Before anything else: "AI" is an old word for a lot of different things, and the thing you use every day is one company's product built on top of it. Get those two straight and half the confusing headlines stop being confusing.
Here is the whole thing, and everything else in this course is a detail hanging off it: AI is software that got very good at finding patterns in huge amounts of data, then using those patterns to predict what comes next.
That is it. The chatbot that writes your emails, the thing that generates images, the feed that somehow knows what you want to watch — all of them are versions of that one move. They look like completely different products because they were pointed at different data and asked to predict different things. Underneath, they are doing the same job.
Notice what is missing from that sentence: understanding. Nothing in there requires the machine to know what it is talking about the way you do. It requires a great many examples and a good sense of what usually comes next. Hold on to that — it explains almost every strange thing AI does later in this course.
It is easy to assume AI arrived in late 2022, because that is when it arrived for most of us. It did not. The term has been stretched over about seventy years of very different technologies: a 1960s program that imitated a therapist by rephrasing your own sentences back at you, chess programs, and now the thing that drafts your emails. Same two letters, wildly different machines.
The field has even had two long winters, roughly the mid-1970s through the early 1990s, when funding collapsed because the promises had badly outrun what the technology could actually do. Worth keeping in your back pocket — it is the most useful context there is for reading today's headlines.
So what actually changed? Two things, together, around 2017 to 2020. A new architecture called the Transformer, which could weigh a whole sentence at once instead of crawling through it word by word. And a willingness to train that architecture on an almost unreasonable amount of text and computing power. Neither was enough alone. Doing both, at that scale, is what produced the version everyone is now arguing about.
ChatGPT is one company's chatbot built on AI — the same way Google is one search engine and not the entire internet. It became the word people use for the whole category the way "Google it" and "Kleenex" did: by getting enormous first, not by being the only one.
Gemini is Google's. Claude is Anthropic's. Copilot is Microsoft's. All four are doing the same basic trick — predicting likely next words from patterns they learned — each with its own training data and its own tuning layered on top. That last part is why you can ask all four the identical question and get four different answers, in four different tones, and none of them is malfunctioning.
This distinction is not pedantry. "AI can now do X" and "one company's product can now do X" are very different claims, and almost every alarming headline you read quietly swaps the second for the first.
AI is pattern-finding at scale used to predict what comes next. It is roughly seventy years old and has disappointed people before. What changed recently was a new architecture plus enormous scale — not the arrival of understanding. And ChatGPT is one product built on all that, not the thing itself.