AI foundations · 01
Is every AI a chatbot?
No. “AI” is a big family of ways to make computers do tasks that seem to need thinking. Chatbots are one kind of product, built with one branch of that family. To understand any AI system, ask separate questions: what job does it do, how was it built, where did its learning come from, and how do people get to use it?
What you will be able to do
Place a spam filter, a route planner, a chatbot and an image generator in more than one fitting category, and explain why a word like “tensor” describes how numbers are stored, not a kind of AI.
Before you start
Nothing. This is the first stop. If you have never used a chatbot, that is fine too.
1 · You have already seen this
Everything is “AI-powered”
Many people picture a chatbot when they hear the word AI.
The words “AI-powered” turn up on phone apps, websites and adverts. The label rarely says what the system does or how it was made, and for many people the picture in their head is still a chatbot that answers questions.
Artificial intelligence is a big family of ways to make computers do tasks that seem to need thinking, and chatbots come from one branch of it. This page is a map of that family, with questions that help you place any AI tool on it. It shows how that works, lets you try it on real examples, and shows where people get it wrong.
2 · What happens, step by step
From a label to a place on the map
Five questions to ask about any AI tool. The section after this one goes deeper into steps 2 to 4, the one after that into steps 1 and 5, and the lab lets you use them on eight real examples.
How to place an AI tool
Five questions
- 1 · What is the job?Say what goes in and what comes out. Text, images, sound or sensor readings can go in; a category, a number, new content or an action can come out.
- 2 · Did a person write the rule?If a person wrote exactly what it does, that part is hand-written rules. With no learning and no search, most people would call it ordinary software.
- 3 · Does it try options or weigh chances?If it explores possible paths and picks a good one, that part is search and planning. If it reasons about how likely things are when facts are uncertain, that part is probability.
- 4 · Did it learn from examples?If it found its patterns in examples instead of rules a person typed in, that part is machine learning.
- 5 · Which part does what?Real products often mix branches, so answer the questions for each part separately. Calling something “AI-powered” tells you very little until you do.
Going deeper · steps 2 to 4
One family, several branches
Machine learning is a branch of AI, and deep learning is a branch of machine learning. Other branches are older and still in daily use.
People in the 1950s already called a chess-playing program “artificial intelligence.” It did not learn from data: it searched ahead through possible moves. Other early systems followed long lists of hand-written rules written with help from experts, such as doctors. Others reasoned with probabilities when facts were uncertain.
Machine learning takes a different route: instead of a person writing every rule, the system finds patterns in examples. Deep learning is machine learning with large neural networks, and it powers most of what makes AI news today, including chatbots.
Real products mix branches. A map app searches for a route and uses machine learning to predict traffic. Calling something “AI-powered” tells you very little until you ask which part uses which approach.
Going deeper · steps 1 and 5
Ask about the job, not the label
These questions are independent. Answering one does not answer the others.
| Question | Examples of answers | Why it matters |
|---|---|---|
| How was it built? approach | Rules, search and planning, probability, machine learning, or a mix | Tells you whether it can learn, and where its mistakes are likely to come from |
| Where does the learning come from? learning setting | Labeled examples (supervised), no labels (unsupervised), filling in hidden parts of the data (self-supervised), rewards (reinforcement) | Tells you what the system was rewarded for getting right |
| What is the output? task | A category (classification), a number (regression), groups (clustering), new content (generation), a ranking, an alert, an action | Tells you how to test it |
| What structure holds what it learned? model family | Linear model, decision tree, nearest neighbors, neural network; transformers are one kind of neural network | Tells you how it can be inspected and how much computing it needs |
| What goes in and out? modality | Text, images, audio, video, sensor readings, clicks | Tells you what kind of data and privacy questions come with it |
| How do people use it? system | A chatbot app, a feature inside a phone, a tool-using agent, a model that runs on your own computer | Tells you who controls it and what else is wrapped around the model |
Predict first
A photo app turns your sketch into a painting. Which of the six questions can you answer just from that sentence?
