AI is a label, not a thing
In short AI is a stack of ideas: AI, machine learning, deep learning, language models. Ask which layer someone means.
"AI" is used for everything from a spam filter to a chatbot to a science-fiction robot. That is not sloppiness so much as history: the term has always been a moving label for software that does something we used to think needed a person.
At a glance
| Layer | What it is | Example |
|---|---|---|
| Artificial intelligence | Any technique that does what we would call intelligent | A rule-based chess program |
| Machine learning | Learns its behaviour from examples | A spam filter |
| Deep learning | Machine learning with large neural networks | Photo recognition |
| Language models | Deep learning on text at huge scale | Chat assistants |
The nesting dolls
The useful picture is a set of nested ideas, each a subset of the one before:
- Artificial intelligence is the broadest term: any technique that makes a machine do a task we would call intelligent. Early AI was mostly hand-written rules, such as chess programs and expert systems.
- Machine learning is the part where the program learns its behaviour from examples instead of being told the rules.
- Deep learning is machine learning with large neural networks, which is what powered the leap in image recognition, speech and translation from the 2010s on.
- Large language models are deep learning applied to text at enormous scale. They are what most people mean by "AI" today.
When someone says "AI", ask which of these they mean. A fraud-detection model, a photo tagger and a chat assistant are all AI, and they have very little in common in practice.
The label keeps moving
There is a well-known pattern: once a task becomes routine, we stop calling it AI. Reading handwritten postcodes, recommending films and recognising faces all counted as AI when they were hard. Now they are just features. The label tends to attach to whatever still feels surprising.
Narrow versus general
Every system in use today is narrow: very good at what it was built and trained for, unreliable outside it. A model that plays Go superbly cannot book a flight. Language models look more general because language touches everything, but they still have patterns of strength and failure that matter in practice.
"General" intelligence, a single system that learns anything a person can, is a goal people discuss and argue about. It is not a product you can buy, and it is not what powers the tools in this series.
What AI is not
- It is not a database of facts that gets looked up. Most systems store patterns learned from data, not a table of answers.
- It is not a mind. The words people use, such as "thinks", "understands" and "wants", are shortcuts that tend to mislead.
- It is not magic. It is software, with inputs, outputs, costs and failure modes, and it can be reasoned about like any other.
This post and the three after it are the appendix of the series: the AI basics every other post assumes.