English:Artificial Intelligence and Machine Learning
Artificial Intelligence and Machine Learning
Introduction
Artificial intelligence or AI is the field of creating computer systems that can perform tasks that usually require human-like abilities such as recognizing patterns, understanding language, making predictions, planning, or generating new content. Machine learning or ML is one important approach inside AI. Instead of writing a separate rule for every possible situation, people can train a machine-learning model on data so that it learns useful patterns.
You already meet AI in everyday life. Recommendation systems suggest music or videos, navigation apps estimate routes, spam filters sort messages, translation tools work with language, and computer-vision systems can identify objects in images. These systems are not human minds. They are designed tools that process information according to mathematical methods, computer programs, and the data used to build them.
This aiMOOC is designed for Grades 7–8. You will learn how data becomes a model, how different kinds of machine learning solve different problems, why models can make mistakes, and why fairness, privacy, safety, and human responsibility matter.
Learning Goals
By the end of this course, you should be able to explain the difference between artificial intelligence and machine learning, describe a basic training process, distinguish supervised and unsupervised learning, explain the role of neural networks at an introductory level, test simple model outputs, and discuss responsible uses of AI.
You will also practice asking an important question whenever you meet an AI system: What data, goals, assumptions, and human decisions may be shaping this result?
AI, Machine Learning, and Data
AI Is the Bigger Field
AI is the broad field. Machine learning is one set of methods used to build AI systems. This means that many modern AI systems use machine learning, but AI and machine learning are not identical terms. Some AI systems can also rely on programmed rules, search methods, logic, or combinations of several techniques.
A model is a mathematical or computational representation that has been adjusted to perform a task. For example, a model might estimate whether an email is spam, predict the next word in a sentence, or identify whether a picture contains a bicycle.
A model does not simply "know" the world. It works with the information represented in its data and the patterns it has learned. This is why the quality and relevance of data matter.
From Data to a Model
A typical machine-learning project follows a cycle. People define a problem, collect or select data, prepare the data, train a model, evaluate it, and then decide whether it is suitable for use. After deployment, the model may need continued monitoring because real-world conditions can change.
- Problem definition: Decide what task the system should perform and what a useful result would look like.
- Training data: Gather suitable examples and check whether the data represents the task fairly enough.
- Data preparation: Clean, organize, label, or transform the data so it can be used.
- Machine learning model: Train a model so that its internal parameters adjust to patterns in the examples.
- Model evaluation: Test the model on data it did not use for training and measure how well it performs.
- Inference: Give the trained model a new input and receive an output such as a prediction or generated response.
The diagram shows a machine-learning workflow with steps for importing data, processing it, visualizing it, building a model, and evaluating the model. Real projects may repeat these steps many times.
Features, Labels, and Examples
A feature is a piece of information used by a model. If you wanted to predict whether a plant needs water, useful features might include soil moisture, temperature, and time since the last watering.
A label is the correct answer attached to a training example in supervised learning. In an image dataset, a photo might have the label "cat" or "dog." Labels can be useful, but they can also contain mistakes or reflect human judgments.
Training data is used to adjust the model. Validation data can help developers choose settings while building the model. Test data is used to estimate how well the final model works on examples it has not already seen. Keeping test examples separate helps prevent an unfair test in which the model is checked on answers it has effectively memorized.
Types of Machine Learning
Supervised Learning
In supervised learning, the training examples include known answers. The model tries to learn a relationship between inputs and labels.
A classification task predicts a category. For example, a model might classify a message as spam or not spam. A regression task predicts a number, such as estimating the energy use of a building from weather and usage information.
Imagine training a model to sort pictures of apples and oranges. If each training picture is labeled correctly, the model can look for patterns in features such as shape, color, or texture. When it receives a new picture, it uses learned patterns to make a prediction.
The image above shows regions created by a classification tree. A decision tree repeatedly splits data using questions about features. The final regions represent the categories the model predicts.
