About

The world’s data centers currently require two percent of global electricity usage and contribute 4% to global greenhouse gas emissions, with these numbers set to double in the next few years. Since we are only starting to comprehend and measure the impact of Artificial Intelligence (AI) on the environment, its contribution towards this number is unclear. What we do know is that AI systems are compute-intensive, which increases the workload of datacenters, increasing their overall energy use, most of which comes from nonrenewable sources.

While some research and documentation on AI’s environmental impacts currently exists, the nature and extent of AI’s effects are under-documented, ranging from its embodied and enabled emissions to rebound effects due to its increased usage.

The Sustainable AI Hub is an initiative that aims to bring together resources, projects and materials around making AI less harmful to people and the planet.

Current State of Knowledge on AI’s Environmental Impacts

Initial work on the environmental impacts of AI models includes the seminal work of [Strubell et al(https://ojs.aaai.org/index.php/AAAI/article/view/7123), which calculated that training a large language model (LLM) was responsible for 626,155 pounds of CO2 emissions, roughly equivalent to the lifetime emissions of five cars: this number was calculated based on an earlier version of current models, which were even smaller than the ones that exist today. Follow-up studies have looked at other types of model architectures, their energy use and emissions, confirming the consequential environmental impact of AI training, as well as its indirect effects on different industries.

Sasha Luccioni’s work has proposed to extend previous analyses of AI’s environmental impacts to consider the broader AI model lifecycle, revealing a fuller picture of AI’s environmental impact that goes beyond training to take into account material extraction, equipment manufacturing, and overhead energy usage. I have also contacted hundreds of authors to gather information about the models they’ve trained [9], analyzing the factors that influence their emissions and proposing ways to reduce them.

In terms of existing communities in the AI and climate space, the largest existing organization is Climate Change AI, a global non-profit that catalyzes impactful work at the intersection of climate change and machine learning. Existing workshops such as SustaiNLP, Climate Change AI and FAccT provide opportunities for publishing and discussing relevant research, but there is no single journal or conference that has centralized this scholarship.