Machine Learning (ML)

Machine learning is the branch of AI where a system derives its behaviour from data rather than from rules a programmer wrote. You do not tell it that emails containing certain phrases are spam; you show it a large set of emails already labelled spam or not, and an optimisation process adjusts the model's internal parameters until its predictions match those labels well enough to generalise to mail it has never seen. That shift — from specifying logic to specifying examples and an objective — is the whole idea. It splits into a few families worth distinguishing, because they need different data and answer different questions. Supervised learning needs labelled examples and predicts a label or a number: will this trial convert, what category is this ticket, what will this house sell for. Unsupervised learning has no labels and finds structure instead: clustering customers into segments, flagging anomalies. Reinforcement learning learns from a reward signal produced by acting in an environment, which is how models are tuned to human preference and how game-playing and robotics systems are trained. Deep learning is not a fourth family but a technique — neural networks with many layers — usable inside any of them. For a SaaS team the hard part is rarely the algorithm, which is usually a library call. It is the data and the evaluation: whether you have enough labelled examples that reflect the traffic you actually get, whether your training data leaked information the model won't have at prediction time, and whether you measure success on data the model never trained on. The classic failure is a model that scores brilliantly in a notebook and poorly in production, and the cause is almost always one of those three rather than a bad choice of architecture. Worth knowing where ML is the wrong tool. If the rule is knowable and stable — tax bands, shipping thresholds, entitlement checks — write the rule. ML earns its complexity when the pattern is real but too intricate or too shifting to write down, and when you can tolerate being wrong sometimes.