• Assessing the environmental impact of AI systems is complex, particularly given the lack of transparency offered by the leading model providers.
• Orange is building the “AI Carbon Value Navigator” tool to guide decision-making around whether or not to incorporate AI into operations.
The boom in popularity of generative AI, which companies and the general public alike have been using on a large scale since late 2022 and the launch of ChatGPT, Gemini, Claude Sonnet, etc., has brought the subject of AI’s environmental impact (back) to the fore. Artificial Intelligence systems have the potential to reduce the carbon footprint of many activities, while simultaneously generating energy and ecological impacts of their own.
Implementing a carbon management tool to guide and simplify decision-making regarding the incorporation of AI into our products and services
Key assessment considerations
Assessing this impact is a business matter. Nathalie Charbonniaud, responsible AI research lead and expert at Orange: “Traditionally, classic AI use cases are some of the most common employed within companies. Their impact is not excessive, and is relatively simple to calculate. Conversely, the impact of generative AI is proving to be much trickier to determine, and even more so to get a grip on, owing to the lack of transparency from the major model providers. Without precise data, we are restricted to making reverse-engineered estimations and projections concerning the training and inference phases”. Orange is conducting independent research and expert analysis on calculation methodologies and is partnering with multi-stakeholder initiatives to drive change. In this vein, the Group has contributed to the development of AFNOR’s frugal AI framework and is working with Boavizta on the EcoLogits tool, which provides measures to assess the impact of an AI system based on model type and use. Its commitment also extends to the field of standardisation, and more generally to platforms urging greater transparency. “In addition, we are working internally on the governance of frugal AI by implementing a dedicated management tool to guide decision-making around the incorporation of AI into our products and services”.
Assessing the carbon/finance delta of AI systems
This tool, named the AI Carbon Value Navigator, has been designed to support decision-makers and operational stakeholders in determining whether a project incorporating AI aligns with the financial and carbon reduction objectives of each BU or division of the company. “The objective is to compare the economic potential for value creation (revenue generation and cost savings) of implementing the AI system against its impact on the overall carbon footprint of the solution into which it must be incorporated”, explained Gilles Deflandre, data analyst at Orange’s Data Studio. “In other words, it enables us to answer the following question: Does the evolution of a service to include AI contribute to creating financial value in a manner compatible with carbon emission reduction ambitions? To do so, we use a carbon intensity reference framework developed by our sustainable performance teams using data drawn from multiple internal and external sources. Those teams assess and correlate the financial gains and the CO2 generated. By way of example, an AI system may reduce the time that machines are running, thereby saving resources, but at the same time could necessitate additional maintenance or development operations. The tool quantifies and balances these factors, and places them on a two-axis compass, one axis corresponding to the financial gain and the other to the carbon emissions generated or avoided. Teams can use the information provided by the tool to gauge whether a project to incorporate AI is a step in the right direction”.
A compass that complements assessment methodologies
This approach represents a first in the telecoms sector, but does not supplant the Group’s efforts relating to assessment methods in the strict sense. On the contrary, the two approaches work hand in hand. The AI Carbon Value Navigator provides an initial understanding of the carbon emission implications of a project. In the most complex cases, it will be supplemented by an assessment of the carbon footprint for the physical operation of the service.
The tool, which is currently in the testing phase, is expected to be rolled out Group-wide in an automated form so that it is accessible and can be used quickly and easily. It should also be expanded beyond carbon to cover other aspects of sustainability. At the same time, the Group is undertaking another similar project relating to a carbon management tool: the Carbon Estimator, for business customers.
Read more :
Nathalie Charbonniaud
Gilles Deflandre