Unsupervised Learning
In unsupervised learning, the training data does not provide a correct label for every example. The system looks for structure or patterns in the data.
One common task is clustering: grouping similar examples together. A shop might group products by buying patterns, or a science project might group measurements that have similar properties. The groups are created from patterns in the data, so people still need to interpret what the groups mean.
The animation illustrates the basic idea of k-means clustering. Points are repeatedly assigned to groups and the group centers are updated. It is a useful example of how an algorithm can discover structure without being given a name for every group.
Reinforcement Learning
In reinforcement learning, an agent learns by taking actions in an environment and receiving rewards or penalties. The goal is to learn a strategy that produces a good long-term result.
A simple classroom analogy is learning a game through repeated attempts. An action that helps reach the goal may receive a positive reward, while an unhelpful action may not. Real reinforcement-learning systems can be much more complex, and designers must be careful because a poorly chosen reward can encourage unwanted behavior.
Neural Networks and Deep Learning
A Network of Simple Units
An artificial neural network is a machine-learning model made from connected computational units often called neurons or nodes. Information moves through layers. Connections have adjustable values called weights. During training, the system changes these weights to reduce prediction error.
The diagram shows an input layer, a hidden layer, and an output layer. The word "neural" is inspired by biology, but artificial neural networks are mathematical systems and should not be confused with real brains.
Learning a Boundary
A very simple kind of artificial neuron is a perceptron. It combines input values, applies learned weights, and produces an output. A perceptron can learn a straight decision boundary for certain classification problems.
The image shows how a perceptron can update a linear boundary as more training examples are added. More advanced neural networks connect many units and can learn much more complicated patterns.
Deep Learning
Deep learning uses neural networks with multiple processing layers. Deep neural networks are especially important in areas such as image recognition, speech processing, language technology, and some generative AI systems.
Deep learning can be powerful, but larger models often need substantial data, computing resources, and careful testing. A complicated model is not automatically a good model. The right method depends on the problem, the data, the risks, and the resources available.
Testing and Evaluating Models
Accuracy Is Not the Whole Story
A model should be tested on examples that were not used for training. Accuracy is the fraction of predictions that are correct, but accuracy alone can hide important problems.
Suppose a medical screening dataset contains far more negative cases than positive cases. A model could achieve high accuracy by usually predicting "negative" while still missing many people who actually need help. This is why developers may also examine different kinds of errors.
A confusion matrix compares predicted classes with actual classes. It helps you see correct positive predictions, correct negative predictions, false positives, and false negatives. Which error matters more depends on the real-world situation.
Overfitting and Generalization
Overfitting happens when a model matches its training data very closely but performs poorly on new examples. It is similar to memorizing answers without learning the general idea.
Generalization means performing well on relevant new data. Good evaluation asks whether the model works beyond the examples it saw during training.
One useful habit is to keep training and testing separate. Another is to test with varied examples that reflect the situations in which the system will actually be used.
AI in Everyday Life
Computer Vision
Computer vision uses computers to process and interpret visual information such as photos or video. Machine-learning systems can be trained to classify images, detect objects, estimate positions, or identify visual patterns.
Applications include checking crops for signs of disease, helping organize large image collections, analyzing scientific images, and supporting driver-assistance systems. Results should still be checked for errors, especially when decisions could affect safety or people.
Language and Generative AI
Language models learn statistical patterns from large amounts of text and other data. Depending on their design, they can generate text, answer questions, summarize, translate, or help with writing and coding.
A generated answer can sound confident and still be wrong. This is sometimes called an AI hallucination. For schoolwork and important decisions, you should verify factual claims with reliable sources rather than treating an AI output as automatic proof.
Generative AI can also create images, audio, code, and other media. The output depends on the model, its training, the instructions it receives, and safety settings. Questions about copyright, privacy, consent, and misinformation are important when generated media is shared.
Recommendation Systems
Recommendation systems try to predict which items a user may find useful or interesting. They can use information about items, patterns across many users, or a mix of methods.
