AI vs machine learning vs deep learning: clear definitions and examples

Artificial intelligence is the broad goal of making machines do tasks that need human-like intelligence; machine learning is the main way to get there today, by learning patterns from data; deep learning is a kind of machine learning that uses many-layered neural networks. This guide gives clear definitions, shows how the three nest, and places generative AI and large language models on the map.

AI vs machine learning vs deep learning: the short answer

Artificial intelligence (AI) is the broad field of building machines that perform tasks we associate with human intelligence, machine learning (ML) is the part of AI where systems learn those tasks from data instead of hand-written rules, and deep learning (DL) is the part of machine learning that uses neural networks with many layers. Picture three nested circles: AI on the outside, machine learning inside it, deep learning at the center.

The rule of nesting runs one way only. Every deep learning system is machine learning, and every machine learning system is AI. A rule-based chatbot is AI but not machine learning; a simple linear model that predicts house prices is machine learning but not deep learning.

What is artificial intelligence?

Artificial intelligence is the science of making machines do things that would need intelligence if a person did them: understanding language, recognizing objects, planning, solving problems and learning. The term comes from a 1955 proposal by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon for a summer workshop at Dartmouth College in 1956.

Their proposal set the agenda in one sentence: the study would proceed "on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it." The list of goals was close to today's: make machines "use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves."

Modern legal definitions are more operational. The EU AI Act (Article 3) defines an AI system as "a machine-based system that is designed to operate with varying levels of autonomy" which, for explicit or implicit objectives, "infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments."

AI covers many techniques. For decades the main ones did not learn at all:

• Rules and expert systems: experts write if-then rules, and the program applies them.
• Search and planning: the program explores possible moves or routes and picks the best, like a navigation app finding the shortest path.
• Logic and knowledge bases: facts and relationships are stored and reasoned over.
• Machine learning: behavior is learned from data. Since the 2010s this approach has dominated.

Chess shows the shift. Classic chess programs searched through possible moves and scored positions with rules written by experts: AI without learning. Later systems learned to evaluate positions by playing games against themselves: AI through machine learning.

What is machine learning?

Machine learning is the branch of AI where a program improves at a task by learning from data rather than following rules a programmer wrote. Google's introduction describes it as "the process of training a piece of software, called a model, to make useful predictions" or generate content from data.

Carnegie Mellon's Tom Mitchell gave the definition most textbooks use. In The Discipline of Machine Learning he writes that a machine learns "with respect to a particular task T, performance metric P, and type of experience E, if the system reliably improves its performance P at task T, following experience E." For a spam filter: the task is sorting email, the metric is the share sorted correctly, and the experience is a pile of emails people have marked as spam or not.

Traditional programming vs machine learning

The difference is where the logic comes from:

Nobody can write down the rules that separate a cat from a dog in a photo, but a model can learn them from a hundred thousand labeled photos.

The main types of machine learning

Google groups machine learning systems into four types:

1. Supervised learning: the model learns from examples with the correct answer attached. Predicting a home's price (regression) or flagging spam (classification) are typical.
2. Unsupervised learning: the model finds patterns in data without answers, such as clustering customers into groups.
3. Reinforcement learning: the model learns by trial, receiving rewards or penalties for its actions, such as a program learning to play Go.
4. Generative AI: models that create new content, text, images, music or code, from a user's input.

Classic machine learning methods, such as linear regression, decision trees, gradient-boosted trees and support vector machines, still run much of the world's fraud detection, credit scoring, demand forecasting and recommendation systems. They work well on structured data, train fast and are easier to explain.

What is deep learning?

Deep learning is a kind of machine learning that uses artificial neural networks with many layers, where each layer turns its input into a slightly more abstract representation. The 2015 Nature review by Yann LeCun, Yoshua Bengio and Geoffrey Hinton defines it this way: "Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction."

In an image model, early layers detect edges, middle layers detect shapes such as eyes or wheels, and later layers detect whole objects. Nobody programs those features. The network adjusts millions or billions of internal numbers, called parameters or weights, during training, using an algorithm called backpropagation to work out how each number should change to reduce its errors.

