Quick Answer: What is an algorithm?
An algorithm is a finite, ordered set of clear rules that turns an input into an output, and it appears everywhere from a recipe for bread to the ranking system behind a social feed. In computing, the same idea becomes a precise sequence of steps a machine can follow without guessing, which is why algorithms sit inside search engines, bank fraud checks, and navigation apps.
The term is often used loosely. Mathematicians treat an algorithm as any effective procedure, while regulators usually write rules about “AI systems” or “automated decision tools” rather than algorithms themselves, so the word means slightly different things depending on who is speaking.
Before trusting or comparing any algorithm, check its inputs and data sources, the steps and stopping rule, its time and memory cost, its accuracy and error rates, who is accountable for its outputs, and whether any law or standard governs its use.
A Word That Started With a Mathematician’s Name
Most people assume “algorithm” is a modern tech word. It is not. The term traces back to Muḥammad ibn Mūsā al-Khwārizmī, a ninth-century mathematician who worked at Baghdad’s House of Wisdom and wrote influential texts on arithmetic and algebra. When those works were translated into Latin in the twelfth century, his name was Latinized, and the resulting term gradually shifted from meaning “the decimal system” to meaning a step-by-step method of calculation. His exact birth and death dates are uncertain, usually given as roughly 780 to 850, so treat any precise date with suspicion. For a fuller picture of the man behind the word, the Encyclopaedia Britannica entry on al-Khwārizmī covers his life and mathematical work.
That slow evolution matters. Claims that “the algorithm” was coined in a specific year are unreliable. Usage drifted for centuries, and the modern computational sense only settled in during the twentieth century.
The Working Definition Most Experts Agree On
Strip away the jargon and an algorithm has a few non-negotiable parts. It takes some input, performs a finite number of well-defined steps, and produces output. The two requirements that do most of the work are finiteness (it must stop) and definiteness (each step must be unambiguous). A procedure that might run forever, or that leaves room for interpretation at each step, is not an algorithm in the strict sense.
Standards bodies have tried to pin the word down too. ISO/IEC 2382, the international vocabulary standard for information technology, defines an algorithm as a finite ordered set of well-defined rules for solving a problem. Britannica’s overview of algorithms is a useful companion for the general definition and history.
A practical test helps here. If two competent people follow the same written steps with the same input and can get different answers, you are looking at vague instructions, not an algorithm. That single test catches most sloppy descriptions in business documents.
What Makes a Set of Steps a Real Algorithm
Five properties separate a genuine algorithm from a rough plan:
- Input. It receives zero or more clearly defined values to work on.
- Output. It produces at least one result.
- Definiteness. Every step is precise, with no hidden assumptions.
- Finiteness. It terminates after a limited number of steps.
- Effectiveness. Each step is simple enough to be carried out in practice.
Miss any one of these and you have something else: a heuristic, a sketch, a policy, or a wish. This is where many workplace “algorithms” fall apart. A document that says “score the customer based on risk” has definiteness problems, because nobody has said how to score or what counts as risky.
Turing’s Machine and the Idea of Computability
The modern notion of an algorithm owes a great deal to Alan Turing. In his 1936 paper “On Computable Numbers, with an Application to the Entscheidungsproblem,” he described a formal model of step-by-step computation, now called the Turing machine. It reads symbols from a tape, follows a fixed rule table, and writes results back. Simple as it sounds, this model captures what it means for something to be computable at all. Turing’s original 1936 paper on computable numbers is readable for anyone who wants the source material rather than a summary.
The practical upshot: an algorithm is not tied to any particular machine, language, or era. It is an abstract recipe, and the Turing machine gave mathematicians a way to reason about which recipes can be carried out at all.
Algorithm vs. Program: The Distinction That Trips People Up
This is the most common mix-up. An algorithm is the logic; a program is one implementation of that logic in a specific language. The same sorting algorithm can be written in Python, Java, or C, and all three programs can be correct. Change the language and you have a new program, not a new algorithm. (Algorithm)
The same logic explains why two apps can “use the same algorithm” and behave differently. Different data, different tuning, different edge-case handling, and different hardware all change the output even when the underlying steps match.
| Concept | What it is | Example |
|---|---|---|
| Algorithm | Abstract, language-free sequence of steps | Sort a list by repeatedly picking the smallest remaining item |
| Program | Concrete code implementing an algorithm | A Python function that performs that selection sort |
| Model | A statistical or machine-learning structure fitted to data | A trained classifier that predicts loan default |
| System | The whole pipeline, including data, code, and people | A bank’s automated lending decision tool |
Keeping these four apart prevents a lot of confused arguments. When someone says “the algorithm is biased,” they usually mean the data, the model, or the surrounding system, not the abstract steps.
How Algorithms Are Measured: Time, Space, and Big-O
Algorithms are compared on two costs: time (how many operations they need) and space (how much memory they use). Both are usually expressed in Big-O notation, which describes how cost grows as the input gets bigger. An algorithm that scales linearly does roughly twice the work on twice the data; one that scales quadratically does about four times the work. That difference is invisible on small inputs and brutal on large ones.
In practice, the biggest mistakes are:
- Optimizing too early. Teams rewrite code for speed before measuring where time actually goes.
- Ignoring the data size. A method that works on a thousand records can collapse on ten million.
- Confusing average and worst case. Some algorithms are fast most of the time and disastrously slow in rare cases, which matters enormously in payment and safety systems.