Show the answer
Only some of them. The task is generation and the modality is images in, images out. You cannot tell the approach, the learning setting or the model family from the description, although today image generators are almost always deep learning models. You would also want to know the system: does your sketch get uploaded to a company’s servers, or does it stay on your phone?
3 · Try it
Sort eight real examples
Pick every approach that fits each example, then check. More than one answer is often right.
Lab
Which approaches does it use?
Choose one or more tags for each example and press “Check.” The explanation tells you what the system actually does.
No AI runs here: the answers are written by ARQIV
Email spam filter
Moves likely junk mail out of your inbox.
Map app route planner
Finds a route from home to school.
Chatbot
Answers questions in everyday language.
Basic thermostat
Turns the heat on below 20 °C and off above 21 °C.
Chess-playing program
Chooses its next move in a chess game.
Phone camera that finds faces
Draws a box around each face before you take a photo.
Tax calculator in a spreadsheet
Applies this year’s tax rates to an income.
Warehouse robot
Carries shelves around a busy warehouse floor.
Here are the eight examples with the approaches each one uses and why. Cover the right-hand part of each line and decide for yourself first.
- Email spam filter: Machine learning, Hand-written rules. Most spam filters are trained on many emails people marked as spam or not spam, so this is machine learning (a classification task). They are usually combined with hand-written rules, such as blocklists of known bad senders.
- Map app route planner: Search and planning, Machine learning. Finding the route is a search problem: the app explores possible paths through a road network and picks a short one. Live travel-time estimates often come from machine learning on traffic data. One product, two approaches.
- Chatbot: Machine learning. The model inside is a large neural network trained on huge amounts of text, so this is machine learning (deep learning). The product around it also uses ordinary software, such as rules for formatting, and sometimes tools like web search.
- Basic thermostat: Not really AI. A fixed rule written by a person, with no learning and no search: this is ordinary automation, and most people would not call it AI. Some “smart” thermostats do learn your schedule, which would add machine learning.
- Chess-playing program: Search and planning, Machine learning. Chess programs search ahead through possible moves and replies. Many strong modern engines also use a small neural network, trained on positions, to score how good each position looks, so machine learning fits too. Older engines used hand-tuned scoring rules instead.
- Phone camera that finds faces: Machine learning. Detecting faces in a picture is a computer vision task, and today it is almost always done with a trained neural network: machine learning.
- Tax calculator in a spreadsheet: Not really AI, Hand-written rules. It follows rules a person typed in. That is useful software, and you could call it a tiny hand-written rule system, but it does not learn or search, so most people would not call it AI. Some old “expert systems” were built from thousands of rules like these, which is why both tags count.
- Warehouse robot: Search and planning, Machine learning, Hand-written rules. Robots are systems, not one method. Planning a path that avoids other robots is search and planning; recognizing objects with a camera is usually machine learning; safety stops are often simple hand-written rules. A robot is a machine that can combine several AI approaches.
Not everything automatic is AI. A thermostat rule or a spreadsheet formula is ordinary software: a person wrote exactly what it does. That is often the best choice, because it is predictable and easy to check. “Uses AI” is not a compliment or a warning by itself.
Words that are not branches
Where do tensors, chatbots and robots fit?
Lists of “AI terms” often put very different kinds of words side by side. Here is what each one actually names.
| Word | What kind of thing it is | Question it answers |
|---|---|---|
| Machine learning | An approach to building AI | How was it built? |
| Supervised learning | A learning setting | Where does the learning signal come from? |
| Classification | A task | What output is wanted? |
| Transformer, decision tree | A model family (architecture) | What structure holds what it learned? |
| Tensor | A way to store numbers inside the software | How is the data represented? |
| Text, images, sensor data | A modality (kind of input or output) | What information goes in and out? |
| ChatGPT, Claude, Gemini | A product built around models | How do people get to use it? |
| Robot | A machine that can combine several approaches | What does the whole system do? |
Predict first
Is a tensor a kind of AI?
