Artificial intelligence refers to computer systems designed to perform tasks that involve abilities such as recognising patterns, interpreting language, making predictions or selecting actions. AI systems usually perform specific tasks rather than possessing general human intelligence.
Example: A photo application may use AI to recognise objects in images.
Practice: Identify three everyday applications that may use AI and state the task performed.
Remember: AI is a collection of computational methods, not a human mind inside a computer.
2
Artificial Intelligence, Automation and Rule-Based Systems
Automation performs tasks with limited human intervention, but not every automated system learns from data. Traditional rule-based systems follow explicitly programmed conditions, while many modern AI systems identify patterns from examples.
Example: A timer switching a light off after ten minutes is automated but does not necessarily use AI.
Practice: Classify five technologies as simple automation, rule-based computing or AI-based systems.
Remember: Automation and AI overlap, but they are not identical concepts.
3
Major Domains of AI
AI applications include data-driven prediction, computer vision, natural language processing, speech technologies, robotics and recommendation systems. Different domains process different forms of information.
Example: Computer vision processes visual information, while natural language processing works with human language.
Practice: Match image recognition, translation and product recommendation with suitable AI domains.
Remember: Real applications may combine several AI domains.
4
The AI Project Cycle
An AI project can be organised into stages such as problem scoping, data acquisition, data exploration, modelling, evaluation and responsible deployment. Iteration is common because results may reveal weaknesses in earlier stages.
Example: Poor model performance may lead a team to collect better training data.
Practice: Arrange the major AI project stages in a logical sequence.
Remember: Building an AI solution begins with understanding the problem, not selecting an algorithm.
5
Problem Scoping and Defining an AI Goal
Problem scoping identifies who is affected, what problem needs solving, what outcome is desired and what constraints exist. A clear problem statement prevents technology from becoming a solution in search of a problem.
Example: “Reduce avoidable food waste in the school canteen” is more useful than “make an AI app”.
Practice: Write a problem statement for an AI-assisted school-library project.
Remember: A useful AI goal should connect technology to a real need.
📊 Data Literacy for AI
6
Why Data Matters in Artificial Intelligence
Many AI systems learn patterns from data. The usefulness of a model therefore depends strongly on whether its data are relevant, representative and suitable for the intended task.
Example: A plant-recognition model needs examples of plants that reflect the conditions in which it will later be used.
Practice: Explain how poor data could produce poor AI predictions.
Remember: Better algorithms cannot automatically repair every weakness in the data.
7
Structured and Unstructured Data
Structured data are organised into defined fields such as rows and columns. Unstructured data include text, images, audio and video that require different processing methods.
Example: A spreadsheet of student attendance is structured data; classroom photographs are unstructured visual data.
Practice: Classify six examples of information as structured or unstructured.
Remember: The form of data influences how an AI system can process it.
8
Data Collection, Labels and Consent
AI data can come from sensors, surveys, transactions, observations or existing datasets. Labels identify target categories or values in supervised learning. Collection should respect permission, privacy and purpose.
Example: Images labelled “cat” and “dog” can be used as examples for a simple image classifier.
Practice: Plan what data would be needed for an AI system that sorts recyclable objects.
Remember: The ability to collect data does not automatically create the right to use it.
9
Data Quality, Missing Values and Imbalance
Datasets can contain errors, duplicates, missing information or unequal representation of categories. These issues can reduce model reliability and may create unfair outcomes.
Example: A classifier trained mostly on one category may perform poorly on categories with few examples.
Practice: Inspect a small sample dataset and list three possible quality problems.
Remember: Data cleaning and checking are essential parts of AI work.
10
Exploring Data with Tables and Visualisations
Before modelling, data should be explored using summaries and visualisations. Charts can reveal distributions, unusual values, relationships and possible errors.
Example: A bar chart can show whether one class label appears far more often than others.
Practice: Choose a suitable chart for comparing the frequency of five categories.
