Why this matters
People say "AI" about a spam filter, a translation app and ChatGPT, as if they were one thing. They are not. They are nested: each sits inside the one before it. Almost every confusing claim you will read about AI comes from mixing these four words up, so it is worth getting the nesting right once.
The idea
AI is the broad term: technology that performs intelligent human behaviour (learning, reasoning, judgment) on a computer. Speech recognition, image recognition and translation all count.
Inside AI sits machine learning: instead of a programmer writing the rules by hand, the system learns patterns from data and uses them to predict and judge. Spam filters and product recommendations work this way.
Inside machine learning sits deep learning, which uses neural networks, systems modelled loosely on brain cells, to learn complex patterns from large-scale data. It is what makes image analysis for self-driving cars and speech synthesis work.
Inside deep learning sits generative AI, which produces new data: text, images, audio, programs. ChatGPT and image generators live here.
So the containment runs one way only. Every generative AI is deep learning, every deep learning system is machine learning, and all of it is AI. The reverse is false, and that is the trap: deep learning is not "a different thing from" machine learning, and AI does not mean the same thing as machine learning.
Picture it
flowchart TD A["AI: intelligent behaviour on a computer"] --> B["Machine learning: learns patterns from data"] B --> C["Deep learning: neural networks, large-scale data"] C --> D["Generative AI: makes new text, images, audio"]
Worked example
A spam filter and a shopping recommender look like completely different products. One reads email and throws things away; the other suggests headphones.
Underneath they do the same job. Both are given a pile of past examples (messages already marked spam, or things people already bought), both learn a pattern from those examples, and both apply that pattern to something new. Neither was handed a list of rules by a programmer. That shared shape is what makes them both machine learning, and it is why the task being different does not put them in different categories.
This also explains a weakness. A system that learned from examples can only be as good as the examples it saw. Show it something rare that barely appeared in its training data and its judgment gets shaky, because there was no pattern there to learn.
Your turn
Try the practice questions for this lesson. Decide which statements about the four terms are true, pick the description that gets the nesting right, and match machine learning, deep learning and generative AI to what each one does.
Take it further
Your school is deciding whether generative AI may be used for homework.
- Investigate. Find one documented case where a generative AI gave an answer that sounded right and was wrong. Record where you found it: a news article, a report, or a prompt you tested yourself with a screenshot. An example you remember but cannot source does not count.
- Weigh it up. Note one benefit and one risk of using it to write a school report, for the student and for the teacher.
- Decide. Should it be allowed? Recommend one rule and give two reasons.
Speed is the obvious benefit. The risk is that the output is built to sound right, which is exactly what makes a wrong answer hard to catch.
Recap
- AI contains machine learning, which contains deep learning, which contains generative AI.
- Machine learning learns rules from data instead of being given them.
- Deep learning is machine learning built on neural networks and large data.
- Generative AI makes new data; it is a part of AI, not a synonym for it.