What actually works is boring and effective: measure first, pick the simplest algorithm that meets the requirement, and only then tune the part that shows up in the profile.
Where Algorithms Show Up in Everyday Life
Algorithms are not confined to code editors. They run in places most people never think about:
- Navigation. Route-finding compares millions of possible paths and picks the fastest under current conditions.
- Search and feeds. Ranking systems order results by predicted relevance, which is why two people see different pages for the same query.
- Banking and insurance. Fraud detection and credit scoring flag unusual patterns for review.
- Hospitals. Triage and scheduling tools allocate scarce beds and staff.
- Compression. Every photo you send is shrunk by an algorithm that removes data the eye is unlikely to notice.
Notice how many of these involve judgment about people. That is exactly where the interesting questions begin.
Why the Word “Algorithm” Causes So Much Confusion
Definitions vary by field, and the slippage is real. Mathematicians may treat an algorithm as any effective procedure. Legal texts often avoid the word entirely: the EU Artificial Intelligence Act, Regulation (EU) 2024/1689, sets risk-based obligations for “AI systems” rather than algorithms, and its transparency duties apply to specific categories of systems. The official text of the EU AI Act on EUR-Lex is the place to check exact obligations and staggered application dates. Regulators elsewhere use “algorithm” loosely enough to include ordinary statistical models.
The result is a word that sounds technical but functions as a catch-all. When you read a headline about “the algorithm,” ask three questions: which specific method, applied to which data, making which decision? Without those answers, the claim is usually too vague to evaluate.
Accountability: Algorithm Registers and Public Scrutiny
Governments have started publishing inventories of automated decision systems, sometimes called algorithm registers. New York City’s Local Law 35 of 2021, effective in 2022, requires an annual report on automated decision-making tools used by city agencies. The New York City automated decision-making reporting requirements from the Office of Technology and Innovation show what that looks like in practice. These registers matter because they turn a hidden process into something the public can inspect, question, and challenge.
The lesson for organizations is straightforward. If an algorithm affects people’s access to housing, credit, benefits, or healthcare, expect to explain it. Documentation written after a complaint is far harder than documentation written at design time.
Common Mistakes People Make When Thinking About Algorithms
A few errors show up again and again, and each one is easy to avoid once named.
Treating algorithms as neutral. They encode choices: what to improve, what to ignore, what counts as an error. Those choices come from people.
Assuming more data always helps. Bad or unrepresentative data produces confident wrong answers. Cleaning and checking data usually beats adding more of it.
Forgetting the stopping rule. Loops without clear exit conditions are a classic source of bugs and runaway costs.
Confusing correlation with cause. A model that predicts well is not automatically explaining why something happens.
Skipping human review. High-stakes decisions need a path for appeal, correction, and override. Automated systems fail in ways their designers did not imagine.
What works in practice is unglamorous: write down the steps plainly, test them on edge cases, keep a log of changes, and have a person accountable for the outcome.
How to Read an Algorithmic Claim Critically
When a product, employer, or public body says an algorithm makes a decision, a short checklist cuts through most of the noise. Ask what the input is and where it came from. Ask what the output actually controls and whether a human can override it. Ask how errors are measured and on whom. Ask whether the process is documented and whether anyone outside the organization has checked it. Ask what happens when it is wrong.
None of these questions requires technical training. They are the same questions you would ask about any process that affects you, and they work because algorithms are, at bottom, procedures written down by people.
Conclusion
An algorithm is a finite, unambiguous sequence of steps that turns input into output, and the idea is older than computing itself, stretching back to a ninth-century mathematician whose name became the word. Turing gave the concept a rigorous foundation, standards bodies gave it a formal definition, and modern software turned it into the invisible infrastructure of daily life. The useful takeaway is not a definition to memorize but a habit of asking sharper questions: which steps, which data, which cost, which errors, and who answers when it goes wrong. Understand those, and “the algorithm” stops being a mysterious force and becomes what it always was, a procedure someone wrote and someone else can examine.
What is the difference between an algorithm and a computer program?
An algorithm is the abstract sequence of steps; a program is one concrete implementation of it in a specific programming language. The same algorithm can appear in many programs, and two programs can implement the same algorithm yet behave differently because of data, hardware, or edge-case handling.
Can an algorithm exist without a computer?
Yes. Algorithms predate computers by more than a thousand years. Long division, a recipe, knitting instructions, and a fire drill procedure are all algorithms when their steps are finite and unambiguous. Computers simply execute them quickly and at scale. (Algorithms)
What does Big-O notation actually tell you?
Big-O notation describes how an algorithm’s running time or memory use grows as the input gets larger. It ignores constant factors and focuses on the trend, so it answers questions like “will this still work on ten million records?” rather than “how many milliseconds does it take on this laptop?”
Are algorithms always objective?
No. Algorithms encode choices about what to improve, what to count as an error, and which data to use. Those choices come from people and can carry bias. Objectivity comes from careful design, testing on diverse cases, transparency, and human review, not from the mere presence of code.
What is a Turing machine and why does it matter?
A Turing machine is a formal model of computation that reads and writes symbols on a tape according to fixed rules. It matters because it gave mathematicians a precise way to define what can be computed at all, and it underpins the modern understanding of what an algorithm is.
Do laws regulate algorithms directly?
Rarely in those exact words. Many regulations target “automated decision systems” or “AI systems” instead, with obligations that depend on risk level and use case. Some cities also publish registers of automated tools used by public agencies, which lets residents see and question how decisions are made.