Remember: Data exploration helps students ask better questions before training a model.
🧠 Machine Learning Fundamentals
11
Machine Learning and Pattern Recognition
Machine learning is an approach in which algorithms use data to identify patterns and improve performance on a defined task. The learned model can then make predictions on new inputs.
Example: A spam filter can learn patterns associated with unwanted messages from labelled examples.
Practice: Explain how learning from examples differs from manually programming every possible rule.
Remember: Machine learning learns statistical patterns; it does not guarantee perfect understanding.
12
Supervised Learning
Supervised learning uses examples containing inputs and known target outputs. The model learns a relationship that can be applied to new data.
Example: Historical house features paired with known prices can train a model to estimate prices.
Practice: Identify the input and target output for a system predicting exam performance from study data.
Remember: Supervised learning requires target information during training.
13
Classification Problems
Classification predicts a category or class. Problems can involve two classes or several possible categories.
Example: An email classifier might predict “spam” or “not spam”.
Practice: Give three school-related problems that could be expressed as classification tasks.
Remember: Classification outputs categories rather than continuous numerical quantities.
14
Regression and Numerical Prediction
Regression models predict numerical values rather than categories. They estimate relationships between input features and a continuous target.
Example: A model could estimate electricity consumption from temperature and building-use data.
Practice: Decide whether predicting rainfall amount is a classification or regression problem and explain why.
Remember: Regression usually answers “how much?” rather than “which class?”.
15
Unsupervised Learning and Clustering
Unsupervised learning explores data without known target labels. Clustering attempts to group examples that are similar according to selected features.
Example: A system might group customers according to patterns in purchasing behaviour without predefined customer types.
Practice: Suggest features that could be used to group books by similarity.
Remember: A discovered cluster is a mathematical grouping and still requires human interpretation.
16
Training Data, Testing Data and Generalisation
A model learns from training data and should be evaluated on separate data it has not used for learning. This helps estimate whether it can generalise beyond memorised examples.
Example: Testing an image classifier only on its training images can give an unrealistically positive impression.
Practice: Explain why training and testing datasets should be separated.
Remember: The goal is useful performance on new data, not memorisation of old examples.
17
Evaluating AI Predictions
Evaluation compares predictions with known outcomes using measures appropriate to the task. Accuracy can be useful, but errors should also be examined because different mistakes may have different consequences.
Example: A model that predicts the majority class every time can appear accurate when a dataset is highly imbalanced.
Practice: If a model correctly predicts 85 of 100 examples, calculate its simple accuracy.
Remember: One performance number does not always tell the whole story.
👁️ Vision, Language and Generative AI
18
Computer Vision and Visual Information
Computer vision uses computational methods to analyse images and video. Tasks include classification, detection, segmentation and visual tracking.
Example: A plant-health application may analyse a leaf image for visible signs associated with disease.
Practice: Identify three situations in which visual recognition could be useful.
Remember: Computer vision processes numerical representations of images, not images in the same way humans experience them.
19
Image Classification
Image classification assigns an overall label to an image from a set of possible classes. Performance depends on training examples and how well they represent real-world variation.
Example: A model could classify an image as paper, plastic, glass or metal for a recycling activity.
Practice: Design four labels for a simple school-object image classifier.
Remember: Classification answers what category best describes the input according to the model.
20
Object Detection and Location
Object detection identifies both what objects are present and approximately where they appear in an image. It is therefore different from assigning one label to the entire image.
Example: A traffic system may detect several vehicles and mark their locations in the same frame.
Practice: Explain the difference between image classification and object detection.
Remember: Detection combines recognition with localisation.
21
Natural Language Processing
Natural language processing enables computers to work with human language in text or speech. Applications include translation, summarisation, search, sentiment analysis and conversational systems.
Example: A translation system processes a sentence in one language and generates an equivalent sentence in another.
Practice: List four NLP applications and identify their input and output.