Deep learning vs classic machine learning

The practical difference is feature engineering. With classic machine learning, a person usually decides which inputs matter (the "features"), such as an email's sender domain or count of links. Deep learning learns useful features from raw data itself.

Deep learning took off in the 2010s when three things arrived together: large labeled datasets, GPUs fast enough to train big networks, and better training techniques. LeCun, Bengio and Hinton note that it "dramatically improved the state-of-the-art in speech recognition, visual object recognition, object detection" and fields such as drug discovery and genomics.

Where do generative AI and LLMs fit?

Generative AI and large language models (LLMs) sit inside deep learning: they are very large neural networks trained to produce new text, images, audio or code. ChatGPT, Claude and Gemini are AI products built on LLMs, which are deep learning models, which are a kind of machine learning.

Most of today's LLMs use the transformer, a network design introduced in the 2017 paper Attention Is All You Need. Its authors proposed an architecture "based solely on attention mechanisms," which trains faster in parallel than the recurrent networks used before. An LLM is trained to predict the next piece of text, a token, across a huge body of writing, then tuned further to follow instructions and be helpful. Our posts on how ChatGPT works and what a large language model is go one layer deeper.

So the full map, from outside in:

1. Artificial intelligence: any technique that makes machines act intelligently.
2. Machine learning: AI that learns from data.
3. Deep learning: machine learning with many-layered neural networks.
4. Generative AI: deep learning models that create content.
5. Large language models: generative models for text and code, such as the models behind ChatGPT, Claude and Gemini.

The terms in this map, plus tokens, embeddings and context windows, are defined in our AI glossary. For more on the generative layer, see what generative AI is.

How widely are these technologies used?

Machine learning in some form is now part of most organizations' operations, and generative AI spread faster than earlier technologies. Stanford's 2026 AI Index reports these figures:

The last figure explains why the newest deep learning models come from a handful of companies: training them takes data and compute that few universities can afford.

Examples of AI, machine learning and deep learning

The easiest way to tell the three apart is to ask how a system got its behavior: written rules, learning from data with simple models, or learning with deep neural networks.

Many real products combine all three: a bank app may use rules for limits, a gradient-boosted model for fraud scores and an LLM for its support chat.

Which one should you use or learn?

Choose the simplest approach that solves the problem: rules when the logic is clear, classic machine learning for structured data, deep learning for images, audio and language, and an existing LLM before you train anything yourself.

A quick decision guide for a new project:

1. Can you write the rules down? If yes, write code. It is cheaper, faster and easier to test.
2. Is your data a table of numbers and categories? Try classic machine learning (for example gradient-boosted trees) before a neural network.
3. Is your data images, audio or free text? Use deep learning, and start from a pretrained model rather than training from scratch.
4. Is the task reading, writing, summarizing or coding? Start with an LLM through an app or API and a good prompt. Fine-tune only if prompting and retrieval fall short.

For learning, a sensible order is: basic statistics and Python, machine learning fundamentals (training, testing, overfitting), then neural networks and deep learning, then transformers and LLMs. Our AI learning roadmap lays this out step by step.

How to explain the difference in one minute

Use this when a colleague asks:

AI is the goal: machines doing tasks that need intelligence.Machine learning is the main method: instead of writing the rules,we show the computer examples and it learns the pattern.Deep learning is machine learning with big, many-layered neural networks,which is what made image, speech and language AI work well.ChatGPT and Claude are deep learning models called large language models.

And a checklist for reading any AI claim:

• Ask whether the system learns from data or follows written rules.
• If it learns, ask what data it was trained on and how it was tested.
• If it is deep learning, expect it to need more data and to be harder to explain.
• If it is an LLM, expect fluent answers that can still be wrong, and check them.

If you go on to use LLMs for software work, coding agents are where they act on their own; Orkio runs agents such as Claude Code on an always-on cloud box so they keep working while your laptop is closed.

Frequently asked questions