Why Sasha Luccioni wants to create Energy Star ratings for AI

Sasha Luccioni was experiencing something of a quarter-life crisis. It was early 2018, and Luccioni was working as an AI researcher at Morgan Stanley, where she spent her days thinking about how to use the rapidly advancing technology to head off financial risks. But over the course of a year in the job, Luccioni couldn’t escape the feeling that there was a far greater risk—namely, climate change—that deserved her attention.

“I remember telling my partner, ‘Maybe I should just quit and go teach kids how to plant trees,’” she tells Fast Company. “He was like, ‘Well, you have a PhD in AI, maybe you can use that?’”

That turned out to be sound advice. As climate and AI lead for the open source machine learning platform Hugging Face, Luccioni is now perhaps the most prominent researcher focused on exposing the outsized climate impact of large language models, which require massive amounts of energy to train. The energy demands of developing these models are so great, in fact, that tech giants like Google now say they may miss their climate goals as electricity requirements, and in turn, emissions grow. “What’s crazy is that they set the goals themselves, then they missed the goals themselves, which is pretty shocking,” Luccioni says. “Companies tend to set goals that they can meet.”

Since shifting her focus to the overlap between AI and climate change, Luccioni has co-created a tool developers can use to estimate the carbon footprint of whatever they’re building and find ways to reduce that impact. She’s helped found an organization focused on ways to use AI to combat climate change. And she’s produced a steady drumbeat of research quantifying the carbon emissions of both training machine learning models and deploying them for everyday tasks.

It’s a job that has only grown more challenging over the last few years, as tech companies have grown more secretive about how and where their coveted large language models are built. “For a while companies were actually pretty forthcoming…until ChatGPT came out,” she says. “It blurred the line between AI research and consumer products.”

But Luccioni’s job doesn’t just involve conducting research. Equally important is the task of communicating that research to a wider audience through media interviews, conference appearances, and, last year, a main stage TED Talk. “Research is great,” she says, “but if people don’t know about your research, then it’s going to have limited impact.”

In that spirit, Luccioni is now working on a project that grades different types of AI models based on their energy efficiency. She envisions it as being similar to the EPA’s Energy Star Rating system for appliances—a way of talking about AI’s impact without having to discuss kilowatt hours and other terms that make most non-scientists’ eyes glaze over. Her goal is to encourage people to use the “right model for the right task,” rather than defaulting to the largest and, therefore, most power-hungry LLMs to carry out small requests. “People still don’t understand that there’s a materiality to AI,” Luccioni says. “I feel compelled to make people understand that when they use ChatGPT like a calculator, that comes with a cost to the planet.”

This story is part of AI 20, our monthlong series of profiles spotlighting the most interesting technologists, entrepreneurs, corporate leaders, and creative thinkers shaping the world of artificial intelligence.