Recommendations can help you find relevant content, but they can also narrow what you see. A responsible user can compare sources, search deliberately, and remember that a recommendation is selected by a system rather than being a complete picture of what exists.
Responsible and Ethical AI
Bias and Fairness
Bias in an AI system can appear when data, labels, measurements, design choices, or the way a system is used create unfair or systematically different outcomes. Bias does not come only from a computer. Human choices shape what problem is defined, which data is collected, what counts as success, and how results are used.
If some groups are missing or poorly represented in training data, a model may work worse for them. Even balanced data does not guarantee fairness, because the task itself or the measurement method may still be problematic.
A useful fairness check is to ask who benefits, who could be harmed, who is represented in the data, who is missing, and whether people can question or correct an important decision.
Privacy and Data Protection
AI systems may use personal information such as text, images, voice, location, or behavior. Before sharing data with an online service, check what information is required and whether you have permission to share it.
Do not upload private information about classmates, family members, or other people into an AI tool without appropriate permission. Schools and organizations may also have rules about which services and kinds of data may be used.
Data protection is not only a technical issue. It is also about respect, consent, security, and giving people appropriate control over information about themselves.
Human Oversight and Responsibility
AI systems can support decisions, but people remain responsible for deciding where and how the systems should be used. Higher-risk uses need stronger testing, documentation, monitoring, and human oversight.
When AI is used for learning, you should understand your school's rules and be able to explain your own thinking. AI can help you explore ideas, but it should not replace your responsibility to learn, check sources, and communicate what work is yours.
Interactive Tasks
Quiz: Test Your Knowledge
Which statement best describes machine learning? (A method in which computer models learn patterns from data) (!A rule that every AI system must use a robot) (!A database that stores only correct answers) (!A type of computer screen)
How are artificial intelligence and machine learning related? (Machine learning is one important approach within artificial intelligence) (!Artificial intelligence is only another name for a spreadsheet) (!Machine learning includes every field of artificial intelligence) (!They are unrelated areas of study)
What is a label in supervised learning? (The known answer attached to a training example) (!A password used to open a model) (!A picture used only for decoration) (!The physical sticker on a computer)
Which task is an example of classification? (Predicting whether an email is spam or not spam) (!Predicting the exact temperature tomorrow) (!Measuring the length of a desk) (!Adding two numbers with a calculator)
What does clustering try to do? (Group similar examples based on patterns) (!Give every example the same label) (!Delete all unusual data automatically) (!Turn text into computer hardware)
What changes during the training of a neural network? (Values such as connection weights are adjusted) (!The laws of mathematics are rewritten) (!Every input becomes identical) (!The computer stops using data)
Why should test data be kept separate from training data? (To check performance on examples the model has not already learned from) (!To make the training set larger) (!To guarantee that every prediction is correct) (!To remove the need for evaluation)
What is overfitting? (Performing well on training data but poorly on new data) (!Learning a useful general pattern) (!Grouping similar points without labels) (!Protecting private information)
Why can biased training data be a problem? (It can contribute to unfair or systematically different model outcomes) (!It always makes a model faster) (!It prevents a model from using any features) (!It turns supervised learning into hardware)
What is a responsible response to an important AI generated factual claim? (Check it against reliable sources before relying on it) (!Assume it is correct because it sounds confident) (!Share it immediately without reading it) (!Ignore all evidence that disagrees with it)
Memory Game
| Training data | Examples used to adjust a model |
| Feature | Information used as an input to a model |
| Label | Known answer attached to a supervised example |
| Clustering | Grouping similar examples without supplied class names |
| Inference | Using a trained model to produce an output for new input |
| Overfitting | Learning training examples too narrowly to work well on new data |
| Bias | A systematic pattern that can contribute to unfair outcomes |
Drag and Drop
| Match the correct terms. | Topic |
|---|---|
| Supervised learning | Learns from examples that include known answers |
| Unsupervised learning | Finds patterns in data without a correct label for every example |
| Classification | Predicts a category |
| Regression | Predicts a numerical value |
| Reinforcement learning | Learns actions through rewards and penalties |
...
