Sasha Luccioni left her job as an AI researcher at Morgan Stanley in 2019 because she was concerned about how to explain and justify the technology’s impact on greenhouse gas emissions and water consumption.
Two years later, she became one of the first researchers to demonstrate the link between AI and energy use, as the first AI and climate leader for Hugging Face, an open source software organization that hosts AI code. Now, as co-founder and chief scientific officer of the new consulting firm Sustainable AI Group, Luccioni is committed to sharing what she’s learned with corporate sustainability professionals.
“More and more, I started to realize that it’s important that research serves people, serves users, serves the general public,” she told me in the latest episode of our Climate Pioneers interview series. “We want to do more research that sheds light on this question and helps companies essentially understand the environmental impact of AI, how to get more information, because it’s not very clear, and how to decarbonize.”
Sustainable AI Group was created to serve that mission. Alongside bespoke consulting services, the firm will create resources that sustainability professionals can use to assess and manage emissions and other AI climate impacts, including water and the minerals used in computing hardware.
“There’s still so many questions that are unanswered about AI’s environmental impacts, and we want to keep answering those questions alongside the community,” Luccioni said.
Rightsized AI
Close to 80 percent are using AI in their day-to-day work, mainly generative AI and chatbots that speed efficiency tasks, according to a Trellis Briefing reader survey. That mirrors Luccioni’s own habits: She’s still seeking the perfect task to outsource to AI, beyond drafting clever headlines for her research papers.
AI’s real superpower is capacity building that complements human experience, Luccioni said, and she’s optimistic about how it might be applied to applications such as the development of cleaner, higher-capacity batteries or biodiversity monitoring.
For example, AI could be used to flag and eliminate bogus information or empty images captured by trail cameras. Often, these applications can be run on local desktop or notebook computers. “These may seem like small contributions but at a global scale or the scale of a whole network of people, it’s really impactful,” she said.
The typical design of large language models used by generative AI is problematic, however, because most process the queries using faraway data centers. “I think that the way we’re doing this with these general purpose, massive models is really not the way to do it,” Luccioni said. “It can be done with smaller models and on-device models.”
Position of power
As more companies commit to enterprise-scale AI licenses, sustainability professionals can regain control by encouraging their technology buyers to require energy and water consumption disclosures from AI mode
Read More
