What is Machine Learning?
How computers learn patterns from data instead of following hand-written rules.
You will learn
- Explain how ML differs from hand-written business logic.
- Name the main learning paradigms and when each applies.
- Connect ML fundamentals to LLMs and agent systems.
The core idea
Machine learning (ML) is a branch of artificial intelligence where systems improve through experience—specifically, by learning from data. Instead of a developer spelling out every rule (“if the email contains these words, mark it spam”), the system discovers patterns in examples and uses them to make predictions on new inputs.
A classic ML pipeline looks like this: collect data, choose a model family, train the model on historical examples, evaluate how well it generalizes, then deploy it to score or classify new data in production.
How it differs from traditional software
In traditional software, behavior is explicit: inputs flow through deterministic logic written by engineers. In ML, behavior is statistical: the same code can produce different outcomes as the model or data changes.
That shift trades predictability for adaptability. ML shines when rules are too complex to write by hand—vision, speech, recommendation, fraud detection—and when you have enough labeled or structured data to learn from.
Main types of learning
Most ML work falls into a few paradigms. Teams often combine them in production systems.
- Supervised learning — learn from input–output pairs (e.g. classify support tickets, predict churn).
- Unsupervised learning — find structure without labels (e.g. clustering customers, anomaly detection).
- Reinforcement learning — learn by trial and error with rewards (e.g. robotics, game playing, some alignment techniques).
Models, training, and inference
A model is the learned artifact—weights and architecture—that maps inputs to outputs. Training adjusts those weights to minimize error on training data. Inference is running the trained model on new data, usually much cheaper and faster than training.
Modern AI stacks often separate training infrastructure (GPUs, experiment tracking, data pipelines) from inference infrastructure (low-latency APIs, batch jobs, edge deployment).
Connection to LLMs and agents
Large language models are machine learning models trained on vast text corpora with self-supervised objectives. They are the engine behind many of today’s AI agents—but agents add orchestration, tools, memory, and policies on top of raw model inference.
Understanding ML fundamentals (data quality, evaluation, drift, failure modes) remains essential even when the interface is natural language.