ISO/IEC 22989 — Artificial Intelligence Concepts and Terminology

What is this document?

ISO/IEC 22989:2022 is a foundational international standard titled "Information technology — Artificial intelligence — Artificial intelligence concepts and terminology" that establishes terminology for AI and describes concepts in the field of artificial intelligence. It was published in July 2022 by ISO/IEC JTC 1/SC 42.

The standard defines over 110 key concepts and terms in the AI domain, making it a foundational reference document upon which all other AI standards from the ISO/IEC JTC 1/SC 42 family rely.

Key points

Purpose of the standard

ISO/IEC 22989 serves as a common vocabulary for the AI ecosystem. Its main purposes are:

• Enabling consistent communication between different stakeholders (engineers, regulators, lawyers, managers)
• Supporting the development of other standards in the AI domain
• Providing foundational definitions used by regulatory initiatives such as the EU AI Act
• Establishing a common reference framework for understanding AI systems

Key concepts defined by the standard

Fundamental AI concepts

• AI system — The definition used as a starting point for the definition in the EU AI Act
• Model — A mathematical or computational representation of a system or process
• Algorithm — A finite set of instructions for solving a problem
• Dataset — Data for training, validation, and testing
• Agent — An entity that acts within an environment

AI system lifecycle

• Development phases: planning, design, development, testing
• Deployment phases: launch, operational use
• Management phases: monitoring, updating, decommissioning

Human roles and oversight

• Human-in-the-loop — A human directly involved in decision-making
• Human-on-the-loop — A human who oversees the system and can intervene
• Human-over-the-loop — A human who defines rules and constraints for system operation

This classification is directly relevant to the human oversight requirements under Article 14 of the EU AI Act.

Key properties of AI systems

• Transparency — Availability of information about system operation
• Explainability — Ability to explain system outputs
• Robustness — Maintaining performance under changing conditions
• Reliability — Consistency of system operation
• Resilience — Ability to recover from disruptions
• Safety — Avoidance of harmful consequences
• Security — Resistance to unauthorised access and attacks
• Privacy — Protection of personal data
• Risk — Effect of uncertainty on objectives

Categorisation of AI technologies

The standard categorises AI technologies including:

• Machine learning — supervised, unsupervised, reinforcement learning
• Natural language processing (NLP)
• Computer vision
• Robotics and autonomous systems
• Expert systems and knowledge-based systems

How does it apply to organisations?

Relevance to the EU AI Act

ISO/IEC 22989 is essential for understanding the EU AI Act because:

1. AI system definition — Article 3(1) of the AI Act defines an AI system in a way that builds upon the work of ISO/IEC 22989. Understanding the ISO definitions helps in correctly interpreting the regulatory scope.

2. Risk management terminology — Terms such as transparency, robustness, and explainability are used throughout the AI Act, and ISO/IEC 22989 provides their standardised definitions.

3. Human oversight classification — The human-in/on/over-the-loop concepts are directly relevant to fulfilling the requirements of Article 14.

Practical application

For compliance teams

• Use standardised terminology in internal documentation and policies
• Ensure consistent use of terms in communication with regulators
• Include a reference to ISO 22989 in the organisation's AI policy

For technical teams

• Adopt standardised definitions in technical documentation
• Use the AI technology classification when categorising your own systems
• Apply lifecycle definitions when planning development

For management teams

• Use the standard as a basis for understanding AI terminology
• Ensure everyone in the organisation uses the same terms consistently
• Include the standard in AI literacy programmes (Art. 4 of the AI Act)

Practical steps

1. Obtain the standard — Available through the ISO web store or national standards bodies
2. Adopt the terminology — Integrate standardised terms into internal documents
3. Educate teams — Conduct training on key AI concepts using the standard as a reference
4. Update documentation — Align existing documentation with standardised terminology
5. Track revisions — The standard is undergoing revision to align with new regulatory developments

Relevant EU AI Act articles

Article	Connection to ISO 22989
Art. 3	Definitions -> Foundational AI system definitions
Art. 4	AI literacy -> Understanding AI concepts
Art. 13	Transparency -> Definition of transparency
Art. 14	Human oversight -> Human-in/on/over-the-loop
Art. 15	Accuracy and robustness -> Property definitions

Source document

Official standard page: ISO/IEC 22989:2022 — AI concepts and terminology

The standard is also freely available through ISO's open access initiative for foundational AI standards.