Standards Development

Scope

This document defines a set of processes and associated terminology for describing the life cycle of AI systems. This document forms the foundation of a detailed AI system life cycle specification. It is based on ISO/IEC/IEEE 15288 and ISO/IEC/IEEE 12207 with substitutes for and additions of AI specific processes, whose foundation is based on ISO/IEC 22989 and ISO/IEC 23053. Selected sets of these processes can be applied throughout the life cycle for managing and performing the stages of an AI system's life cycle.

This document provides processes that support the definition, control and improvement of the AI system life cycle processes used within an organization or a project. Organizations and projects can use these processes when developing or acquiring AI systems. When an element of an AI system is traditional software or a traditional system, the software life cycle processes in ISO/IEC/IEEE 12207 and the system life cycle processes in ISO/IEC/IEEE 15288 may be used to implement that element.

Purpose

Artificial Intelligence (AI) systems such as machine translation, speech recognition, image classification and information retrieval have achieved remarkable successes. To build an AI system, it is necessary to consider both the AI-specific life cycle processes as well as the life cycle processes for a traditional software system.

The life cycle processes for a traditional software system can have a significant effect on any of Q (Quality), C (Cost), and D (Delivery), which is called QCD. The AI-specific life cycle processes of an AI system can also affect QCD of the AI system. ISO/IEC/IEEE 15288 (System life cycle processes) and ISO/IEC/IEEE 12207 (Software life cycle processes) describe the life cycle of software and system.

In the case of an AI system that uses machine learning, part of the system is defined by inductive learning using training data instead of being implemented directly in code as per the system’s specifications. Furthermore, when the data being processed by the AI system changes over time, it may be necessary to update the AI system by retraining.

By providing a life cycle standard based on these characteristics of AI systems, it becomes easier to secure social acceptance of AI systems by promoting mutual understanding between developers and users of AI systems.

Furthermore, it is easy to determine whether an AI system development project is completed or not completed. Troubles caused by inconsistency between users and providers in understanding the functions and performance of AI and AI systems during development, delivery, and operation can also be prevented.

Regarding the processes of AI system life cycle, there are three considerable types of processes in this document:

1. the definition of processes of AI system life cycle corresponds to the definition in traditional system (ISO/IEC/IEEE 15288).

2. the processes include AI-specific characteristic, but the definition of processes should be modified from the definition of processes in ISO/IEC/IEEE 15288.

3. the processes include AI-specific characteristic but cannot adopt any existing processes in ISO/IEC/IEEE 15288. 

To identify the type of processes, analysing use cases of AI system is an effective approach. SC 42/WG 4 collected more than 130 use cases from various application domains. We consider that developing this document in SC42/WG 4 is suitable. WG4's use cases can be used to identify the type of processes in AI system life cycle and the contributors of use cases can provide additional information for analysis.

Consider the following: Is there a verified market need for the proposal? What problem does this standard solve? What value will the document bring to end-users? See Annex C of the ISO/IEC Directives part 1 for more information. See the following guidance on justification statements on ISO Connect: https://connect.iso.org/pages/viewpage.action?pageId=27590861