How to Achieve Sustainability in AI

Sustainable AI development appears to be a distant aspiration, as major technology firms that once pledged to reduce emissions accelerate the construction of large data centers reliant on fossil fuels.

The frenzy to advance AI technology regardless of costs has been fueled by the Trump administration’s rollback of environmental safeguards.

However, despite these challenges, Sasha Luccioni, an AI sustainability researcher, believes there is an unprecedented demand for transparency in AI from both companies and consumers.

Luccioni has emerged as a key advocate for increasing awareness of AI’s environmental effects during her four-year tenure at Hugging Face, an AI firm, where she has led initiatives like a leaderboard tracking the energy efficiency of open-source AI models. She has also been a vocal critic of leading AI companies that, according to her, intentionally conceal energy and sustainability data from the public.

She is now launching the Sustainable AI Group, a new initiative alongside former Salesforce sustainability chief Boris Gamazaychikov. Their mission will involve assisting companies in exploring questions like, “what levers can we adjust to mitigate negative impacts?” Luccioni is also keen on investigating the energy requirements of various AI tools, from speech-to-text translation to photo-to-video conversions, areas she feels are currently underexplored.

Luccioni recently spoke with WIRED about the increasing call for sustainable AI and her expectations from major tech firms.

This interview has been edited for length and clarity.

WIRED: I often hear from individuals concerned about environmental issues and AI, but companies seem less vocal. What insights have you gained from those engaged with AI in business settings, and what concerns do they express?

Sasha Luccioni: First off, there’s significant pressure from employees—and from boards and directors—demanding quantifiable impacts. Employees are asking, “If we’re using Copilot, how does this align with our ESG objectives?”

For many organizations, AI has become integral to their offerings. Thus, it’s crucial for them to understand associated risks and the operational environments of their models. They cannot afford to utilize models without knowing the locations of their data centers or the power grids involved. It’s essential to assess supply chain emissions, transportation emissions, and other factors.

It’s no longer about abandoning AI; that conversation has evolved. It’s about selecting appropriate models, for instance, and signaling that the energy source is significant enough that consumers are willing to invest more in data centers powered by renewable energy. There are viable routes forward; it’s simply about identifying the right proponents.

Moreover, the sustainability landscape for global enterprises must be quite different from that in the US, correct? While the US government may not prioritize this, other nations are more engaged.

In Europe, the EU AI Act has incorporated sustainability as a vital element from the outset. Numerous clauses are included, and initial reporting initiatives are just beginning.

Even in Asia, there’s a push for greater transparency. The International Energy Agency has been producing reports on AI and energy consumption. During discussions, officials indicated that other nations realize the IEA sources their data from member countries, and many of these countries lack specific figures for data centers. This gap hinders their ability to make informed future projections about capacity needs. Some nations have begun advocating against the practices of data center developers.