The broader field
Machine learning is a branch of artificial intelligence in which a model improves at a defined task by finding statistical structure in examples. It includes linear models, decision trees, support-vector machines, clustering methods, neural networks, and many other approaches.
The word learning is technical. It means that an optimization or estimation process changes model parameters; it does not establish human-like understanding.
Training, validation, and testing
Training data is used to fit a model. Validation data helps choose settings and compare candidates. A separate test set estimates performance on examples not used to fit or select the model. Keeping these roles separate reduces the risk of mistaking memorization for general ability.
Metrics must match the real task. Accuracy alone may conceal failures affecting rare events or particular groups, and performance can decline when the world differs from the training data.
Major learning approaches
Supervised learning uses labeled examples, unsupervised learning searches for structure without target labels, and reinforcement learning adjusts behavior using rewards received through interaction. Self-supervised learning creates learning signals from the structure of the data itself.
These approaches may be combined. The choice depends on the available evidence, the desired output, the cost of mistakes, and whether feedback is reliable.
Strengths and limits
Machine learning can detect patterns too complex or numerous for hand-written rules and can adapt when retrained with relevant evidence. It is widely used for ranking, forecasting, recognition, anomaly detection, and decision support.
A correlation learned from past data may not survive a changed environment and does not prove causation. Models need representative data, appropriate evaluation, monitoring, and accountable human decisions around their use. A category describes how a system is built or used; it does not by itself prove accuracy, safety, intelligence, or consciousness. Real performance must be tested on the actual task and conditions of use.
Primary research and institutions
Sources are linked directly so readers can examine the underlying evidence. Numerical statements identify the reporting organization and year. Projections are presented as estimates, not established future facts.