UNE Specification — Environmental Impact Assessment of Artificial Intelligence Systems
UNE Specification — Environmental Impact Assessment of Artificial Intelligence Systems
This UNE Specification has been developed to address the assessment of the environmental impact of Artificial Intelligence (AI) systems, focusing on the energy and water consumption of models and algorithms, as well as the carbon emissions produced throughout their life cycle.
The environmental assessment of other natural resources is outside the scope of this UNE Specification, including the hardware on which these systems run and the materials used in their manufacturing, as well as the environmental assessment of the facilities where they operate—whether simple offices or dedicated data centers—since other recommendations and standards specifically address these aspects.
Nevertheless, this specification should be considered complementary to those other recommendations and standards in order to obtain the most realistic possible approach to the environmental impact of the AI systems under consideration and of the solutions, products, and services in which they are used. This approach is essential to optimize efficiency and minimize environmental impact, while ensuring adequate performance of such AI solutions.
In parallel with this work, other specifications and normative documents are being developed that address the measurement of AI’s environmental impact. The most relevant among these are:
– ISO/IEC TR 20226 Technical Report, Information technology — Artificial intelligence — Environmental sustainability aspects of AI systems.
– AFNOR SPEC 2314, June 2024, Référentiel général pour l'IA frugale. Mesurer et réduire l'impact environnemental de l'IA.
– CEN/CLC Technical Report TR 18145:2025, Environmentally Sustainable Artificial Intelligence, by CEN/CENELEC.
– Technical Specification prCEN/TS, Sustainable Artificial Intelligence. Guidelines and metrics for the environmental impact of artificial intelligence systems and services, by CEN/CENELEC.
The objective of this UNE Specification is to provide a detailed guide for measuring energy consumption, carbon footprint, water consumption, and performance of Artificial Intelligence systems, establishing an appropriate framework for the measurement and evaluation of these parameters in the models and algorithms that constitute such systems during the different phases of their life cycle.
It defines the methods and guidelines for measuring energy consumption, carbon footprint, water consumption, and performance both at the hardware level and through the libraries and frameworks used in their development and operation. It also considers the specific characteristics of data centers that typically host their execution, and in particular, the specific aspects of cloud-based operation.
The aim is to provide a consistent and reliable approach for the measurement and optimization of these parameters in all aspects related to AI operations, facilitating comparison and continuous improvement in balancing operational performance and environmental efficiency.
This specification therefore considers the environmental life-cycle assessment of AI systems based on machine learning and deep learning techniques and models, as these are the ones with the greatest environmental impact.
This UNE Specification provides a detailed guide for measuring the energy consumption, carbon footprint, water consumption, and performance of Artificial Intelligence systems. It is intended to be applied to systems directly involved in AI tasks, and more specifically to machine learning and deep learning models (including Generative AI and Large Language Models — LLMs, neural networks, etc.), as these have the most significant environmental impact, as well as to the management of energy resources within these environments.
It is not intended to be applied to components or systems that are not directly engaged in AI operations or that do not comply with the specified measurement guidelines.
Reference Standards
The documents listed below, in whole or in part, are indispensable reference standards for the application of this document.
For dated references, only the edition cited applies.
For undated references, the latest edition (including any amendments) applies.
-
UNE-EN 50600-4-2:2016 + A1:2019, Information technology — Data centre facilities and infrastructures — Part 4-2: Energy usage effectiveness.
-
UNE-EN 50600-4-3:2016 + A1:2019, Information technology — Data centre facilities and infrastructures — Part 4-3: Renewable energy factor.
-
UNE-EN 50600-4-8:2022, Information technology — Data centre facilities and infrastructures — Part 4-8: Carbon usage effectiveness.
-
UNE-EN 50600-4-9:2022, Information technology — Data centre facilities and infrastructures — Part 4-9: Water usage effectiveness.
.jpeg)
Comments
Post a Comment