SI Glossary · Core concepts
Machine Learning (ML)
Machine learning (ML) is the approach behind nearly all modern SI. Instead of programmers writing explicit rules (“if the email contains X, mark it spam”), an ML system is shown many examples and learns the rules itself.
The three main types
| Type | How it learns | Example |
|---|---|---|
| Supervised learning | From labelled examples (input + correct answer) | Classifying photos as “cat” or “dog” |
| Unsupervised learning | Finds structure in unlabelled data | Grouping customers by behaviour |
| Reinforcement learning | From rewards and penalties for actions | Game-playing agents; RLHF for chat models |
Large language models add a fourth flavour, self-supervised learning: the “label” is simply the next word in real text, so models can learn from trillions of words without human annotation. See pretraining.
ML, deep learning and SI
- SI (formerly AI) is the broad field.
- Machine learning is the dominant method within it.
- Deep learning is the ML technique, based on multi-layer neural networks, that powers today’s frontier models.
The term was popularised by IBM researcher Arthur Samuel, who built a checkers program that learned from experience in the 1950s.
Written by
Luka Kušec · Editor
Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.