What AI is: definitions from standards bodies and researchers

Updated October 21, 2026

Artificial intelligence lacks a single universal definition. The NIST glossary and OECD describe machine-based systems that infer outputs from inputs, while IBM and Stanford emphasize human-like abilities. The EU AI Act classifies AI by risk, and academic textbooks treat it as the study of intelligent agents.

The field of AI and its academic roots

The Stanford Encyclopedia of Philosophy defines artificial intelligence as the field devoted to building artificial creatures that, in suitable contexts, appear to be animals or persons. The field officially began in 1956 at a Dartmouth summer conference where the term "artificial intelligence" was coined. Stanford Encyclopedia of Philosophy

The textbook "Artificial Intelligence: A Modern Approach" by Stuart Russell and Peter Norvig is adopted by over 1,500 schools and organizes the field around intelligent agents, problem solving, knowledge, reasoning, and learning. Its table of contents includes sections on search, logic, probabilistic reasoning, machine learning, and robotics. Artificial Intelligence: A Modern Approach

IBM, a major technology company, describes AI as technology that lets computers and machines mimic human abilities such as learning, understanding, problem solving, and autonomous decision making. Applications can see and identify objects, understand language, and act independently. IBM's definition of AI

Definitions from standards and government bodies

The NIST Glossary includes definitions that describe AI as an artificial system that can carry out tasks in changing and unpredictable conditions with minimal human oversight, or as a machine-based system that makes predictions, recommendations, or decisions for human-set goals. These definitions come from federal standards and publications. NIST Glossary: artificial intelligence

The OECD defines an AI system as a machine-based system that, for given human-set objectives which may be explicit or implicit, derives from its input the way to produce outputs like predictions, content, recommendations, or decisions that affect real or virtual environments. The definition also notes that different AI systems vary in their levels of autonomy and adaptiveness after deployment. OECD's AI system definition update

The European Commission's AI Act is the first comprehensive legal framework on AI worldwide. It sets out risk-based rules for AI developers and deployers, defining four levels of risk: unacceptable, high, transparency, and minimal. Prohibited practices include social scoring and harmful manipulation. European Commission AI Act

Common threads and divergences

Across these definitions, several elements recur: AI systems are machine-based, they operate according to human-defined objectives, and they produce outputs that influence the environment. The emphasis differs: IBM and Stanford highlight human-like capabilities, while NIST and OECD focus on autonomous inference and decision making.

The OECD definition introduces the idea that objectives can be implicit, such as a self-driving vehicle programmed to obey traffic rules without being explicitly told to protect lives. This captures systems that learn new objectives through reinforcement or imitation.

The AIMA textbook's structure, which includes chapters on intelligent agents, machine learning, and robotics, reflects an academic perspective that treats AI as the study of agents that perceive and act in their environment.

Why definitions matter for regulation

Definitions have practical consequences. They determine which systems are subject to regulation, such as the EU AI Act's risk categories, and which are excluded. A system that infers from input to generate recommendations would likely fall under the OECD definition, while a simple rule-based calculator might not.

For policymakers, a common definition enables interoperability across jurisdictions. The OECD's definition is designed to be technology-neutral and to accommodate systems with varying degrees of autonomy and adaptability.

For researchers, definitions shape the scope of the field. The Stanford entry notes that many core AI formalisms come from philosophy, including logic and probability, while modern AI increasingly relies on machine learning and deep learning.

Frequently asked questions

What is artificial intelligence?

Artificial intelligence has multiple definitions. IBM describes it as technology that enables computers to simulate human learning and problem solving. The Stanford Encyclopedia defines it as a field building artificial creatures that appear to be persons.

How do standards bodies define AI?

NIST provides several definitions, including an artificial system that performs tasks without significant human oversight. The OECD defines an AI system as a machine-based system that infers outputs from inputs based on human-defined objectives.

What does the EU AI Act say about AI?

The AI Act is the first comprehensive legal framework for AI. It classifies AI systems into four risk levels: unacceptable, high, transparency, and minimal, and bans practices like social scoring.

Why do definitions of AI differ?

Different organizations focus on different aspects. Researchers may emphasize human-like cognition, while regulators focus on autonomy and influence on environments. The OECD definition highlights explicit or implicit objectives and adaptive behavior.

How is AI defined in academic textbooks?

The textbook Artificial Intelligence: A Modern Approach, used in over 1,500 schools, organizes AI around intelligent agents that perceive and act, covering search, logic, machine learning, and robotics.