Types of artificial intelligence: from spam filters to chatbots
“Artificial intelligence” is not one program but a large number of different programs. Some recognize images, some speech, some write text or control a robot. In this lesson we make a “map” of the types of artificial intelligence.
Narrow and general artificial intelligence
- Narrow (specialized) AI does one job or one kind of job: it plays chess, recognizes faces, translates. All the artificial intelligence we use today is basically like this, even though some modern models (chatbots) are good at many different tasks.
- General artificial intelligence (often called AGI) would be able to learn and do almost any thinking job a person can do. It doesn't exist yet, and experts don't agree on when it could appear – or whether it is possible at all.
What AI does: the main kinds of tasks
| Type | What it does | Example |
|---|---|---|
| image recognition (computer vision) | finds and recognizes objects, faces, text in a picture | face unlock, plant recognition, self-driving cars |
| speech recognition | turns speech into text and back | voice assistants, video subtitles |
| language processing | understands and writes text, translates | translator, spell checker, chatbots |
| recommendations | predicts what you will like | YouTube, online shops, music |
| prediction | uses data to estimate what will happen | weather forecast, traffic jams, machine breakdowns |
| games and decision-making | chooses the best move | chess, go, opponents in video games |
| generative AI | creates new content | text, images, music, video, program code |
| robots | AI controls movement and arms | robot vacuum cleaner, factory robots, drones |
How AI learns: three basic ways
| Option | How it learns | An example from life |
|---|---|---|
| supervised learning | from examples with the correct answer (“this is a cat”, “this is spam”) | like learning with a workbook that has solutions |
| unsupervised learning | finds groups and similarities by itself, without correct answers | like sorting stickers into piles that “look alike” |
| reinforcement learning | tries, gets a reward or a penalty and improves | like learning to ride a bike or play a game |
We will do this in detail in the module How a machine learns. For now it's enough to remember: AI always learns from data – and how well it works depends on the data.
Robots and artificial intelligence
The word “robot” often makes us imagine a humanoid machine from the movies. In reality:
- A robot is a machine that moves and acts in the real world (a motor, an arm, wheels) – like the devices we built with Arduino and ESP32 (SR).
- Artificial intelligence is a program. It can run in a robot (a robot vacuum cleaner that makes a map of the apartment), but most often it runs on a phone or a server, without any body.
- Many robots don't use AI at all – they work by rules (e.g. a factory robot that always repeats the same movement).
Activity: make a map
For each example, write the kind of task from the table: (1) an app recognizes the song that is playing, (2) the phone suggests the next word, (3) a program draws a dragon from a description, (4) a shop suggests products, (5) a car brakes by itself in front of a pedestrian, (6) an app translates a sign from a photo.
Solution
(1) speech/sound recognition; (2) language processing (text prediction); (3) generative AI; (4) recommendations; (5) image recognition + robotics (decision-making); (6) image recognition (text in a picture) + language processing (translation). Many systems combine several types.
Draw a big “AI map” with the kinds of tasks as branches; students add examples from their own experience.
Common misconceptions
- “All robots are artificial intelligence” – many robots work by fixed rules; and most AI has no body.
- “A chatbot is general artificial intelligence” – chatbots are very versatile, but they still have big limitations; whether and when general AI can appear is an open question.
Practice tasks
Try it yourself first, and only then open the solution.
- What is the difference between narrow and general artificial intelligence?
- To play a video game, an AI gets points when it wins and loses them when it loses. Which way of learning is this?
- Name one device that is a robot but doesn't use AI, and one program that uses AI but is not a robot.
Solution 1: answer
Narrow AI is made for one task or one kind of task (e.g. face recognition). General AI could learn almost any thinking task like a person – it doesn't exist yet.
Solution 2: answer
Reinforcement learning – it learns from the rewards and penalties it gets after its moves.
Solution 3: possible answers
A robot without AI: a factory robotic arm that repeats the same movement, or a cart that follows a line by a rule. AI without a robot: a translator, a spam filter, a chatbot.
Check your knowledge
Open the quiz
Quiz: types of artificial intelligence
1. What is narrow artificial intelligence?
2. A program draws a picture from a text description. That is…
3. AI learns from examples with correct answers. That is…
4. Does every robot use artificial intelligence?