ISO/IEC 22989:2022 - Artificial intelligence concepts and terminology

Summary

ISO/IEC 22989:2022 is the global standard that brings order to the chaos of AI terminology. As artificial intelligence exploded across industries, organizations found themselves speaking different languages—what one company called "machine learning," another might call "automated decision-making." This standard cuts through the confusion by establishing official definitions for over 100 AI-related terms, from basic concepts like "artificial intelligence" itself to nuanced distinctions between "supervised learning" and "reinforcement learning." Think of it as the authoritative dictionary for AI professionals worldwide, ensuring everyone from Silicon Valley startups to European regulators is using the same vocabulary.

The terminology goldmine: What's actually defined

This isn't just a collection of buzzwords—it's a carefully structured taxonomy of AI concepts organized into logical categories:

Core AI concepts: Artificial intelligence, machine learning, deep learning, neural networks, and the fundamental building blocks that underpin all AI systems.

Learning paradigms: Clear distinctions between supervised, unsupervised, semi-supervised, and reinforcement learning, including hybrid approaches that combine multiple methods.

Data and processing terms: Everything from "training data" and "test data" to "feature engineering" and "data preprocessing"—the vocabulary of AI's data pipeline.

Model types and architectures: Specific definitions for convolutional neural networks, recurrent neural networks, transformers, and other architectural patterns.

Performance and evaluation: Standardized language for accuracy, precision, recall, bias, fairness, and other metrics that determine AI system quality.

AI system lifecycle: Terms covering development, deployment, monitoring, and maintenance phases of AI systems.

Why this standard is a game-changer

The real power of ISO/IEC 22989 lies in its potential to eliminate miscommunication across the AI ecosystem. When procurement teams, technical teams, legal departments, and vendors all use the same definitions, projects run smoother and compliance becomes clearer.

For organizations implementing AI governance frameworks like ISO/IEC 23053 or responding to regulations like the EU AI Act, having standardized terminology eliminates ambiguity in policies and procedures. Instead of spending meeting time clarifying what someone means by "AI system," teams can focus on the actual work.

The standard also bridges cultural and linguistic gaps in global organizations, providing English definitions that can be consistently translated and applied across different markets and regulatory environments.

Who this resource is for

AI governance teams building internal policies and procedures who need consistent language across all documentation and training materials.

Technical writers and documentation specialists creating AI-related content who want to ensure their terminology aligns with international standards.

Legal and compliance professionals drafting AI-related contracts, terms of service, or regulatory compliance documents where precise definitions matter.

Procurement and vendor management teams writing RFPs or evaluating AI solutions who need clear, standardized criteria and terminology.

Consultants and auditors working across multiple organizations who need a common vocabulary to communicate effectively with diverse clients.

Standards developers and regulators creating AI-related standards, regulations, or guidelines who want to build on established, internationally recognized definitions.

Getting your hands on the standard

ISO/IEC 22989 is available for purchase directly from ISO or national standards bodies like ANSI in the US or BSI in the UK. Pricing typically ranges from $150-300 depending on your location and whether you need individual or site licenses.

Many organizations find it worthwhile to purchase institutional access, especially if multiple teams will reference the standard. Some technical libraries at universities or large corporations may already have access through broader ISO subscriptions.

The standard is about 30 pages long and designed for reference use rather than cover-to-cover reading. Most teams find it valuable to extract key definitions relevant to their specific use cases and incorporate them into internal glossaries or training materials.

Watch out for: Common implementation pitfalls

Don't assume everyone in your organization needs to memorize every definition—focus on the terms most relevant to your specific AI applications and governance needs.

Be careful about version control when incorporating these definitions into your own documentation. As AI evolves, future versions of the standard may refine or expand definitions.

Remember that standardized terminology is most valuable when consistently applied. Half the organization using ISO definitions while the other half uses informal terminology defeats the purpose.

The standard provides definitions but not implementation guidance—you'll need to pair it with other resources for practical AI governance and development practices.