Machine learning

Machine learning (ML) is a branch of artificial intelligence focused on building systems that learn from data, identify patterns, and make decisions with minimal human intervention. The term was coined by Arthur Samuel in 1959, who defined it as the "field of study that gives computers the ability to learn without being explicitly programmed."¹ Central to ML is the concept of a model, which is trained on a dataset to capture underlying relationships. The training process adjusts model parameters to minimize a loss function, often using optimization algorithms like gradient descent. ML is typically categorized into supervised learning (where labeled examples are provided), unsupervised learning (where the algorithm finds structure in unlabeled data), and reinforcement learning (where an agent learns by interacting with an environment and receiving rewards).²

The roots of machine learning trace back to statistics and early cybernetics. In the 1940s and 1950s, pioneers like Warren McCulloch and Walter Pitts proposed mathematical models of neural networks. In 1957, Frank Rosenblatt invented the perceptron, an early neural network capable of binary classification.³ However, interest waned after Marvin Minsky and Seymour Papert's 1969 book Perceptrons highlighted limitations. The field revived in the 1980s with the development of backpropagation, which enabled multi-layer networks. The 1990s saw the rise of support vector machines and ensemble methods like random forests. The 2010s brought deep learning, fueled by large datasets and GPU computing, leading to breakthroughs in image and speech recognition.⁴

Supervised learning includes regression and classification algorithms such as linear regression, decision trees, and neural networks. Unsupervised learning encompasses clustering (e.g., k-means) and dimensionality reduction (e.g., PCA). Reinforcement learning has achieved notable successes in game playing, such as AlphaGo. Deep learning, a subfield, uses multi-layered neural networks to model complex patterns; architectures include convolutional neural networks (CNNs) for images and transformers for natural language processing.⁵ Ensemble methods like gradient boosting often win Kaggle competitions. Each approach has trade-offs in interpretability, accuracy, and computational cost.

Machine learning is deployed across industries: in healthcare for diagnostic imaging and drug discovery; in finance for fraud detection and algorithmic trading; in e-commerce for recommendation systems; and in autonomous vehicles for perception and control.⁶ It also powers natural language interfaces like virtual assistants and machine translation. The societal impact is profound, raising questions about privacy, bias, and employment. For instance, biased training data can perpetuate discrimination, as seen in some facial recognition systems.⁷ Regulatory frameworks, such as the EU's AI Act, are emerging to address these concerns.

Beyond the mainstream, machine learning has niche applications: in astronomy for classifying galaxies, in archaeology for detecting looting sites from satellite imagery, and in musicology for analyzing style. A notable edge case is the phenomenon of "catastrophic forgetting," where neural networks lose previously learned knowledge when trained on new tasks. The field also has a rich history of overlooked contributors, such as the mathematician Grace Hopper, whose work on compilers influenced early AI, and the lesser-known researcher Alexey Ivakhnenko, who developed deep learning in the 1960s.⁸ Another surprising fact: the term "machine learning" was popularized by Arthur Samuel, but the concept of learning machines dates back to Ada Lovelace's notes on the Analytical Engine.

Supervised learning

A type of ML where the model is trained on labeled data, learning to map inputs to outputs.

Unsupervised learning

ML that finds patterns in unlabeled data without predefined labels.

Reinforcement learning

An agent learns by interacting with an environment, receiving rewards or penalties.

Deep learning

A subfield of ML using multi-layer neural networks to model complex abstractions.

Overfitting

When a model learns the training data too well, performing poorly on new data.