Skip to main content

Quality model for AI Systems - ISO/IEC 25059

Building Trustworthy AI: A New Standard Emerges (ISO/IEC 25059)​

Artificial intelligence (AI) is rapidly transforming our world, impacting everything from healthcare diagnoses to self-driving cars. As AI becomes more integrated into our daily lives, ensuring its quality and trustworthiness becomes paramount. To address this critical need, the International Electrotechnical Commission (IEC) and the International Organization for Standardization (ISO) have joined forces to develop a new international standard: ISO/IEC 25059.

This groundbreaking standard establishes a quality framework specifically designed for AI systems. But what exactly does it do?

Leveraging Proven Methods for a New Frontier

ISO/IEC 25059 builds upon the established SQuaRE (Systems and software Quality Requirements and Evaluation) methodology used in other ISO/IEC standards. SQuaRE provides a comprehensive approach to software quality, encompassing aspects like requirements, design, testing, and maintenance. By adapting this proven framework, ISO/IEC 25059 ensures a consistent and reliable approach to evaluating AI systems.

A Common Language for Measuring AI Quality

One of the key strengths of ISO/IEC 25059 is the introduction of a standardized language for specifying, measuring, and evaluating AI system quality. This includes:

  • AI-specific characteristics: AI-specific characteristics, as outlined in the ISO/IEC 25059 standard, focus on the unique qualities that differentiate AI systems from traditional software. Here's a breakdown of some key characteristics:

  • Learning and Adaptation: AI systems can learn from data and experience, improving their performance over time. The standard considers how effectively an AI system learns, adapts to new data, and avoids negative learning biases.

  • Handling Incomplete Data: Real-world data often contains missing information or uncertainties. The standard evaluates how well an AI system deals with incomplete data, makes robust predictions despite these gaps, and avoids generating misleading results.

  • Probabilistic Reasoning: Unlike traditional software with definitive outputs, AI systems often make probabilistic predictions with varying degrees of certainty. The standard assesses how accurately AI systems represent these probabilities and how effectively they communicate them to users.

  • Explainability and Transparency: Understanding how AI systems arrive at decisions is crucial for trust and ethical considerations. The standard evaluates how well an AI system can explain its reasoning process and the factors influencing its decisions.

  • Fairness and Bias Mitigation: AI systems can inherit biases from the data they are trained on. The standard considers how effectively an AI system identifies and mitigates potential biases to ensure fair and non-discriminatory outcomes.

These are just some of the AI-specific characteristics addressed by ISO/IEC 25059. By focusing on these unique aspects, the standard ensures a comprehensive and rigorous evaluation of AI system quality.

  • Building upon existing standards: ISO/IEC 25059 integrates seamlessly with existing software quality standards like ISO/IEC 25010. This ensures a holistic evaluation process that considers both general software quality principles and the specific characteristics of AI systems.

Benefits of Implementing ISO/IEC 25059

By adhering to ISO/IEC 25059, developers can ensure their AI systems are built with the highest quality standards in mind. This translates to several key benefits:

  • Enhanced Trust and Transparency: Following the standard's guidelines promotes rigorous testing and evaluation, fostering trust in the reliability and accuracy of AI systems. Additionally, the standard emphasizes transparency, encouraging clear communication about how AI systems function and make decisions.
  • Reduced Risks: Implementing ISO/IEC 25059 helps mitigate potential risks associated with AI, such as biases or errors. By proactively addressing these concerns, developers can create more robust and responsible AI systems.
  • Ethical Considerations: The standard acknowledges the importance of ethical considerations in AI development. Following its guidelines can help ensure AI systems are developed and used in a responsible and ethical manner.

While ISO/IEC 25059 establishes a strong foundation for AI system evaluation, it's important to note that it focuses on the quality models themselves. Specific guidance on evaluation procedures and metrics is being developed in a separate technical specification (TS).

The emergence of ISO/IEC 25059 marks a significant step towards ensuring the responsible and ethical development of AI. By adopting this standard, developers can build trust, mitigate risks, and pave the way for a future powered by high-quality, reliable AI systems.

Learn More: