What is machine learning?

Machine learning is a way of building software that learns from examples. In a traditional program, a person writes the rules, like exactly what makes an email spam. In machine learning, you show the system lots of examples, and it works out the patterns itself. It then uses what it learned to make predictions about data it has never seen.

How it works

1. Collect data: numbers, images, text or sounds, often with the right answers attached.
2. Train a model: the software adjusts its internal settings, called parameters, until its guesses match the examples.
3. Test it: some data is held back to check how well the model does on things it hasn't seen.
4. Use it: the trained model makes predictions on real, new data.

Google's example: a classic weather program needs physics equations, while a machine learning model finds the patterns in past weather data.

The three main types

• Supervised learning learns from labeled examples with known answers, like predicting a house price or spotting spam.
• Unsupervised learning finds structure in unlabeled data, such as grouping customers with similar habits.
• Reinforcement learning learns by trial and error, collecting rewards for good outcomes. It powers game-playing AI and helps train robots.

Where you meet it every day

• Recommendations on Netflix, YouTube and social feeds
• Fraud alerts on your bank card
• Face unlock and photo search on your phone
• Speech recognition, translation and chatbots

ML, AI and deep learning

All machine learning is AI, but not all AI is machine learning. Deep learning is a kind of machine learning that uses neural networks with many layers, and it's behind today's large language models and image generators.

The term itself dates to 1959, when IBM researcher Arthur Samuel wrote about a checkers program that could learn to play better than its creator.