Toward Responsible AI Use: Considerations for Sustainability Impact Assessment

🔬 Research Summary by Eva Thelisson, General Secretary, AI Transparency Institute in Switzerland.

[Original paper by Eva Thelisson, Grzegorz P. Mika, Quentin Schneiter, Kirtan Padh, Himanshu Verma]

Overview: This paper discusses the development of the ESG Digital and Green Index (DGI). This tool is designed to help stakeholders assess and quantify AI technologies’ environmental and societal impacts. DGI offers a dashboard for assessing an organization’s performance in achieving sustainability targets. This includes monitoring the efficiency and sustainable use of limited natural resources related to AI technologies (water, electricity, etc). It also addresses the societal and governance challenges related to sustainable AI. The DGI is part of the Care AI tools developed by the AI Transparency Institute. It aims to incentivize companies to align their pathway with the Sustainable Development Goals (SDGs).

Introduction

What is the environmental impact of AI?

This question has become increasingly relevant as AI, especially generative AI, becomes widely available to users worldwide. A recent study found that using AI to create images requires as much energy as charging a smartphone and comes at the cost of significant carbon emissions ¹. As AI systems have become more versatile and ubiquitous, AI models have become massive, requiring immense amounts of data and other computational resources to train and maintain. They have also pervaded many aspects of our lives, from recommending movies, shopping, and dating, to banking, credit scoring, fraud detection, and traffic management.

While the use of these systems is based on the promise of increased reliability, efficiency, and fairness, it is now well known that things often go wrong², with countless examples of algorithms turning out to be racist or sexist. Moreover, training and deploying AI consumes as much energy³ as a small country, and the rate is only increasing over time.

Moreover, in February 2022, the European Commission proposed a Directive on corporate sustainability⁴ due diligence. By May 2024, the Council of the European Union approved the political agreement, finalizing its adoption. This Directive aims to promote sustainable and responsible corporate behaviour across global value chains and create new obligations for companies. The new regulations require companies to identify and address adverse human rights and environmental impacts within and outside Europe, ensuring comprehensive accountability in their operations. The DGI could help companies in this context.

The Digital and Green Index (DGI) aims to increase transparency and awareness of the environmental impacts of developing and deploying AI on energy, carbon emissions, and the use of natural resources and rare earth minerals. The outcome can be seen as a measure of how well a business adheres to both social and environmental sustainability best practices. The DGI is based on Kate Raworth’s doughnut model⁵, which combines the concept of planetary boundaries with the complementary concept of social boundaries. This paper refers to this compass.

The conclusion is that by aligning business interests with global sustainability objectives, we can go from a limitless growth of digital technologies to a balanced prosperity that takes into account the limits of the Earth’s natural resources and human rights. The DGI provides an exploratory tool to incentivize this alignment.

Embracing Sustainable AI: The Digital and Green Index

**Figure 1:** Conceptual framework for the ESG Digital and Green Index

The Digital and Green Index (DGI) represents an innovative approach to developing AI services and products that are both environmentally sustainable and socially beneficial. This methodology provides a comparative advantage for AI technologies that are low-carbon, climate-resilient, and contribute to pollution mitigation, ecosystem health, and productivity.

In pursuit of sustainable artificial intelligence, we have crafted a comprehensive set of indicators to assess the DGI. The development of these indicators has provided valuable insights into achieving the goals of sustainable AI. Central to the DGI are four interlinked concepts: a low-carbon economy, a healthy ecosystem, a resilient society, and inclusive growth.

These pillars guide the creation of AI solutions that not only advance technology but also support environmental and social well-being.

Design Rationale

**Figure 2:** The stepwise approach to developing the ESG Digital and Green Index.

The Digital and Green Index (DGI) is designed to measure, track, and communicate the performance of digital and green AI technologies. This composite index aims to raise awareness and sustain momentum for sustainable AI initiatives across public and private sectors. Rooted in a robust sustainability framework, the DGI highlights the achievements of the Sustainable Development Goals (SDGs) related to digital and green AI.

By improving current knowledge of green AI and its driving factors, the DGI offers an interactive learning experience, enhancing users’ understanding of green AI and strategy development. The tool can simulate and assess the impacts of various policy and investment options, providing valuable insights for planning and supporting the creation of green AI policies in crucial sectors.

Global Sustainability Targets

Our commitment to transforming organizational economies focuses on four key areas: a) sustainable energy and water, b) responsible governance of sustainability, c) AI, and d) the health and well-being of employees. Leveraging the Sustainable Development Goals (SDGs) as a framework, we aim to foster responsible innovation and growth in both public and private sectors. The Digital and Green Index (DGI) provides metrics to create a virtuous growth cycle benefiting multiple stakeholders. Utilizing the comprehensive SDG indicators, the DGI establishes a robust assessment mechanism to evaluate AI’s impact across various dimensions. The selection of sustainability targets is guided by SDG indicators and OECD⁶ report recommendations.

System Description

**Figure 3:** The Digital & Green Index App Dashboard.

The design process for the Digital and Green Index (DGI) involves developing and applying relevant frameworks, including an index and dashboards, while engaging a range of institutions. Two primary processes guide the creation of green AI growth frameworks: the fit-for-purpose principle and stakeholder consultations. According to the OECD Handbook, the fit-for-purpose principle ensures the selection of indicators that meet end users’ needs, relying on a solid theoretical foundation and scientifically driven indicators.

Developing composite indices requires meticulous steps to ensure transparency, replicability, and credibility, as outlined by Greco and colleagues⁷ (2019) Our stepwise approach begins with Phase I, involving four steps: a) identifying models interlinking indicators, b) requiring data, c) presenting underlying concepts and assessing indicator weights, and d) developing the DGI app to calculate green scoring. The index scoring is based on survey responses, with refined questions tailored for each indicator.

The DGI framework is built on four dimensions of sustainable AI: environmental ceiling, social floor, governance, and transversal aspects. These dimensions are interconnected, focusing on a low-carbon economy, resilient society, ecosystem health, and inclusive growth. Each dimension encompasses essential indicator categories, such as climate change, natural resources, pollution, and biodiversity for the environmental ceiling; health and well-being, education, and security for the social floor; and global sustainability for the transverse dimension. This comprehensive approach ensures a transition to green and sustainable AI pathways.

Between the lines

The Digital and Green Index (DGI) provides a multi-dimensional and multi-layered framework to assess the broader sustainability impacts of AI systems. Unlike existing metrics that focus solely on environmental indicators like carbon footprints, the DGI incorporates environmental, societal, and governance aspects of digital and AI systems. This comprehensive approach helps organizations and stakeholders evaluate the sustainability impacts of their AI systems, fostering responsible and sustainable digital ecosystems.

The DGI framework quantifies an organization’s impact on energy, natural resources, societal and ethical considerations, and governance. It positions itself at the intersection of societal and technological evolution, offering a sandbox for AI researchers to test and improve AI sustainability with diverse stakeholders.

Initially, the DGI serves as a generic self-assessment tool to raise awareness about the challenges of training and using large AI models. Over time, it is expected that researchers will refine and adapt the framework for various contexts and sectors. Policymakers, civil societies, and communities can also leverage the DGI to develop and implement policies promoting sustainable AI practices, potentially creating sustainability scoring systems similar to food nutrition labels like Nutri-Score.

Footnotes