Show the answer
No. A tensor is how machine learning software stores numbers: a single number, a list, a grid, or a grid with more directions. An image becomes a tensor of pixel values before a model can use it. It describes the data, the way “spreadsheet” describes how numbers are laid out. See How does a picture become numbers?
4 · Common mix-ups
What people get wrong about AI
Three ideas that trip people up, and what to think instead.
Mix-up: “AI” means chatbots.
Better: Chatbots are one product type. Spam filters, route planners, recommendation feeds and chess programs are AI too, built in different ways.
Mix-up: Machine learning and AI are two separate things.
Better: Machine learning is one branch inside AI. All machine learning is AI; not all AI is machine learning.
Mix-up: If a product “uses AI,” the whole product is AI.
Better: Usually one part uses a model and the rest is ordinary software, data and people. Ask which part.
5 · Check yourself
Three quick questions
Answer in your head or on paper before you open each one.
A music app suggests songs based on what millions of listeners played after the songs you like. Name the task and the approach.
Check your answer
The task is ranking (or recommendation): choosing an order of songs to show you. The approach is machine learning, because the pattern comes from listening data, not from rules a person wrote.
Your friend says, “The robot vacuum is AI, so it learned everything it does.” What is wrong with that?
Check your answer
A robot is a system that can combine approaches. Its route around the room can come from search and planning, its bump-and-turn behavior from simple rules, and only some parts, like recognizing a sock or a cable, from machine learning.
Put these in order from biggest group to smallest: deep learning, artificial intelligence, machine learning.
Check your answer
Artificial intelligence, then machine learning, then deep learning. Each one sits inside the one before it.
For younger learners
A grades 3–5 version
No tags and no vocabulary list: one sorting question, done together.
15 minutes · teacher-led
Did a person write the rule?
- Show four cards: a light that turns on when it gets dark, a game that finds the shortest path through a maze, a photo app that finds faces, and a chatbot.
- For each card, ask: Did a person write exactly what it does, or did it learn from lots of examples?
- Sort into two piles, then talk about the maze: it does not learn, it tries paths. Make a third pile for “it tries lots of options.”
Exit question
Say it in one sentence
“Is every AI a chatbot?” Good answers name one example that is not a chatbot and say how it works: “No, the maze game tries paths, and the photo app learned from pictures.”
Next activity: Train the Machine shows what “learning from examples” means.
Honest limits
What this map leaves out
What this page leaves out
Many branches get only a line here: knowledge representation, logic, planning under uncertainty, evolutionary methods and more. AI researchers also disagree about exact boundaries, and a textbook may draw the map differently. The point is not the perfect map; it is asking separate questions about any system.
Where to go next
Follow one branch
Next stop · 02
How can examples become a rule?
Machine learning, with a tiny fruit table: classification, regression and finding groups.
The neural branch
How AI works
Neurons, layers and the training loop behind deep learning.
Search and planning at work
AI in engineering & robots
How a million robots plan paths without bumping into each other.
Sources
Where this comes from
- Stuart Russell and Peter Norvig, Artificial Intelligence: A Modern Approach, 4th edition, table of contents: the breadth of AI beyond machine learning (search, knowledge, uncertainty, learning, robotics).
- Google for Developers, “What is machine learning?” (Machine Learning Crash Course).
- AI4K12, the Five Big Ideas in AI (perception, representation and reasoning, learning, natural interaction, societal impact).
- Chess Programming Wiki, “Stockfish NNUE”: a search-based chess engine that scores positions with a small neural network.
AI words on this page
Artificial intelligence, Machine learning, Deep learning, Neural network, Algorithm, Symbolic AI and expert systems, Search and planning, Classification, Tensor, Chatbot
All definitions live in one place: AI words in plain language →
Review status
Written by ARQIV. Last checked October 2026. Independent subject review and classroom pilots are still to come, so treat the activities as drafts and tell us about anything that is wrong or unclear.