Crossword Puzzle
| Model | What trained system represents patterns for making outputs? |
| Feature | What input information can a machine learning system use? |
| Label | What known answer is attached to a supervised training example? |
| Cluster | What group contains examples judged similar by an unsupervised method? |
| Inference | What is the process of using a trained model on new input called? |
| Overfitting | What problem occurs when a model memorizes training patterns too narrowly? |
LearningApps
Cloze Text
Open-Ended Tasks
Easy
- AI Around You: Find three everyday systems that may use AI, create a one-page poster, and explain what input and output each system might use.
- Human or Machine Rules: Write five examples of tasks that could be solved with fixed rules and five that might benefit from learning patterns in data; explain one difficult choice.
- Data Detective: Keep a short data diary for one school day and identify what kinds of data a learning app could collect, then mark which items should stay private.
- Explain Machine Learning: Create a 60-second audio or video explanation of machine learning for a student one grade below you, using one clear example.
Standard
- Image Classification Experiment: With a teacher-approved tool, train a tiny image classifier using safe non-personal images, test it with new examples, and record where it succeeds or fails.
- Interview About AI: Interview a teacher, family member, or local professional about one way AI affects their work or daily life, then summarize benefits, limits, and one concern.
- Bias Audit: Design two small imaginary training datasets for the same classification task, make one dataset less representative than the other, and predict how the difference could affect results.
- Build a Mini Dataset: Create a small labeled dataset about classroom objects, explain your chosen features and labels, and describe how you would split it into training and test examples.
Advanced
- Decision Tree Challenge: Design a decision tree that sorts at least eight objects into categories, test it with new objects, and revise any question that causes repeated mistakes.
- Recommendation Investigation: Compare recommendations from two teacher-approved services or two different starting profiles, record noticeable differences, and explain what data or design choices might cause them.
- Model Evaluation Project: Use a set of at least twenty teacher-provided predictions and correct answers to build a simple confusion matrix, calculate accuracy, and explain which type of error matters most in the scenario.
- Responsible AI Proposal: Choose a realistic school use of AI and write a proposal that includes a goal, possible data, benefits, risks, fairness checks, privacy protections, human oversight, and a way to appeal an incorrect result.
Learning Assessment
- Explain the Relationship: Use your own example to explain why machine learning can be part of AI without being identical to the whole field of AI.
- Choose a Learning Method: For three different problems, decide whether supervised learning, unsupervised learning, or reinforcement learning is the best starting point and justify each choice.
- Evaluate a Dataset: Examine a sample dataset and identify missing groups, possible labeling problems, privacy concerns, and at least two improvements.
- Interpret Model Errors: Given a confusion matrix, explain what false positives and false negatives mean in the scenario and argue which error deserves more attention.
- Diagnose Overfitting: Compare training and test results from two imaginary models and decide which one shows stronger evidence of overfitting, using evidence from the results.
- Design for Responsible Use: Create rules for a school AI system that balance usefulness with fairness, privacy, transparency, and human review.
Evidence of Learning
Knowledge: You can accurately explain AI, machine learning, training data, features, labels, supervised learning, unsupervised learning, reinforcement learning, neural networks, inference, evaluation, overfitting, bias, and human oversight.
Skills: You can identify inputs and outputs, organize a simple dataset, compare learning approaches, interpret basic model results, spot possible sources of bias, and verify important AI-generated factual claims.
Products: Useful evidence may include your AI poster, mini dataset, classifier experiment log, interview summary, confusion matrix, decision tree, or responsible AI proposal.
Reasoning: You can explain why data quality, task design, evaluation methods, and context affect whether a model is useful or fair.
Transfer: You can apply these ideas to unfamiliar AI systems by asking what the system is trying to do, what data it uses, how it was tested, who may be affected, and where human judgment is needed.
OERs on the Topic
Linked Learning Areas