Foundations
Algorithm
Also known as: procedure
A step-by-step set of instructions for solving a problem; the basic building block of all software, including AI.
An algorithm is just a recipe for a computer, a precise list of steps to get something done, like sorting names alphabetically or finding the shortest route. All software runs on algorithms. In everyday talk people say "the algorithm" to mean whatever an app uses to decide what you see, and with AI the "recipe" is increasingly one that the computer learned from data rather than one a person wrote out by hand.
An algorithm is a well-defined sequence of steps that transforms input into output to solve a problem. It is the foundational concept of computing.
Traditional vs learned
- Traditional algorithms: a human writes the exact rules (e.g. a sorting algorithm).
- Machine learning: instead of writing the rules, you use a learning algorithm to find patterns in data, and the resulting model is, loosely, a learned "algorithm".
"The algorithm" in daily life
When people blame "the algorithm" (on social media, say), they mean the automated system deciding rankings or recommendations, often ML-based. These systems can carry bias from their data and design, which is why algorithmic accountability matters.
Formally, an algorithm is a finite, unambiguous procedure computing a function, analysed for correctness and for resource complexity (time/space, e.g. Big-O). Classical AI was largely explicit-algorithm design (search, planning, constraint solving); the machine-learning shift replaced hand-specified rules with learning algorithms (e.g. gradient descent) that fit models from data, yielding learned decision procedures whose logic is implicit in parameters rather than human-authored.
This distinction matters for accountability: learned systems resist inspection (explainability) and can encode bias from data, motivating "algorithmic accountability" in law and ethics (the colloquial "the algorithm" usually denotes such deployed ML systems). Both senses coexist: modern AI stacks combine classical algorithms (data structures, search, optimisation, retrieval) with learned components, and algorithmic efficiency remains central to making large-scale training and inference feasible.