Overview
EN ISO/IEC 23053:2023 specifies a standardized framework for describing Artificial Intelligence (AI) systems that utilize Machine Learning (ML) technologies. Developed by CEN and based on ISO/IEC 23053:2022, this standard provides a structured approach to the components and their functions within the AI system ecosystem. The framework is applicable across a wide range of organizations, including public and private enterprises, government agencies, and non-profits, regardless of their size or sector.
With the rapid expansion of AI and ML in multiple industries, a common framework helps organizations adopt, implement, and govern AI systems in a consistent, transparent, and interoperable manner. EN ISO/IEC 23053:2023 establishes common terminology and concepts, supporting better communication, integration, and compliance in AI system development and deployment.
Key Topics
- Core Components of ML Systems: Defines the essential elements of an AI system based on machine learning, including models, tools, data, and their interdependencies.
- Machine Learning Tasks:
- Classification: Assigning inputs to defined categories (e.g., spam detection, species identification).
- Regression: Predicting continuous variables (e.g., price forecasting).
- Clustering: Grouping unlabeled data by similarity (e.g., customer segmentation).
- Anomaly Detection: Identifying outliers (e.g., fraud detection).
- Dimensionality Reduction: Simplifying data while retaining meaningful attributes.
- Structured Prediction & Other Tasks: Handling complex outputs such as language parsing or image segmentation.
- ML System Lifecycle:
- Model Development: Training, evaluation, and updating models based on new or changing data (concept drift/data drift).
- Data Management: Distinguishing between training, validation, and production datasets.
- Pipeline Stages: Data acquisition, preparation, modeling, deployment, validation, and continuous operation.
- ML Approaches: Outlines different learning paradigms, such as supervised, unsupervised, semi-supervised, self-supervised, reinforcement, and transfer learning, and their typical applications.
- Evaluation & Optimization: Covers the techniques and metrics for assessing ML model performance and ensuring ongoing reliability.
Applications
Organizations can leverage EN ISO/IEC 23053:2023 to:
- Design and Document AI Systems: Employ standardized terminology and a common framework when specifying AI projects, ensuring clarity and alignment among stakeholders.
- Ensure Interoperability: Support integration with other systems and compliance with international AI standards.
- Mitigate Risks: Address system reliability, data quality, model drift, and bias concerns via systematic processes recommended by the standard.
- Guide Procurement and Compliance: Use the framework as a reference for assessing vendor solutions and ensuring third-party AI systems meet recognized norms.
- Train Staff and Stakeholders: Benefit from the clear descriptions of roles, system components, and workflows when onboarding teams to AI projects.
- Support Innovation: Foster the development and deployment of flexible, modular, and robust AI/ML systems suitable for a wide variety of use cases across domains such as finance, healthcare, manufacturing, and government.
Related Standards
Organizations implementing EN ISO/IEC 23053:2023 should also consider the following relevant standards:
- ISO/IEC 22989: Information technology - Artificial intelligence - Concepts and terminology (key reference for terminology).
- ISO/IEC 23894: AI Risk Management (practices for managing risks in AI deployments).
- ISO/IEC 20546: Big Data - Overview and vocabulary (for data-centric approaches in AI/ML).
- ISO/IEC 38507: Governance implications of the use of AI.
- ISO/IEC TR 24028: Overview of trustworthiness in AI.
By following EN ISO/IEC 23053:2023, organizations can strengthen their AI strategy, drive compliant and effective ML system implementation, and keep pace with evolving AI technologies and regulations.