Is ChatGPT AI or machine learning?

Both. ChatGPT is a large language model, which is a kind of generative AI, built with deep learning, which is a kind of machine learning, which is a kind of AI. Each term is a narrower part of the one before it.

What is the difference between generative AI and agentic AI?

Generative AI creates content, such as text or images, when you ask. Agentic AI uses a generative model in a loop with tools, so it can plan and take actions, such as searching, booking or updating records, to complete a goal.

Is all AI machine learning?

No. Older AI systems use hand-written rules and search, with no learning from data. Most AI you use today, though, is built with machine learning.

The AI Family Tree: AI vs Machine Learning vs Deep Learning vs Generative AI vs Agents

AI terms get thrown around as if they mean the same thing. They don’t. The easiest way to understand them is as a family tree: each term is a branch of the one above it.

The tree at a glance

Artificial Intelligence
├── Rule-based / expert systems
├── Search & planning
├── Machine Learning
│   ├── Supervised learning (classification, regression)
│   ├── Unsupervised learning (clustering)
│   ├── Reinforcement learning
│   └── Deep Learning (neural networks)
│       ├── Computer vision
│       ├── Speech & audio
│       └── Generative AI
│           ├── Image, audio & video generation
│           └── Large Language Models (LLMs)
│               ├── RAG: answers from your own data
│               ├── Tool use & MCP: connecting to apps
│               └── AI Agents: plan → act → observe
└── Robotics (sensing + planning + control)

Read it top to bottom: every LLM is generative AI, all generative AI is deep learning, all deep learning is machine learning, and all machine learning is AI. The reverse is not true.

1. Artificial intelligence: the whole field

AI is the broad goal of making machines do things that normally need human intelligence: understanding language, recognising images, making decisions. Early AI used hand-written rules (“if the temperature is above 30, turn on the fan”) and search (chess programs exploring possible moves). These still exist in route planners and game AI.

2. Machine learning: learning from examples

Instead of writing rules, machine learning learns patterns from data. Show a model thousands of emails marked spam or not spam, and it learns to spot spam on its own.

Supervised learning learns from labelled examples: predicting house prices or classifying photos.
Unsupervised learning finds structure without labels: grouping customers with similar habits.
Reinforcement learning learns by trial and reward: game-playing AI and some robots.

Most “smart” features in apps (recommendations, fraud alerts, spam filters) are machine learning.

3. Deep learning: big neural networks

Deep learning is machine learning with large neural networks: many layers of simple units whose connections are adjusted during training. With enough data and computing power (usually GPUs), deep learning became far better than older methods at images, speech and language. It powers face unlock, voice assistants and translation.

4. Generative AI: creating new content

Generatieve AI uses deep learning to create: text, images, music, video and code. Image generators usually use diffusion models, which start from noise and refine it step by step into a picture. Text generators use transformers.

5. Large language models (LLMs)

LLMs are generative models trained on huge amounts of text to predict the next word (more precisely, the next token). After extra training to follow instructions, they become assistants that can write, summarise, translate, reason and code. Want the details? Read how generative AI works.

On their own, LLMs have two limits: they only know their training data, and they can only produce text. The next three branches fix that.

6. RAG: answering from your own data

Retrieval-augmented generation (RAG) searches your documents for the passages relevant to a question and gives them to the LLM, so it answers from your data with sources. It is the most common AI feature in companies. See our guide on how to build a RAG app.

7. Tools and MCP: connecting to the world

Tool calling lets an LLM ask your code to run a function: search the web, check an order, run a calculation. The Model Context Protocol (MCP) is an open standard for connecting AI apps to tools and data, so one integration works with many AI apps. See how to build an MCP server.

8. AI agents: from answering to doing

An AI agent puts an LLM in a loop with tools: it plans, takes an action, looks at the result and decides what to do next, until the goal is done. This is the move from generative AI (creating content) to agentic AI (getting work done). See how to build an AI agent and what is agentic AI?

Where robotics fits

Robotics combines AI with the physical world: sensors to perceive, AI and planning to decide, and motors to act. Modern robots increasingly use deep learning for vision and LLMs to understand instructions.

A simple way to remember it

AI = the goal.
Machine learning = learning from data.
Deep learning = learning with big neural networks.
Generatieve AI = deep learning that creates.
LLMs = generative AI for language.
RAG = gives LLMs knowledge. Tools/MCP = gives them abilities. Agents = gives them autonomy.

Learn it step by step

Our free AI basics roadmap walks through each branch with notes and quick study tips. When you’re ready to build, follow the AI Engineer roadmap.