AI vs ML vs Deep Learning vs Generative AI
Four terms you hear every day, and how each one sits neatly inside the one before it.
By the end of this lesson you will be able to
- define AI, Machine Learning, Deep Learning and Generative AI in one sentence each
- explain why each one is a subset of the one before it
- place a real product (ChatGPT, a spam filter, Face ID) in the right category
The big picture: circles inside circles
People often use these four terms as if they mean the same thing. They don't. The easiest way to remember them is as nested circles: each one is a special, more powerful part of the one outside it.
As you move right, capability goes up, and so does the need for data and computing power.
Each term in one sentence
Artificial Intelligence
The broad field of building machines that can mimic human intelligence. It includes rule-based systems that never “learn” at all.
Machine Learning
A part of AI where machines learn from data instead of being programmed with every rule.
Deep Learning
A part of ML that uses neural networks with many layers to discover complex patterns by itself.
Generative AI
A part of Deep Learning that creates new content: text, images, code, audio and video.
- AI is the entire subject of Mathematics.
- ML is learning algebra by solving many examples.
- Deep Learning is mastering advanced maths by discovering patterns yourself across thousands of problems.
- Generative AI is when, after mastering maths, you can write new questions, explain concepts and produce a whole textbook.
Side-by-side comparison
Scroll the table sideways on a phone. Read it one row at a time and notice how each column “grows” compared with the one on its left.
| Feature | AI | Machine Learning | Deep Learning | Generative AI |
|---|---|---|---|---|
| Definition | A broad field of creating machines that mimic human intelligence | Machines learn from data without being explicitly programmed | Uses multi-layer neural networks to learn complex patterns | Creates new content such as text, images, code, audio and video |
| Main goal | Make machines intelligent | Learn patterns from data | Learn highly complex patterns automatically | Generate content that looks human-created |
| Works by | Rules, logic, search algorithms, expert systems, ML | Training algorithms on historical data | Artificial neural networks with many hidden layers | Large neural networks: LLMs, diffusion models, GANs, Transformers |
| Needs training data? | Sometimes | Yes | Yes, usually huge datasets | Yes, very large datasets |
| Feature engineering | Mostly manual | Often manual | Learns features automatically | Learns representations automatically |
| Data requirement | Low to medium | Medium | High | Very high |
| Computing power | Low | Medium | High (GPU/TPU) | Very high (large GPU clusters) |
| Accuracy | Depends on the rules | Better than rule-based systems | Very high on complex tasks | Excellent for content generation |
| Creates new content? | Usually no | Usually no | Limited | Yes |
| Typical methods | Rule engines, expert systems, A* search | Linear regression, decision trees, random forest, SVM, K-Means | CNN, RNN, LSTM, Transformer | GPT, Claude, Gemini, Stable Diffusion, DALL·E |
| Example applications | Chess, expert systems, robotics | Spam and fraud detection, recommendations | Face and speech recognition, self-driving cars | ChatGPT, image generation, code generation, AI assistants |
| AWS services | Rekognition, Lex, Polly, Comprehend | SageMaker AI, Personalize | SageMaker AI, Trainium, Inferentia | Bedrock, Amazon Q, Amazon Nova, SageMaker AI |
| Everyday example | A robot following predefined instructions | An email spam classifier | Face unlock on your phone | ChatGPT writing an email or code |
The family tree
Here is the same relationship drawn as a tree, with the main branches you will meet in later lessons.
One example that uses all four
Imagine you are building the software for a bank. Each technology plays a different role:
| Technology | What it does for the bank |
|---|---|
| AI | Makes the overall banking system “smart”. |
| Machine Learning | Learns which transactions are fraudulent from past examples. |
| Deep Learning | Recognises customers from their face or voice. |
| Generative AI | Answers customer questions, writes emails, summarises documents and generates reports. |
Put each into the smallest circle it belongs to: a chess program that searches moves by fixed rules, Netflix recommendations, Face ID, a chatbot that writes poems, Google Translate.
Show suggested answers
Chess by fixed rules: AI (rule-based). Netflix recommendations: ML. Face ID: Deep Learning. Poem-writing chatbot: Generative AI. Google Translate: Deep Learning (a Transformer), and it is also generative because it produces new sentences.
Key takeaways
- AI → ML → Deep Learning → Generative AI: each is a specialised subset of the one before.
- ML learns from data; Deep Learning does it with many-layer neural networks; Generative AI uses those networks to create new content.
- Moving inward needs more data and more computing power, but unlocks harder tasks.
Check your understanding
1. Which statement is always true?
Deep Learning sits inside Machine Learning. Not all AI learns (rule-based systems don't), and most ML models predict rather than create.
2. An email spam filter trained on thousands of labelled emails is best described as…
It learns patterns from labelled historical data, which is the definition of Machine Learning.
3. What does Deep Learning do that classical ML usually doesn't?
Classical ML often needs humans to hand-craft features. Deep networks discover their own features, which is why they need more data and GPUs.