Kinds of AI
AI isn't one thing. It's a toolbox. Some tools recognise, some predict numbers, some make new things, some take actions. All of them predict.
First, what do you think?
Is a chatbot the same kind of AI as the face unlock on your phone? Have a guess.
Which kinds of AI are there?
The whole story in three sentences. AI is not one machine but a toolbox. Each tool learned to predict something different. Know the tool, and you know what to expect from it.
You already use several. Your phone unlocks when it sees your face. Your weather app says 80% chance of rain. A chatbot writes a birthday poem. A navigation app picks a route.
Four kinds cover most of what you meet.
- Recognising. The AI predicts what something is. Is this email spam? Is this a cat? Whose face is this? Experts call this classification.
- Predicting a number. The AI predicts an amount. What will this house sell for? How long will this ride take? How much rain will fall?
- Making. The AI makes something new: text, an image, music, code. This is called generative AI. Chatbots belong here.
- Doing. The AI takes steps by itself to reach a goal. A robot learns to walk around a room. An AI agent searches, clicks and fills in forms for you.
Most AI is narrow. It does one job well. A spam filter can't write a poem. A face unlock can't tell you the weather. Chatbots seem different, because they can write about almost anything. But they're still predictors: they predict text.
A first small step: pick one app on your phone. Which tool from the box does it use?
What the predictor does
Every tool in the box predicts. What it predicts is different.
- A recogniser predicts a label: spam, cat, your face.
- A number predictor predicts an amount: €310,000, 25 minutes, 80% chance.
- A maker predicts what comes next, again and again. The next word, the next bit of an image. Put enough predictions together, and you get a poem or a picture.
- A doer predicts the next step: which button to click, which page to open.
An example. You ask a chatbot for a poem. It doesn't look the poem up. It predicts one word, then the next, then the next. That's why you get a different poem each time.
AI is a toolbox. Every tool predicts, but each predicts something different.
Try it yourself
Which tool from the box is at work? Sort the examples.
Everything runs in your browser. Nothing you type is sent anywhere.
In practice
Three people pick a tool from the box. The right tool depends on what you want predicted.
Case 1The talk about volcanoes
Daan (12) has to give a talk about volcanoes. He asks a chatbot: "How many volcanoes are there?"
Copy the number into the slides.
A chatbot is a maker. It predicts text that sounds right. A number in its answer is a predicted word, not a counted fact.
Use the chatbot to plan the talk. Look the numbers up on a reliable website or in a book.
Each tool does what it was trained for. A maker makes text; a source gives facts.
Case 2The price of a flat
Esra (29) wants to sell her flat. A website gives an AI estimate of the price.
Take the estimate as the price.
It's a number predictor. It learned from flats that sold before. Her new kitchen or her noisy neighbours aren't in its examples.
Use the estimate as a starting point. Ask an estate agent who visits the flat.
The predictor knows the average. A person sees what's special about this flat.
Case 3The agent that books
Piet (71) hears a chatbot can book a train ticket for him. He wants to try.
Let it go ahead and pay.
An agent is a doer. It predicts the next click. One wrong prediction early on, and the next steps build on it.
Let it prepare the booking, and check every detail before you pay.
Agents can make errors that pile up. A check before the payment catches them.
By situation
Different tasks need different tools. A chatbot can explain or help you plan. A calculator calculates. A search engine or a book gives facts you can check. Using the wrong tool is the most common mistake.
Ask which kind of AI a tool is before you trust it. A recogniser that sorts invoices makes different mistakes from a maker that writes emails. The first can put something in the wrong pile. The second can write something that isn't true.
At home you use all four. Face unlock recognises. The weather app predicts a number. A chatbot makes text. A smart speaker that switches on lights does things. Knowing which is which helps you guess when it might go wrong.
Myths and mistakes
| Claim | True? | Why |
|---|---|---|
| All AI is a chatbot | No | Chatbots are one kind: makers. Recognisers and number predictors are older and everywhere. |
| A chatbot can do anything | No | It predicts text. That covers a lot, but it doesn't count, check or look things up unless it gets a tool for that. |
| Narrow AI is stupid AI | No | Narrow means one job. A spam filter can do its one job very well. |
| A chatbot looks up its answers | No | It predicts them, word by word. Some chatbots can also search the web, but the answer is still predicted. |
Related topics
Read first
- How AI learnswhere the patterns come from.
Next
- AI around youwhere you meet each tool in daily life.
Related
- Language modelsthe maker behind every chatbot.
- AI agentsthe doers.
Sources
- explainerGoogle for Developers. What is Machine Learning?.Machine learning trains software, called a model, to make useful predictions from data instead of hand-written rules. Four kinds: supervised, unsupervised, reinforcement learning and generative AI. Examples: spam, cats in photos, weather.
- governmentEuropean Parliament. What is artificial intelligence and how is it used?.AI in daily life: recommendations in online shops, web search, digital assistants, machine translation, smart thermostats, safety features in cars, cybersecurity.
- documentationAnthropic. Building effective agents.An agent is a language model that decides its own next steps and which tools to use. It starts from a task from a person. More freedom means higher costs and errors that pile up, so test it and set limits.
Test
You get 5 questions about the whole page. Get 4 right, and you've mastered the topic.
About the tools in this course. Every tool is described in the same way, in alphabetical order, without a recommendation. Names, prices and features change often: the maker's own site is always right. This site was built with the help of Claude, one of the tools described here.