AI foundations · Guided lesson
Learn machine learning from the beginning
Start with AI you already use, understand how machines learn from data, and then build a clear mental model of supervised, unsupervised, and reinforcement learning.
Step 1 · Begin with experience
You already use AI every day.
Before learning definitions, connect AI to familiar experiences. These systems perform tasks that normally require human abilities such as seeing, understanding language, recognizing patterns, and making predictions.
Face unlock
Computer vision recognizes your face.
Recommendations
Stores and streaming apps predict what may interest you.
Arrival time
Maps and delivery apps estimate a numeric travel time.
AI assistants
Language models interpret and generate human language.
Step 2 · Build the map
AI, machine learning, and deep learning are related, but not identical.
Artificial intelligence is the broad field. Machine learning is an AI approach that learns from data instead of relying only on explicitly written rules. Deep learning is a machine-learning approach built with multilayer neural networks.
Largest field
Artificial intelligence
Systems that perform tasks associated with human intelligence.
Subset of AI
Machine learning
Algorithms that learn useful patterns from data.
Subset of ML
Deep learning
Learning with multilayer neural networks.
Step 3 · Understand the learning process
Data goes in. A learned model comes out.
A machine-learning system studies examples, adjusts its internal parameters, and produces a model. We then give that model new data and ask it to predict, group, or choose an action.
Input
Training data
Past examples, features, labels, or interactions.
Learning
Training
The algorithm adjusts parameters to improve an objective.
Result
Model
A learned function or policy used on new cases.
Feature
An input such as distance, email text, or customer activity.
Label
A known answer such as “spam” or a delivery time.
Inference
Using the trained model on new, unseen input.
Step 4 · Learn the first type
Supervised learning
The model studies examples that include the correct answer. It learns a mapping from inputs to a known target.
Classification
Predict a category
Spam or not spam, loan approved or declined, cat or dog.
Output: one of a finite set of classes
Regression
Predict a number
Delivery time, property price, temperature, or demand.
Output: a continuous numeric value
Step 5 · Remove the answer key
Unsupervised learning
The data has features but no predefined target label. The algorithm searches for useful structure, and people interpret what the discovered pattern means.
Clustering
Place similar observations into groups, such as customer segments based on purchase behavior.
Association
Find items or events that frequently occur together, such as products commonly purchased in one basket.
Step 6 · Learn through consequences
Reinforcement learning
An agent takes actions inside an environment. Rewards encourage useful consequences; penalties discourage poor ones. The goal is to maximize cumulative reward across a sequence, not simply make one correct prediction.
Game intuition: a move can lose a piece now but still help win the match later.
1 · Observe
Read the current state
2 · Act
Choose an available action
3 · Receive feedback
Observe reward and next state
4 · Improve
Update the policy and repeat
Step 7 · Compare after learning
Now place the three approaches side by side.
| Question | Supervised | Unsupervised | Reinforcement |
|---|---|---|---|
| Learning signal | Correct answers | No target answers | Rewards and penalties |
| Main question | What should we predict? | What structure exists? | Which action works best over time? |
| Output | Category or number | Groups or relationships | Action policy |
| Example | Spam detection | Customer segments | Game-playing agent |
Step 8 · Check understanding
Can you identify the learning type?
Predict tomorrow’s sales from labeled history: supervised regression.
Discover groups of similar customers: unsupervised clustering.
Learn a sequence of game moves from reward: reinforcement learning.
Teach or revise offline
Use the complete guided presentation.
The PowerPoint follows this same teaching sequence and includes speaker notes, examples, diagrams, a glossary, discussion questions, and knowledge checks.
Next: continue through the AI foundations path.