Overview
ISO/IEC 5338:2023 - Information technology - Artificial intelligence - AI system life cycle processes - defines a structured set of life cycle processes and concepts for AI systems based on machine learning (ML) and heuristic approaches. Built on established system and software life cycle standards (ISO/IEC/IEEE 15288 and 12207) and aligned with AI-specific frameworks (ISO/IEC 22989 and ISO/IEC 23053), this standard describes how to define, control, manage, execute and improve AI systems throughout their life cycle. It is intended for use by organizations and projects developing or acquiring AI systems, and it clarifies how traditional software life cycle processes apply when AI elements coexist with conventional software.
Key Topics and Requirements
- Life cycle process taxonomy: Agreement processes, organizational/project-enabling processes, technical management processes, and technical processes tailored for AI systems.
- AI-specific processes: Examples include knowledge acquisition, AI data engineering, continuous validation, and retraining-related activities.
- Modified processes: Adaptations of ISO/IEC/IEEE 15288 and 12207 to address AI characteristics (e.g., model drift, training-data lifecycle).
- Technical controls and governance: Risk management, quality assurance, configuration & information management, and measurement processes adapted for AI outputs and data.
- Verification & Validation (V&V): Processes for verification, validation, transition, operation, maintenance, and disposal, including ongoing monitoring for measurable potential decay and retraining needs.
- Conformance and integration: Guidance on applying traditional software/system life cycle processes where AI system elements are conventional software components.
Practical Applications and Users
Who benefits:
- AI system architects and machine learning engineers implementing robust ML lifecycle management.
- Systems engineers and software developers integrating AI elements with traditional systems.
- Project managers, procurement teams and suppliers using acquisition/supply processes for AI procurement.
- MLOps, DevOps and data engineering teams responsible for deployment, monitoring, and retraining pipelines.
- Quality, compliance and risk management teams needing documented processes for safety-critical AI (e.g., healthcare, autonomous vehicles).
How it’s used:
- Designing AI development workflows that include data engineering, knowledge acquisition, and continuous validation.
- Establishing organizational policies for AI life cycle model management, infrastructure, and human resources.
- Managing model drift, measurement and performance decay through planned maintenance and retraining strategies.
- Ensuring traceability, configuration control, and auditability for regulatory and governance needs.
Related Standards
- ISO/IEC/IEEE 15288:2023 - System life cycle processes
- ISO/IEC/IEEE 12207:2017 - Software life cycle processes
- ISO/IEC 22989:2022 - AI concepts and terminology
- ISO/IEC 23053 - Framework for AI systems using ML
- ISO/IEC TR 5469 (safety-related considerations)
Keywords: ISO/IEC 5338:2023, AI system life cycle, AI life cycle processes, machine learning lifecycle, AI governance, AI data engineering, model validation, continuous validation, ISO AI standards.