ISO/IEC 22989:2022 - AI Terminology and Concepts Standard

Summary

ISO/IEC 22989:2022 is the foundational terminology standard that brings order to the chaotic world of AI language. If you've ever been in a meeting where "machine learning," "artificial intelligence," and "cognitive computing" get tossed around interchangeably, this standard is your antidote. It provides precise, globally recognized definitions for 120+ AI terms and establishes a common vocabulary that spans from basic concepts like "training data" to complex topics like "explainability" and "AI bias." This isn't just an academic exercise—it's the linguistic foundation that enables clear communication across disciplines, industries, and borders in AI governance and development.

The terminology crisis this solves

Before ISO/IEC 22989, the AI field suffered from what linguists call "semantic chaos." The same concept would have different names across industries, or worse, the same term would mean completely different things to different communities. "Deep learning" meant one thing to computer scientists, something slightly different to data scientists, and was often misunderstood entirely by business leaders and policymakers.

This standard tackles that head-on by providing authoritative definitions that are vendor-neutral, technology-agnostic, and designed to remain stable as AI evolves. Each definition includes context, relationships to other terms, and examples that clarify usage across different domains.

What you'll find inside

The standard organizes AI terminology into logical clusters that mirror how practitioners actually think about AI:

Core AI concepts: Fundamental terms like artificial intelligence, machine learning, neural networks, and algorithms—with precise boundaries between each concept.

Data and learning: Comprehensive coverage of training data, validation sets, supervised learning, unsupervised learning, and reinforcement learning, including the subtle but crucial distinctions between them.

AI system components: Clear definitions for models, datasets, features, parameters, and hyperparameters that eliminate confusion between technical and business teams.

Performance and evaluation: Standardized language for accuracy, precision, recall, bias, fairness, and other metrics that are essential for AI governance and compliance.

Trustworthy AI: Definitions for explainability, interpretability, transparency, robustness, and accountability that align with emerging regulatory requirements globally.

Who this resource is for

AI governance teams who need precise language for policies, risk assessments, and compliance documentation that will be understood consistently across their organization and by external auditors.

Technical standards writers developing AI-specific standards who need a stable foundation of terminology that won't create conflicts with other standards or regulations.

Legal and compliance professionals working on AI contracts, privacy assessments, or regulatory submissions who must use terminology that has clear, defensible meanings.

Cross-functional teams including business leaders, engineers, data scientists, and ethicists who need a common vocabulary to enable productive collaboration on AI projects.

International organizations and multinational companies that need terminology translations and concepts that work across different legal and cultural contexts.

How this integrates with your AI governance stack

This terminology standard isn't meant to sit on a shelf—it's designed to be the linguistic backbone of your entire AI governance program. When you're implementing ISO/IEC 23053 (AI use cases and applications) or working toward ISO/IEC 23894 (AI risk management), you'll use ISO/IEC 22989 definitions to ensure consistency.

The standard also aligns with regulatory frameworks like the EU AI Act, which references similar concepts but doesn't always define them precisely. Having ISO/IEC 22989 as your baseline ensures your governance documentation uses globally recognized language that regulators and auditors will understand.

Most importantly, this standard helps you avoid the governance nightmare of having different definitions of critical terms like "bias," "transparency," or "high-risk AI" across different policies, procedures, and systems within your organization.

Quick implementation wins

Start by adopting the standard's definitions in your AI policy documents, replacing any homegrown terminology with ISO/IEC 22989 language. Create a glossary based on the standard for your organization, focusing on the 20-30 terms most relevant to your AI use cases.

Use the standard's conceptual relationships to structure your AI governance framework—the way it organizes concepts often reveals gaps in typical governance approaches. Train your teams on key definitions, especially those that differ from common usage in your industry or region.