Starting at the beginning
You hear the term everywhere. In the news, at work, at a birthday party: Artificial Intelligence. It often sounds like something magical, or like something that will take over your job tomorrow. Neither is quite right.
In this chapter we calmly peel the term apart. No science fiction and no marketing talk, just the question: what is AI, and what does it actually do when you use it?
Learning objectives
After this chapter you can explain in your own words what AI is. You understand the difference between a computer that follows fixed rules and a computer that learns from examples. And you know why almost all the AI you come across today is "narrow AI".
AI in one sentence
Put simply: AI is a collective term for software that mimics human intelligence to perform tasks. Think of tasks we normally use our heads for: understanding language, recognising something in a photo, making an estimate.
So AI is not one device. It is not a robot, and it is not a particular computer. It is software, and that software is especially good at two things:
- Recognising patterns. AI can search through enormous amounts of data and find connections that we as humans overlook. You might look at a thousand photos before your eyes glaze over. An AI system looks at ten million without getting tired.
- Making predictions. Based on those patterns, the system makes an estimate. What probably comes next? Which answer fits best? Is this email spam or not?
First recognise a pattern, then predict what probably comes next.
Keep that word estimate in mind. We will come back to it a few times in this module.
A cookbook or a chef in training
How is AI different from ordinary software? Mostly in its approach.
Traditional software works like a cookbook. A programmer writes down step by step what the computer should do, and the computer carries it out exactly. If there is a mistake in the recipe, the dish fails. Every single time. The computer does not notice.
AI works more like a chef in training. The chef does not just read the recipe, but also tastes. Too salty? Less salt next time. Too bland? A bit more seasoning. The dish keeps getting better. An AI system learns in a similar way: through training with lots of examples and feedback on the result, not because someone thought up every rule in advance.
That also means nobody wrote down all the rules of such a system. The system extracted them from the examples itself. That is exactly what makes it powerful, and sometimes unpredictable.
Cookbook or chef in training
Traditional software
- Follows a fixed recipe, step by step
- A mistake in the recipe fails every time
- Learns nothing from the result
AI system
- Learns from lots of examples
- Tastes the result and adjusts
- Improves through training and feedback
Narrow AI and general AI
Time for an important distinction. Many people picture the robots from films when they hear "AI": computers as smart as we are, with feelings and a will of their own. We are not there.
In practice we distinguish two kinds:
- Narrow AI is what you see all around you today. Such a system is extremely good at one specific task, but cannot do anything else. A chess computer beats the world champion but cannot fry an egg. A spam filter recognises unwanted email but cannot chat about the weather. Narrow in what it can do, strong at that one thing.
- General AI is the theoretical form: an AI as versatile as a human. One that can learn, reason, plan and understand in every new situation, just like you. It does not exist yet. Researchers are working on it, but experts disagree strongly on whether and when it will happen.
Narrow AI: world class at chess, but it cannot fry an egg.
And the chatbots you may already use, such as ChatGPT, Copilot, Gemini or Claude? They can write about almost any topic, which makes them feel very broad. Still, they too count as narrow AI, and they are certainly not general AI: they have no human understanding, do not keep learning on their own in every situation, and make mistakes a person would rarely make. You will discover why in the next chapters.
An example you know: Netflix
Take something you probably use yourself: Netflix or Spotify.
You open Netflix and immediately see a row of recommendations. "Because you watched The Crown." How does Netflix know you probably feel like watching a historical series? That is narrow AI in action, in three steps:
- First, the data. The system keeps track of what you watch, how long you watch, when you drop off and what you give a thumbs up. Together that adds up to billions of data points.
- Then, the pattern. The system compares your viewing behaviour with that of millions of other viewers. People who liked The Crown often liked Downton Abbey too. No human draws that conclusion by hand. An algorithm does it in a fraction of a second.
- Finally, the prediction. Downton Abbey moves forward on your screen. If you click it, the system learns from that again.
How Netflix recommends a series
- Data
- Pattern
- Prediction
The system does not understand what drama is, or history, or why you enjoy period pieces. It sees patterns in viewing behaviour, nothing more. Ask the same system to do your tax return and you will get nothing useful back. It has only been trained on films and series.
Pause and think, no score: which AI did you use this week without noticing? Think of your email, your phone or your route to work.
Possible answer: the spam filter in your inbox, face recognition to unlock your phone and a route planner that estimates traffic. All narrow AI: each system is good at one thing.
In short
What you take away from this chapter:
- AI is a collective term for software that mimics human intelligence. It recognises patterns and makes predictions based on them.
- Ordinary software follows fixed rules. AI learns from examples and feedback.
- Almost all the AI you encounter today is narrow AI: very good at one task.
- General AI, with the versatility of a human, does not exist yet. Even clever chatbots are narrow AI, however broad they feel.
