What is an algorithm?

An algorithm is simply a finite, well-defined sequence of instructions for accomplishing something. Long division is an algorithm. A recipe is an algorithm. The word predates computing by a millennium, deriving from the name of the ninth-century Persian mathematician al-Khwarizmi, whose work also gave us the word algebra. In computing, algorithms describe procedures for sorting lists, finding routes, compressing files, and encrypting messages, and the field studies how efficiently they use time and memory as inputs grow.

What people usually mean in conversation is narrower and worth naming precisely: the ranking and recommendation systems that decide what appears in a social feed, a search result, or a video queue. Those systems predict what you are likely to engage with based on your behavior and the behavior of similar users, then order content accordingly. They are not neutral pipes, because someone chose what to optimize for, and optimizing for engagement is a decision with consequences: content that provokes strong reaction performs well by that measure regardless of whether it is accurate or good for the person consuming it.

Two things worth holding onto. First, an algorithm can be entirely correct in its execution and still produce harmful results if it was trained on biased data or pointed at the wrong objective, which is the central concern in algorithmic accountability work covering hiring, lending, and criminal justice applications. Second, modern machine learning systems complicate the classical picture, because the rules are learned from data rather than written by a person, which means even the engineers cannot always explain a specific output. That is why explainability and auditing have become their own fields, and why the phrase the algorithm decided should prompt the question of who chose the objective rather than being accepted as an answer.