion or recommendation.
MACHINE LEARNING
Use data to adapt a model or process, enabling behaviour such as pattern classification.
AI is not limited to one fashionable application. A rule-based expert system can simulate expert reasoning without using a learned model.
04 | EXPERT SYSTEM EXAMPLE
Imagine a troubleshooting system for a printer. It asks whether power is present, whether paper is loaded and whether an error indicator is showing.
It uses stored knowledge and rules to determine which check to recommend next. This simulates part of a human expert’s problem-solving behaviour in a defined area.
Important: this is not just a list of answers shown at random. Inputs and relevant rules determine the conclusion.
The conclusion depends on the available knowledge, rules and input accuracy. The system does not automatically know every possible fault.
05 | MACHINE LEARNING EXAMPLE
Imagine a system trained on labelled images of recyclable materials. During training, it adapts its model to patterns associated with the labels.
It can then classify a new image according to those learned patterns. The result is a prediction, not a guarantee that the image has been correctly understood.
The model’s performance depends on relevant training data, its design and testing. An unfamiliar or unclear image can lead to a wrong classification.
A trained model need not change itself during every use. Training and using the trained model can be separate stages.
06 | AI IS NOT THE SAME AS AUTOMATION
| System | Main idea |
|---|---|
| Timer turns a light on at a set time | A simple programmed automation rule |
| Thermostat switches heating according to a threshold | Feedback control using a defined rule |
| Expert system uses facts and rules to infer a fault recommendation | Simulation of expert reasoning in a defined domain |
| Learned image classifier predicts a category | Pattern recognition using a model adapted from data |
AI can automate a task, but automation does not automatically imply AI. A system need not use simulated reasoning or learning simply because it runs without a person performing every step.
There is no single visible test that classifies every borderline system. Explain its method and intended behaviour rather than judging only by appearance.
07 | AI IS NOT THE SAME AS ROBOTICS
Robotics concerns the design, construction and operation of physical robots. AI concerns simulated intelligent behaviour.
A robot can follow a fixed movement sequence without AI. An AI classifier can run on a computer without a moving physical robot.
The two can combine: a robot might use image classification to identify an object before its mechanisms act.
A human-shaped body or a spoken response does not establish that the machine has human awareness.
08 | DATA, PROCESSING AND OUTPUT
INPUT INFORMATION
Facts, questions, measurements or images
↓ Apply the appropriate method
AI PROCESSING
Knowledge and rules, or a learned model
↓ Produce a result
OUTPUT
Classification, prediction or recommendation
The method gives the information meaning within the system’s task. Input data alone is not AI, and an output should be checked against the intended use.
This is a simplified view. Different AI approaches organise their processing differently.
09 | WHAT AI DOES NOT GUARANTEE
Correctness: a system can produce a plausible but wrong answer.
Fairness: unsuitable data or rules can lead to uneven or biased results.
General ability: success on one task does not prove competence on every other task.
Human understanding: intelligent-looking output is not proof of consciousness or emotion.
Continuous learning: many systems use fixed rules or a trained model that does not change during ordinary use.
These limits help you describe AI accurately without treating it as either magic or useless.
10 | HUMAN RESPONSIBILITY
People choose a system’s task, data, rules, training method and evaluation criteria. They also decide how its output will be used.
In a school image-classification project, students should test results on examples not used to fit the model and investigate mistakes.
A recommendation can support a decision, but the system’s capability and consequences determine how much human checking is appropriate.
Using the label AI does not remove responsibility for design, testing or safe use.
11 | A CLEAR EXAM ANSWER
Definition: AI is a branch of computer science dealing with the simulation of intelligent behaviours by computers.
Example: an expert system applies stored facts and rules to infer a troubleshooting recommendation, simulating reasoning in a defined area.
Identify the behaviour and method. “It uses a computer” is too broad, and “it thinks exactly like a person” claims more than the definition establishes.
12 | COMMON MISTAKES
- All automation is AI: simple timers and threshold control can work without the AI approaches studied here.
- AI must be a robot: software-only systems can simulate intelligent behaviour.
- Every AI learns continuously: expert-system rules or trained models may stay fixed during use.
- A fluent answer is necessarily correct: output needs appropriate checking.
- AI must have human emotions: simulation does not establish consciousness.
- More data always means better output: relevance and quality matter.
- AI is only machine learning: this course also includes expert systems.