What AI actually is.
You will know what AI, machine learning and generative AI each mean, where you already use them, and what intelligent means here.
Think about the last hour of your day. Your email app quietly dropped a few messages into the spam folder. Your maps app picked a route around a traffic jam. Your phone suggested the next word as you typed a text. None of these announced themselves as artificial intelligence. All of them were.
This lesson sorts out what the term actually covers, which parts of it you already rely on, and what people mean (and do not mean) when they call a machine intelligent.
Three words that get mixed up
Artificial intelligence, or AI, is the broad field: getting machines to do things that look like they need intelligence when a person does them. Recognizing a face, planning a route, translating a sentence, playing chess. The word "look" is doing real work in that definition. It says nothing about how the machine does it, only that the result resembles something we usually credit to a mind.
Inside that field sits machine learning, the part where the machine learns its behavior from data instead of following rules a programmer wrote by hand. The older approach was to write the rules down: if the email contains these words, mark it as spam. Machine learning turns that around. You show the system many examples of spam and not-spam, and it works out the pattern itself. Most of what gets called AI today is machine learning. How that learning happens is the subject of a later lesson.
Inside machine learning sits generative AI, the part that produces new things: text, images, audio, video and code. A spam filter sorts; it does not write. ChatGPT, Claude and Gemini write. An image model paints something that did not exist before you described it. "Generative" means the output is a fresh creation rather than a label or a decision.
So the three terms nest like bowls. AI is the biggest bowl. Machine learning sits inside it. Generative AI sits inside that. When a headline says "AI" today, it usually means the smallest bowl.
Narrow and general
Every AI system in use today is narrow AI: it does one kind of task, or a related cluster of tasks, and nothing else. The program that beats every human at chess cannot drive a car. The system that spots tumors in scans cannot answer your email. Even a chatbot that seems able to talk about anything is narrow in this sense. It produces words and pictures on request. It does not also drive, perform surgery, or keep pursuing goals of its own after you close the window.
General AI, often called artificial general intelligence or AGI, is the idea of a system that could learn and perform any intellectual task a person can, across every domain, without being rebuilt for each one. It does not exist. Whether it is close, far off, or not reachable along the current path is a genuine disagreement among serious researchers. When you read that AGI has arrived or is imminent, remember that the term has no agreed test, and that people using it often mean different things by it.
For this course, the practical point is simple. The tools you will use are narrow. They are very capable inside their lane and unreliable outside it, and the edges of the lane are not always where you expect.
The systems you already use
Most AI is invisible, because it works. A few examples you probably met today:
- Spam filters. Your email provider learns what spam looks like from the messages people report, and sorts new mail accordingly.
- Maps and routing. Predicting how long a drive will take, given the time of day and current traffic, is a learned estimate, not a lookup in a table.
- Recommendations. The next video, the next song, the "you might also like" row. These systems learn from what millions of people watched and bought.
- Phone cameras. Modern phones find faces to focus on, brighten dark scenes and blur backgrounds using models trained on huge numbers of photos.
- Autocomplete and autocorrect. The keyboard that guesses your next word is a small language model, a distant cousin of the chatbots.
None of these feel like science fiction, and that is the point. There is an old observation in the field: once something works reliably, people stop calling it AI and start calling it software. The chatbots feel different mostly because you talk to them directly, in your own words, and they talk back.
What intelligent means here
When a system is called intelligent, it is tempting to picture something like a person's mind inside the machine. That is not what is happening, and it is not what the researchers who build these systems mean.
What these systems do, at bottom, is find patterns in very large amounts of data and use them to make predictions. A spam filter has seen so many examples that it can predict whether a new message is junk. A chatbot has absorbed so much text that it can predict which words plausibly come next in almost any conversation. The scale is what makes this remarkable. Humans find patterns too, but we cannot read a sizable slice of everything ever written and keep track of every regularity in it. A large model can, and the results often look like understanding.
Two consequences are worth holding on to. First, the system is only as good as the patterns in its data. Where the data is thin, skewed or wrong, so is the output. Second, the system is doing something substantial, not nothing. Calling it "just pattern-matching" undersells it, because the patterns involved are deep enough to write working code, summarize a contract or explain a joke. Pattern-finding at this scale is a new kind of capability, and we do not yet have a tidy word for it.
Does it understand?
Here is the honest answer: nobody knows, and the question may be badly posed.
One camp argues that a system which predicts text, however well, is shuffling symbols without any grasp of what they refer to. It has never seen a dog, held a cup or felt cold; it has only seen the words. Another camp replies that to predict text this well about dogs, cups and cold, the system must have built some internal picture of how those things relate, and that this is a form of understanding, even if it is not the human form. Both camps include people who have spent their careers on the question.
You do not need to settle this to use the tools well. What you need is a working stance, and here is one that holds up: treat the system as something that is extremely good at producing plausible output and has no built-in guarantee that plausible means true. That stance keeps you curious about what it can do and careful about what it tells you. The rest of this course is about putting it into practice.
Try it
Open any chatbot and type: Give me three examples of AI I probably used today without noticing, and for each one say whether it is generative AI or not.
Read the answer against this lesson. Did it separate the broad field from the generative part? Then ask: Do you understand what you just wrote, or are you predicting text? Give me the strongest argument on each side, two sentences each. Notice how it handles a question with no settled answer. A good reply presents both sides without pretending the debate is over.
Key terms
- Artificial intelligence (AI)
- The broad field of making machines do tasks that look like they need intelligence when a person does them.
- Machine learning
- The part of AI where a system learns its behavior from examples in data instead of following rules written by hand.
- Generative AI
- Machine learning systems that produce new text, images, audio, video or code rather than sorting or labeling what already exists.
- Narrow AI
- A system that does one kind of task well and nothing else, which describes every AI system in use today.
- Artificial general intelligence (AGI)
- The idea of a system that could learn and do any intellectual task a person can; it does not currently exist.
Check yourself
Today's AI is narrow, learns patterns from data at enormous scale, and whether that counts as understanding is still an open question.