Machine learning (ML)

Machine learning is a branch of computer science in which systems learn from data to improve their performance on a task, without being explicitly programmed with rules. Rather than following hand-written instructions, a machine learning model is trained on examples and learns to generalize from them.

ML encompasses a wide range of techniques and problem types:

  • Supervised learning: The model is trained on labeled examples (input/output pairs) to learn a mapping from inputs to outputs. Used for classification (eg. spam detection, computer vision) and regression (eg. predicting house prices). Suppose you give a model a dozen photos, six labeled "car" and six labeled "van", and ask it to find a visual pattern separating the groups. With so few images it may seize on color (red for car, white for van) rather than shape, and then misclassify a white car as a van and a red van as a car. Two remedies reinforce each other. Provide more varied training examples to correct the bias in the data, and give the model feedback on its mistakes so it refines its decision-making. This kind of AI bias is a central concern for supervised systems.
  • Unsupervised learning: The model finds structure in unlabeled data with no human guidance, and is used for clustering, dimensionality reduction, and anomaly detection. ChatGPT is the prominent example. The volume of text on the internet and in digitized books is so vast that, over many months, the model learned to combine words meaningfully by itself, with humans then fine-tuning its responses. The analogy is giving someone a big pile of books in a foreign language, some with images, and eventually they work out that a particular word appears whenever a tree is pictured, another when a house is, and a nearby word that might mean "a" or "the". This self-directed learning from raw text is the foundation of large language models and modern natural language processing.
  • Reinforcement learning: An agent learns by interacting with an environment and receiving rewards or penalties. Used in robotics, game-playing, and fine-tuning language models.

Generative AI is a subfield of machine learning focused on models that produce new content – text, images, audio, and video – rather than classifying or predicting from existing data. Large language models (LLMs), such as GPT and Claude, are the most prominent example. They are neural networks trained using ML techniques on vast text datasets to generate human-like language. Most modern LLMs are built on the transformer architecture.

See also

References