Machine learning is the subfield of artificial intelligence concerned with building systems that improve their performance on a task through exposure to data, rather than through explicit rule programming. A machine learning model is defined by a parameterised function; training adjusts those parameters using an optimisation algorithm that minimises a Loss Function measuring prediction error. The trained model then generalises its learned patterns to inputs it has never seen before.

Think of it like teaching someone to recognise fraudulent invoices not by handing them a rulebook but by letting them study thousands of examples of genuine and fake documents until they can reliably spot the difference on a new one. Their mental model of "what fraud looks like" emerges from pattern exposure, not dictated criteria.

The three main learning paradigms are supervised learning (labelled input-output pairs guide training), unsupervised learning (the model finds structure in unlabelled data), and reinforcement learning (an agent learns by receiving reward signals from an environment). A simple supervised example:

from sklearn.linear_model import LogisticRegression
model = LogisticRegression()
model.fit(X_train, y_train)          # learns from labelled data
predictions = model.predict(X_test)  # generalises to new inputs

Modern deep learning is a branch of machine learning that uses multi-layer Neural Networks as the parameterised function, with Gradient Descent and backpropagation handling optimisation. Key failure modes include Overfitting (performing well on training data but poorly on new data) and underfitting (the model is too simple to capture the underlying pattern). The scale of data and compute now available has produced Foundation Models and LLMs, but every one of them is still a machine learning model at its core: a function whose weights were adjusted by gradient signals computed from data.

You will encounter machine learning as the foundational layer beneath virtually every AI product — from spam filters and recommendation engines to voice recognition and language generation — and understanding it is essential for reasoning about why models succeed or fail.