Remember: Human language contains ambiguity, context and cultural meaning that make NLP challenging.
22
Chatbots and Conversational AI
Chatbots interact with users through text or speech. Some follow predefined conversation flows, while more advanced systems use language models to generate responses.
Example: A school-information chatbot could answer questions about opening hours and admission procedures.
Practice: Write five requirements for a safe and useful school-help chatbot.
Remember: Fluent responses can still contain mistakes and should be checked when accuracy matters.
23
Generative Artificial Intelligence
Generative AI systems produce new outputs such as text, images, audio or code based on patterns learned during training. Generated content can be useful for drafting and exploration but may be inaccurate or unsuitable.
Example: A generative system can draft a study outline from a user’s instructions.
Practice: List two useful applications and two risks of generative AI in education.
Remember: Generated output should not automatically be treated as verified fact.
24
Prompt Design and Evaluating AI Output
A prompt communicates the user’s task and constraints to a generative AI system. Clear context, desired format and success criteria can improve usefulness, while evaluation remains necessary.
Example: “Explain photosynthesis for Class 9 in 150 words with three key terms” is more specific than “tell me about plants”.
Practice: Improve a vague prompt by adding audience, task, constraints and output format.
Remember: Better prompting can improve output, but it cannot guarantee truth or remove the need for verification.
⚖️ Responsible AI and Future Skills
25
Bias and Fairness in AI Systems
AI systems can produce uneven outcomes when data, labels, assumptions or design choices reflect imbalances. Fairness requires examining who may benefit, who may be disadvantaged and how errors are distributed.
Example: A model trained on data that poorly represents a group may perform less reliably for that group.
Practice: Describe one way dataset imbalance could create unfair results.
Remember: AI does not automatically become neutral simply because decisions involve mathematics.
26
Privacy, Personal Data and AI
AI applications may use personal information, so data collection should be limited to legitimate purposes and protected appropriately. Users should understand what information is being collected and why.
Example: A school AI project should avoid collecting sensitive student information when anonymous or less intrusive data would be sufficient.
Practice: Identify three privacy questions to ask before collecting data for an AI project.
Remember: Collecting less unnecessary personal data can reduce privacy risk.
27
Explainability, Transparency and Human Oversight
Users may need to understand how an AI-assisted decision was reached, what information influenced it and what limitations apply. Human oversight is especially important when decisions affect people significantly.
Example: A teacher should not rely blindly on an automated prediction about a student’s future performance.
Practice: Explain why human review matters in high-impact AI decisions.
Remember: A prediction can support a decision without being allowed to make the entire decision.
28
Principles of Responsible AI
Responsible AI considers fairness, privacy, safety, reliability, transparency, accountability and human wellbeing throughout the project lifecycle. These principles should influence design choices from the beginning.
Example: Testing an AI tool with diverse users before deployment can reveal usability and fairness problems.
Practice: Create a six-point responsible-AI checklist for a student project.
Remember: Ethics is part of AI design, not an optional final step.
29
AI Careers and Interdisciplinary Skills
AI-related work includes data science, machine learning engineering, software development, research, product design, cybersecurity, ethics and domain-specific applications. Mathematics, communication and critical thinking complement technical skills.
Example: Healthcare AI projects may require collaboration among software specialists, data experts and medical professionals.
Practice: Choose one AI career and list five skills that could support it.
Remember: AI is interdisciplinary; strong solutions often require technical and human-domain knowledge.
30
Designing a Responsible AI Project
A student AI project should define a real problem, identify suitable data, choose an appropriate modelling approach, test results and consider risks. Documentation should explain both what the system can do and where it may fail.
Example: A recycling classifier project can define categories, gather balanced example images, test predictions and document limitations.
Practice: Draft a mini project plan containing problem, users, data, AI task, evaluation method and two ethical considerations.
Remember: A successful AI project combines useful performance with responsible design.