At school, there is a difference between a multiple choice exam and an essay exam.
In the first, you pick from a ready list: A, B, C, D. In the second, you write something that did not exist before you wrote it.
Most AI systems we knew for years were like the first exam: they classify, they pick, they decide. This message is spam or normal. This photo has a cat or it does not.
Generative AI is like the second exam. It does not pick from a list. It produces something new.
What is generative AI?
It is a type of artificial intelligence that creates content instead of only analyzing it: text, images, audio, video, or even code.
When you ask ChatGPT to write you an email, it is not looking for a ready email somewhere and copying it for you. It is building the sentence word by word, in that moment, based on your request.
This is why if you ask the same question twice, you may get two different answers. Because it does not retrieve. It produces.
How does it produce something it has never seen?
The idea is closer to educated guessing than you might think.
Imagine you have read thousands of formal letters in your life. If I asked you to write one now, you would automatically know that it opens with a greeting, that the tone is polite, that it ends with thanks. You would not be copying a specific letter, but you know the shape of letters in general.
A generative model does something similar. It read enormous amounts of text and learned which words usually follow which. When you ask it for something, it starts predicting the next word, then the one after that, until the answer is complete.
This simplification applies to text. Images and audio work differently in technical terms, but the general principle is the same: learn patterns from a huge number of examples, then produce something new that resembles them without being a copy.
Where do you find it today?
In chat tools like ChatGPT, Claude, and Gemini. In image generation like Midjourney. In writing code, editing video, cloning voices, and translating text in a way that sounds much closer to human.
It is also what moved AI out of the labs and into everyone’s phone in about two years.
What does it not do?
This is where the biggest misunderstanding sits.
A generative model does not know it is right. It builds an answer that has the shape of a correct answer, and that is not the same thing.
It might invent the title of a book that does not exist, attribute a quote to someone who never said it, or give you a precise number with no basis. And it presents all of that with the same confidence it uses for correct information. This is called hallucination.
The reason is simple: it was not designed to tell the truth. It was designed to continue text in a plausible way. Most of the time those two things match. But not always.
How do you use it well?
Treat it as a fast assistant writing a first draft, not as a final reference.
It is excellent when you want a starting point instead of a blank page, another phrasing for an idea you already have, a simple explanation of something you partly understand, or a review of text you wrote yourself.
It is dangerous when you ask it for numbers, dates, or sources and copy the answer without checking, especially in medical, legal, and financial topics.
The practical rule: the higher the cost of an error, the more human review it needs.
In short
Generative AI moved the machine from the role of analyst to the role of producer. It now writes, draws, and codes, and it is available to anyone with an internet connection.
But being able to produce is not the same as knowing.
It is very good at sounding convincing. Telling the difference between convincing and correct is still your job.