What is an Algorithm? — AI Glossary

An algorithm is a step-by-step set of instructions for solving a problem or completing a task. In AI and computing, algorithms are the logical procedures that tell a computer exactly what to do — from sorting a list to training a neural network to generating a response to your question.

Every piece of software, every AI model, and every app on your phone is built on algorithms. The word comes from the 9th-century Persian mathematician Muhammad ibn Musa al-Khwarizmi, whose name was Latinized as “Algoritmi.” Algorithms have been the foundation of mathematics and computation ever since.

How Algorithms Work

An algorithm has three essential properties:

• Input — the data or problem it receives
• Process — the defined sequence of operations to apply
• Output — the result or solution produced

A simple algorithm might be: “If the number is greater than 10, print ‘large’; otherwise print ‘small’.” A complex algorithm might be the procedure for training a deep neural network — defining how backpropagation updates weights, how gradient descent iterates toward a solution, and how convergence is measured.

In machine learning, “algorithm” often specifically refers to the learning method — the procedure that adjusts model parameters based on training data. Linear regression, decision trees, k-nearest neighbors, and backpropagation are all learning algorithms. The algorithm defines the learning strategy; the data provides the examples; the trained model is the result.

Why Algorithms Matter

Algorithms are the logic layer of AI. Without algorithms, raw compute power is useless — you need a procedure to tell the hardware what to calculate. Good algorithms can solve problems faster (better time complexity), with less memory (better space complexity), or with fewer errors than alternatives.

Algorithmic choice also matters for fairness. The same data fed through different algorithms can produce different levels of bias. An algorithm that optimizes purely for accuracy might perform excellently on majority groups while failing on minorities — a key concern in AI governance. This is why “the algorithm did it” is not a sufficient explanation — humans choose the algorithm and its objectives.

Algorithms in Practice

AI and software rely on many categories of algorithms:

• Search algorithms — finding items in data (binary search, depth-first search)
• Sorting algorithms — ordering data (quicksort, mergesort)
• Optimization algorithms — finding the best solution from many possibilities (gradient descent, simulated annealing)
• Machine learning algorithms — learning patterns from data (random forest, backpropagation, PPO)
• Cryptographic algorithms — securing data (AES, RSA)

In everyday life, algorithms decide what you see in your social media feed, how your email is sorted, what ads appear on websites, and how Siri or Google Assistant interprets your voice. The recommendation algorithms of YouTube and TikTok are among the most behaviorally influential pieces of software ever written.

Common Misconceptions

Misconception: An algorithm is the same as an AI. An AI model uses many algorithms to learn and make predictions, but a simple algorithm (like a recipe or flowchart) is not AI. AI typically involves learning algorithms that update a model based on data.

Misconception: Algorithms are neutral. Algorithms encode choices made by humans — what to optimize for, what data to use, what trade-offs to accept. Those choices reflect values, and the outputs reflect any biases baked into those choices.

---

Key Takeaways

• An algorithm is a step-by-step procedure for solving a problem with defined inputs and outputs.
• In ML, the algorithm is the learning procedure that trains the model on data.
• Algorithm choice affects speed, accuracy, resource use, and fairness.
• Common AI algorithms include gradient descent, backpropagation, and random forest.
• Algorithms are never purely neutral — they encode the choices of their designers.

Frequently Asked Questions

What is the difference between an algorithm and a model?

An algorithm is the procedure used to train a model. A model is the result of running that procedure on data — a set of learned parameters that make predictions. The algorithm is the recipe; the model is the dish it produces.

What makes a good algorithm?

Correctness (produces the right output), efficiency (runs fast with minimal resources), scalability (handles large inputs well), and robustness (handles edge cases and noisy data gracefully). For AI systems, fairness and interpretability are increasingly required properties too.

What algorithm does ChatGPT use?

ChatGPT uses many algorithms in combination: the transformer architecture for the neural network structure, backpropagation and gradient descent for training, and a form of reinforcement learning (RLHF) for alignment. No single algorithm is responsible for the full behavior.

Are algorithms patentable?

Pure mathematical algorithms are generally not patentable under US law (Alice Corp. v. CLS Bank). However, specific software implementations of algorithms can be patented in some jurisdictions. This remains a contested legal area with significant implications for AI.

What is algorithmic bias?

Algorithmic bias occurs when an algorithm produces systematically unfair outcomes for certain groups. This can arise from biased training data, biased objective functions, or biased feature selection. See What is Bias in AI? for more.

---

Sources: Grokipedia — Algorithm · CMU: Algorithmic Foundations · MIT Press: Introduction to Algorithms (CLRS)

You May Also Like

• What Is Artificial Intelligence
• AI Glossary
• AI for Dummies
• How Does AI Work
• Essential AI Skills

---

Get free AI tips daily → Subscribe to Beginners in AI

Sources

This article draws on official documentation, product pages, and industry reporting. Specific sources are linked inline throughout the text.

Last reviewed: April 2026