top of page
  • Writer's pictureMichael

International Standards (ISO/IEC 23053) for Artificial Intelligence: Part 3/4

Worldwide standardizations for Artificial Intelligence (AI)

  • ISO (International Organization for Standardization)

  • IEC (International Electrotechnical Commission)


Focus on ISO/IEC 22989:2023: Framework for Artificial Intelligence (AI) Systems Using Machine Learning (ML)





Framework


The ISO/IEC 23053:2023 standards establishes a framework for describing a generic AI system using ML technology. It describes the system components and their functions in the AI ecosystem. The framework is applicable to all types and sizes of organizations, including public and private companies, government entities, and not-for-profit organizations, that are implementing or using AI systems.

The set of standards aims to provide a basis for the clear explanation of AI systems and various considerations that apply to their engineering and use. It is intended for a wide audience, including experts and non-practitioners. However, some clauses include more in-depth technical descriptions.


ISO/IEC 23053:2023 is an international standard that provides a framework for describing artificial intelligence (AI) systems that use machine learning (ML) technology. The set of standards aims to establish a common terminology and set of concepts for AI systems that use ML, providing a basis for clear explanations of the systems and various considerations that apply to their engineering and use. The standard is applicable to all types and sizes of organizations, including public and private companies, government entities, and not-for-profit organizations that are implementing or using AI systems.


Ecosystem

 

The standards begins by providing an overview of the AI ecosystem, which is presented in terms of its functional layers. ML is identified as a significant component of this ecosystem, and the set of standards provides a breakdown of the ML system into its components of model, software tools and techniques, and data. The standard describes the different components of the ML system in detail, including the different ML approaches and their dependency on training data, and the processes involved in developing, deploying, and operating an ML model.

 

The standards emphasizes the importance of data in the ML model life cycle, as both training and evaluation rely on it. The acquisition and preparation of data require special care to ensure that the acquired data is appropriate to the business purpose of the model. The standard provides guidance on the type of data needed, which depends on the problem being solved, and the sources from which data can be acquired. The standards also highlights the importance of validation and test data to determine the accuracy and other performance metrics of ML models.

 

The standard discusses the fundamental challenges with ML, including statistical analysis, algorithm design and optimization, and understanding the potential and possibilities of ML. It emphasizes the need for risk management and governance, security and privacy, accountability, transparency, and explainability, safety, resilience, robustness, and fairness across the entire ML pipeline.

 

The set of standards categorizes tools used in ML model creation into data preparation, ML algorithms, optimization methods, and evaluation metrics. It highlights the significance of these tools in different stages of the ML model development lifecycle, from data preprocessing to model evaluation. The standard emphasizes the importance of high-performance compute workloads due to computational demands and the use of large training datasets in the creation of ML models. It also highlights the importance of assessing the performance of ML models through the generation of evaluation metrics.

 

The standard provides guidance on the different categories of ML algorithms, including supervised, unsupervised, and reinforcement learning. It emphasizes the importance of hyperparameters, which are characteristics of an ML algorithm that affect its learning process, and their role in estimating model parameters. The set of standards also provides guidance on the different types of ML models, including classification and regression models, and the importance of generalization, which is the ability of a trained model to make correct predictions on previously unseen input data.

 

The standards emphasizes the importance of understanding the potential and possibilities of ML, as it relies on data and can replicate, amplify, and expedite existing faults and inequities in many cases. It highlights the need for accountability, transparency, and explainability in the development and use of AI systems that use ML technology. The standard provides guidance on the use of ISO/IEC 19944-1 to describe data flows and develop data use statements for ML processes, which can help in developing the building blocks and fundamental elements needed for AI accountability and transparency.





Key Topics


The key topics covered in the ISO/IEC 23053:2023 standards include:

 

1. Framework for AI Systems Using ML: The standards establishes a framework for describing a generic AI system using ML technology, covering the system components and their functions in the AI ecosystem.

 

2. Model Development and Use: It includes terms and definitions related to machine learning, such as classification model, regression model, and generalization.

