How AI is helping address the climate crisis

Emerging technologies are playing a pivotal role in the global effort against climate change, helping the world decarbonize energy production, boost the efficiency of industrial systems, and reduce the carbon footprint of everyday life. The development of these technologies represents one of the most urgent engineering challenges of our lifetimes and one that we believe will be powered in significant ways by advances in AI.

We’ve already seen, in so many technical fields, how AI can reshape our approach to solving problems. It can supercharge computational approaches to scientific research, enabling calculations that once took days to happen in hours and transforming the ability to model complex systems or process vast amounts of data. As a force multiplier for scientific research, AI is helping accelerate the rate of progress across many domains, including those most important to solving the climate crisis.

Computational approaches to chemistry, for example, allow researchers to simulate millions of experiments in a way that would be impossible to attempt if each needed to be done in a lab. As scientists search for new compounds and materials that could help scrub carbon dioxide from the atmosphere or convert it into a clean fuel source, AI will help accelerate the search.

One promising attempt at this is the Open Catalyst Project, run by a partnership between Meta AI and Carnegie Mellon University’s Department of Chemical Engineering. The project is bringing together AI researchers from across the world to design new machine learning models capable of predicting the result of chemical reactions, with the aim of making complex computer simulations that currently take hours or days happen instead in a matter of seconds. This leap in efficiency could help researchers identify the new materials needed to enable a massive leap forward in technologies for combating climate change. The team has already released the world’s largest training data set of materials for renewable energy storage, and will soon release a new data set with over 8M data points from 40K+ unique simulations across a variety of materials for green hydrogen production. We believe this is the largest data set for oxide catalysis to date. Here’s a conversation I recently had with Larry Zitnick, the researcher leading this amazing project:

Something Went Wrong

We're having trouble playing this video.

Learn more

The potential for AI to boost decarbonization efforts extends to the ability to monitor, measure, and model the effects of climate change and our efforts to mitigate them. This kind of work will grow in importance as governments and organizations seek to verify the effectiveness of carbon offset schemes or track the progress of environmental commitments.

Even a task as fundamental as understanding how much of the Earth is covered with forests, and how quickly these forests are changing in size and density, could be supercharged by better AI. Researchers at Meta are exploring using cutting-edge computer vision systems to interpret satellite imagery and produce extremely detailed, high-resolution maps of the world’s forests. The goal of the project is to estimate the carbon stored in forested areas, which would improve the ability to assess the impact of reforestation efforts.

This type of progress is incredibly exciting, but there’s equally vital work to be done in making AI itself become more carbon- and energy-efficient. Current approaches to training AI models depend on vast amounts of data processing and computing power, both of which are sources of significant energy consumption. Once these models are trained, putting them into operation is equally energy-intensive, especially when they operate at large scale, as they do at technology companies like Meta. And the physical infrastructure needed to support all this work produces significant emissions of its own, from the construction materials needed to build data centers and other facilities to the computing hardware and equipment that powers them. As AI use grows exponentially, the environmental impact of its growing infrastructure footprint needs to be accounted for.

Meta’s global operations (including projects like the Research SuperCluster, which will be the fastest AI supercomputer in the world) are supported by 100 percent renewable energy, but efficiency is still crucial as we build new, more powerful AI systems and grow our overall use of the technology across the company. Researchers at Meta and throughout the industry are currently exploring a number of approaches to Green AI, which includes things like developing the standards needed to measure the energy efficiency of an AI system to the algorithms and computing hardware needed to operate AI at scale. We believe this work will enable AI systems to grow sustainably and with lower infrastructure needs.

In a recently published paper, Meta researchers highlighted the scale of the challenge: The pursuit of more powerful AI models comes with an exponential scaling of the size of these models and the computing resources needed to train and operate them. But the potential to optimize these models to achieve the same performance gains with a significantly reduced carbon footprint is equally large: In an experiment on one AI model used for language translation in Meta’s products, the researchers identified optimizations that led to an 800x reduction in the infrastructure resources needed to serve the model at scale. A similar study by researchers at Google found that through the use of a number of best practices, they were able to reduce energy usage by 100x and carbon emissions by 1,000x.

The impact of such a reduction will be magnified by AI’s overall ability to drive major efficiency gains across a range of industries from farming to manufacturing. In agriculture, AI systems are helping optimize water and fertilizer usage and increase the productivity of farm equipment and systems. In manufacturing, AI tools are driving major improvements in robotics systems and process controls, enabling factories to be more productive with less waste and lower energy consumption. As industries race to decarbonize, AI will help drive the efficiency gains that make it possible.

While making today’s economy more efficient will be one of AI’s most important environmental contributions, the inventions it enables in the future could be even more consequential. Scientists around the world are learning to harness AI’s power to accelerate their research, and the result could be fundamental breakthroughs that are transformative to the effort to decarbonize.

We’re incredibly optimistic about the impact AI is going to have on climate and sustainability, and the role that our researchers and engineers can play in helping build it. And by applying that progress to our own operations, we hope to set an example for just how much impact this incredible technology can have on creating a sustainable global business.