ISO-IEC 25059:2023

Standardizing AI System Quality

ISO/IEC 25059 introduces a standardized framework for defining, assessing, and measuring the quality of AI systems. It defines a practical quality model specifically for AI systems, building on the SQuaRE series of standards. It introduces clear and consistent terms for describing, measuring, and assessing the quality of AI systems. Additionally, the standard outlines quality traits that can be used to check if quality requirements are thorough and complete.

Key AI-Specific Characteristics

The standard emphasizes the unique qualities that set AI systems apart from traditional software. Below are some of the critical characteristics it evaluates:

• Learning and Adaptation: AI systems have the ability to learn from data and experiences, enhancing their performance over time. The standard examines how effectively a system learns, adapts to new information, and avoids undesirable learning biases.
• Handling Incomplete Data: In real-world scenarios, data often contains gaps or uncertainties. The standard assesses how well AI systems manage incomplete data, deliver reliable predictions, and minimize the risk of misleading results.
• Probabilistic Reasoning: AI systems often produce probabilistic outputs rather than definitive answers. The standard evaluates how accurately these systems represent probabilities and communicate their level of certainty to users.
• Explainability and Transparency: For AI systems to be trusted and ethically aligned, users must understand how they arrive at decisions. The standard reviews the clarity of an AI system’s reasoning and its ability to explain decision-making processes.
• Fairness and Bias Mitigation: Bias in AI systems can lead to unfair or discriminatory outcomes. ISO/IEC 25059 focuses on how effectively a system identifies and mitigates biases, ensuring equitable results.
• These characteristics highlight the standard’s commitment to addressing the distinct challenges posed by AI technologies, enabling a comprehensive quality assessment.

Integration with Existing Standards

ISO/IEC 25059 builds on established software quality standards like ISO/IEC 25010, creating a cohesive framework that combines general software quality principles with AI-specific requirements. This integration ensures that all aspects of software performance, including those unique to AI, are rigorously evaluated.