 

3. ML Processes and Lifecycle: The set of standards illustrates specific ML processes involved in developing, verifying, deploying, and operating an ML model, and how they relate to the AI system life cycle stages. It addresses aspects such as risk management, governance, security, privacy, accountability, transparency, explainability, safety, resilience, robustness, and fairness.

 

4. Data Acquisition: It discusses the initial stage for the development, deployment, and operation of an ML model, emphasizing the acquisition and preparation of data, including training, validation, and test data. It also highlights the types of data needed depending on the problem being solved and the sources of training data.

 

5. Machine Learning System Components: The standards describes the elements of an ML system, delineating the roles and their ML-specific functions that can be implemented by different entities. It also provides an overview of the ML system and its components.

 

6. Data Flows and Data Use Statements: It provides guidance on describing data flows and developing data use statements for ML processes, utilizing the taxonomy and data use statement format described in ISO/IEC 19944-1.

 

These topics collectively provide a comprehensive framework for understanding and implementing AI systems using ML technology.




 

Related Frameworks

 

The ISO/IEC 23053:2023 set of standards references the following frameworks:

 

1. ISO/IEC 19944-1: This standard is utilized to describe data flows and develop data use statements for ML processes. It provides the taxonomy and data use statement format for this purpose.

 

2. ISO/IEC 22989: This standard is referenced for terms and definitions related to artificial intelligence. It is part of the terminology databases maintained by ISO and IEC for use in standardization [T2, T6].

 

These frameworks provide essential guidance and terminology for the development and understanding of AI systems using machine learning.

 

 

The ISO/IEC 23053:2023 set of standards covers several key terms related to artificial intelligence and machine learning. Some of these terms include:

 

1. Classification Model: A machine learning model whose expected output for a given input is one or more classes.

 

2. Regression Model: A machine learning model whose expected output for a given input is a continuous variable.

 

3. Generalization: The ability of a trained model to make correct predictions on previously unseen input data. It is related to the prediction accuracies using previously unseen input data.

 

4. AI Ecosystem: The standards presents an AI ecosystem in terms of its functional layers, with machine learning being a significant component of this ecosystem.

 

5. Data Acquisition: The process of acquiring data, which is a key component in the ML model life cycle, including training, validation, and test data.

 

These terms are fundamental to understanding the concepts and processes outlined in the set of standards and are essential for practitioners and stakeholders involved in AI systems using machine learning.

 

Artificial Intelligence Tools

 

In the context of ISO/IEC 23053:2023, the standards discusses the use of tools in the creation and assessment of machine learning (ML) models. It categorizes tools into various types and emphasizes their significance in the ML model development process.

 

The standards highlights the following aspects related to tools:

 

1. Categorization of Tools: Tools used in ML model creation are categorized into data preparation, ML algorithms, optimization methods, and evaluation metrics. These tools play a crucial role in different stages of the ML model development lifecycle, from data pre-processing to model evaluation.

 

2. Performance Considerations: The creation of ML models often requires high-performance compute workloads due to computational demands and the use of large training datasets. The performance of compute and storage resources can significantly impact the speed at which ML models are developed and trained.

 

3. Model Assessment: Tools are utilized to assess the performance of ML models through the generation of evaluation metrics. These metrics provide insights into the effectiveness and accuracy of the trained models.

 

Overall, the set of standards underscores the importance of tools in the development, evaluation, and performance assessment of ML models, acknowledging their role in enabling effective machine learning processes.


Closing thoughts


ISO/IEC 23053:2023 provides a comprehensive framework for describing AI systems that use ML technology. The standard emphasizes the importance of data, tools, and performance considerations in the creation and assessment of ML models. It provides guidance on the different categories of ML algorithms, types of ML models, and the importance of generalization. The standards also highlights the need for accountability, transparency, and explainability in the development and use of AI systems that use ML technology. Overall, the standard provides essential guidance and terminology for the development and understanding of AI systems using machine learning.


24 views

Comments

Couldn’t Load Comments
It looks like there was a technical problem. Try reconnecting or refreshing the page.

BOTS of LONDON town

bottom of page