Is Sustainable Artificial Intelligence an oxymoron?

The contributions published in the latest issue of BioLaw Journal offer a timely and thought-provoking perspective on what it means to design, regulate, and deploy AI systems in an environmentally responsible way.

While artificial intelligence is often celebrated as a powerful ally in tackling climate change (optimizing energy use, supporting climate modeling, and improving resource efficiency) there is growing awareness of the environmental costs behind the digital curtain (B.Marchetti) .

Recent reflections from researchers affiliated with LUMSA, Trento and other universities shed light on this crucial but underexplored issue and offer a multifaceted view of the ecological impact of AI, pointing out how enthusiasm for its potential benefits has often overshadowed the significant energy, water, and resource demands it entails.

One of the emerging concerns is the massive energy consumption associated with training large AI models, especially in the field of generative AI. From natural language processing to image generation, these systems require powerful data centers, whose electricity needs are forecasted to increase dramatically over the coming years. The environmental footprint is not limited to carbon emissions, water use for cooling, mining of rare earth elements, and the generation of electronic waste are all part of the equation (B. Marchetti).

The current European regulatory framework, including the recently approved AI Act, still lacks the necessary ambition to address these impacts. Although the regulation recognizes environmental protection as a fundamental value, concrete obligations related to sustainability remain vague or optional. This raises the question: can AI truly support the green transition if its development and deployment are not themselves green (N. Rangone).

Public administrations, as major adopters of AI systems, are also called upon to play a leading role. Sustainable criteria in public procurement processes could become a driving force for more responsible innovation (L. Parona). Similarly, the application of the "Do No Significant Harm" principle (already used in other EU policies) could help prevent AI systems from contributing to environmental degradation (L. De Gaetano). In the international arena, the debate is evolving, with growing demands for transparency from developers and operators, especially in the United States. There is increasing pressure for companies to disclose data on energy and water usage, and to assess the environmental risks associated with their models throughout the entire lifecy (M. Merler).

Readers can explore the full articles here:

• I costi ambientali dell’IA (The Environmental Costs of Artificial Intelligence) – Barbara Marchetti
• Intelligenza artificiale, tutela dell’ambiente e regolazione europea (Artificial Intelligence, Environmental Protection and European Regulation) – Nicoletta Rangone
• La ponderazione dei costi ambientali nell’approvvigionamento di sistemi di intelligenza artificiale da parte delle amministrazioni pubbliche (Considering Environmental Costs in the Procurement of AI Systems by Public Authorities) – Leonardo Parona
• Il principio Do Not Significant Harm (DNSH) e i costi ambientali dell’intelligenza artificiale (The Do Not Significant Harm Principle (DNSH) and the Environmental Costs of Artificial Intelligence) – Lorenzo De Gaetano
• I costi ambientali dell’intelligenza artificiale: il dibattito negli Stati Uniti (The Environmental Costs of Artificial Intelligence: The Debate in the United States) – Marianna Merler