Foreword

ISO (the International Organization for Standardization) and IEC (the International Electrotechnical Commission) form the specialized system for worldwide standardization. National bodies that are members of ISO or IEC participate in the development of International Standards through technical committees established by the respective organization to deal with particular fields of technical activity. ISO and IEC technical committees collaborate in fields of mutual interest. Other international organizations, governmental and non-governmental, in liaison with ISO and IEC, also take part in the work.

The procedures used to develop this document and those intended for its further maintenance are described in the ISO/IEC Directives, Part 1. In particular, the different approval criteria needed for the different types of document should be noted. This document was drafted in accordance with the editorial rules of the ISO/IEC Directives, Part 2 (see www.iso.org/directives or www.iec.ch/members_experts/refdocs).

ISO and IEC draw attention to the possibility that the implementation of this document may involve the use of (a) patent(s). ISO and IEC take no position concerning the evidence, validity or applicability of any claimed patent rights in respect thereof. As of the date of publication of this document, ISO and IEC had not received notice of (a) patent(s) which may be required to implement this document. However, implementers are cautioned that this may not represent the latest information, which may be obtained from the patent database available at www.iso.org/patents and https://patents.iec.ch. ISO and IEC shall not be held responsible for identifying any or all such patent rights.

Any trade name used in this document is information given for the convenience of users and does not constitute an endorsement.

For an explanation of the voluntary nature of standards, the meaning of ISO specific terms and expressions related to conformity assessment, as well as information about ISO's adherence to the World Trade Organization (WTO) principles in the Technical Barriers to Trade (TBT) see www.iso.org/iso/foreword.html. In the IEC, see www.iec.ch/understanding-standards.

This document was prepared by Joint Technical Committee ISO/IEC JTC 1, Information technology, Subcommittee SC 42, Artificial intelligence.

Any feedback or questions on this document should be directed to the user’s national standards body. A complete listing of these bodies can be found at www.iso.org/members.html and www.iec.ch/national-committees.

Introduction

Unprecedentedly large and ever-growing deep learning models, large language models, natural language understanding networks and generative AI applications require vast data storage capacities, take weeks to train, are running continuously and require a lot of compute power as well as memory to load the models. And once completed they consume substantial amounts of network connectivity bandwidth in operation. Sixty per cent of IT industry carbon emissions come from the downstream use of products by customers.

The use of power intensive GPUs to run machine learning training (and non-AI uses such as crypto currency mining) has already been cited as contributing to increased carbon dioxide emissions.^[1^] Many machine learning packages have been modified to take advantage of the extensive parallelism available inside the average graphics processing unit. Often this resource intensity is used to exemplify environmental concerns with AI systems.

According to the World Economic Forum and experts in the field, AI has “the potential to accelerate environmental degradation, and is already doing so” ^[1^,2^]. In 2022, the OECD’s Policy Observatory ^[3^] that provided input into basic framework for understanding, measuring and benchmarking domestic AI computing capacity by country and region, did not consider environmental sustainability in its charter ^[4^].

The AI system life cycle does provide opportunities to consider and positively influence the environmental sustainability aspects of the system: for example, using and applying teacher–student models ^[5^] in deep neural networks represents a trade-off between more learning and better inference performance when in production.

Improving in-operation product performance can, conversely, aid sustainability. Publications from the European Union,^[6^,7^] the United States,^[8^-10^] the United Nations ^[11^,12^] and other regional ^[13^] and global think tanks ^[14^] have called for better understanding and disclosure with regards to ICT’s environmental footprint and that of AI systems in particular.

1 Scope

This document provides an overview of the environmental sustainability aspects (e.g. workload, resource and asset utilization, carbon impact, pollution, waste, transportation, location) of AI systems during their life cycle, and related potential metrics.

NOTE 1 This document does not identify opportunities on how AI, AI applications and AI systems can improve environmental, social or economic sustainability outcomes.

NOTE 2 This document can help other projects related to AI system environmental sustainability.

2 Normative references

3 Terms and definitions

4 Symbols and abbreviated terms

5 Purpose

5.1 General

5.2 The London Declaration

5.3 Existing International Standards

5.4 Environmental sustainability in the European Artificial Intelligence Act

5.5 Enacted climate change legislation and market response

5.6 Access to affordable, reliable, sustainable and modern energy

6 Environmental sustainability

6.1 General

6.2 Energy consumption

6.3 Geographic considerations

6.4 Water use

6.5 Cooling

6.6 Carbon footprint

6.7 Consumption patterns and rebound effects

6.8 Waste

7 Perspectives of environmental sustainability of AI systems

7.1 General

7.2 AI system ecosystem

7.3 AI system life cycle

7.4 AI system supply chain

7.5 Other perspectives

8 AI system ecosystem perspective

8.1 General

8.2 AI system

8.3 AI application

8.4 Machine learning

8.5 AI engineering

8.6 Big data and data sources

8.7 Cloud and edge computing

8.8 Resource pools

9 AI system life cycle perspective

9.1 General

9.2 Inception

9.3 Design and development

9.4 Verification and validation

9.5 Deployment

9.6 Operation and monitoring

9.7 Continuous validation

9.8 Re-evaluation

9.9 Retirement

10 AI system supply chain approach to determine environmental sustainability aspects

10.1 General

10.2 Elements

10.3 Mines

10.4 Smelters and refiners

10.5 Component manufacturers

10.6 Assemblers

10.7 Distributors

10.8 Domestic infrastructure

10.9 Internet infrastructure

10.10 AI Training

10.11 Data preparation, labelling and dataset size

10.12 Device abandonment

10.13 Collection of abandoned devices

10.14 Shipping of abandoned devices

10.15 Recovering of abandoned devices

10.16 Disposing of abandoned devices

11 Measuring the environmental sustainability aspects of AI systems

11.1 General

11.2 AI system ecosystem measurements

11.3 AI system life cycle measurements

11.4 AI system supply chain measurements

11.5 Other measurements

12 Approaches to reducing AI systems’ negative environmental impacts

12.1 General

12.2 Ecosystem approaches

12.3 Life cycle approaches

12.4 Supply chain approaches

Bibliography