What is AI, really?

Let's strip the hype

When people hear "artificial intelligence," they picture robots, sentient machines, or the sci-fi villain that takes over the world. That picture is entertaining, but it has almost nothing to do with the AI you will actually use.

Here is a one-sentence definition you can carry through the rest of this course:

AI is software that learned patterns from huge amounts of data and now produces useful outputs from new inputs.

That is it. No magic. No consciousness. No feelings. Just a very powerful pattern machine.

The pattern machine

Think of a child learning to recognize dogs. Nobody sits a toddler down and explains fur density, ear shape ratios, or breed classification charts. Instead, the child sees hundreds of dogs -- big ones, small ones, fluffy ones, short-haired ones -- and over time, something clicks. They point at a new animal and say "dog!" even though they have never seen that specific dog before.

AI works in a remarkably similar way. A model is shown millions (sometimes billions) of examples. Over time, it picks up on patterns in that data. Once trained, it can take a new input it has never seen before and produce a useful output based on the patterns it absorbed.

When your email app catches a spam message, it is not following a hand-written rule that says "if the email mentions a Nigerian prince, delete it." It learned from billions of emails which patterns tend to be spam -- certain word combinations, sender behaviors, link structures -- and it applies those patterns to every new message that arrives.

That is AI. That is the whole trick.

Narrow AI vs. general AI

Here is an important distinction.

Narrow AI (also called "weak AI") is built for one specific job. Your spam filter is narrow AI. So is the system that recommends your next Netflix show, the autocorrect on your phone, and the voice assistant that sets your morning alarm. Every AI that exists today is narrow AI. It does one thing, or a related cluster of things, extremely well.

General AI (often called AGI, for "artificial general intelligence") would be a system as flexible as a human mind -- one that could write poetry, diagnose a medical condition, fix a car engine, and negotiate a business deal, all without being specifically trained for each task. This does not exist. It is an active area of research and debate, but as of today, nobody has built it. When people worry about AI "taking over," they are usually imagining general AI. That is science fiction for now.

The AI tools you will use in this course -- ChatGPT, Claude, Gemini, and others -- are impressively versatile, but they are still narrow. They are language models, trained on text. They are very good at text-based tasks and surprisingly good at reasoning about ideas expressed in text. But they do not truly "understand" the world the way you do.

You have been using AI for years

This might surprise you, but you have been using AI for a long time already. Here are a few examples:

• Spam filter: Every major email provider uses AI to keep junk out of your inbox. You probably do not think about it, but it works thousands of times a day on your behalf.
• Autocorrect and predictive text: When your phone suggests the next word as you type, that is a small language model predicting what you are most likely to say.
• Maps and navigation: Google Maps, Waze, and Apple Maps use AI to predict traffic, suggest routes, and estimate arrival times. The reason your "25 minutes" estimate is usually right is because a model learned traffic patterns for that road at that time of day.
• Streaming recommendations: When Netflix says "because you watched..." or Spotify builds you a weekly playlist, those are recommendation models finding patterns between your behavior and millions of other users.
• Photo organization: When your phone groups photos by the people in them, that is image recognition AI at work.

None of these feel like "artificial intelligence" because they are quiet. They do their job in the background. But every single one is a trained model applying learned patterns to new data. You have been living with AI for years. This course is about learning to use it intentionally.

So what changed recently?

If AI has been around for years, why is everyone suddenly talking about it? The answer is language models got dramatically better between 2020 and 2024. Models like GPT-4, Claude, and Gemini can now hold a conversation, write essays, summarize documents, brainstorm ideas, and help with an enormous range of text-based tasks. They went from niche research tools to everyday assistants almost overnight.

The core idea did not change -- it is still the pattern machine. But the scale changed. More data, bigger models, and smarter training methods produced systems that feel qualitatively different from what came before. That is why you are here, and that is what the rest of this course is about: learning to work with these tools effectively.

What is next

In the next lesson, we will look under the hood. How does an AI model actually learn those patterns? We will use a cooking analogy that makes the whole process click -- no math required. Head to How AI learns when you are ready.

---