How to Use AI to Combat Climate Change

On a recent snowy evening at Bowdoin, members of the campus community braved the weather to hear Dan Loehr’s talk “Addressing Climate Change with AI.”

Loehr is an adjunct lecturer at Georgetown University, where he received his PhD in computational linguistics. He currently teaches courses on AI and climate change, and previously spent thirty years overseeing research and development and the application of AI for several large organizations.

Loehr’s talk focused on breaking down the climate crisis into the numerous sectors where it looms large—like energy, transportation, carbon dioxide removal from the atmosphere, land use, and weather forecasting—and supplementing each area with a specific AI method that is currently being applied to address issues within that sphere.

Reducing airplane contrails: For example, contrails from airplanes contribute to climate change because they act as clouds and trap heat on Earth. Loehr showed an example of scientists using image recognition with images from satellites and weather data to develop “contrail forecast maps” that anticipate pockets of humid air where contrails are more likely to form. As a result, pilots can adjust the altitude they are flying at to avoid those pockets.

New formulas for producing building materials: In the industry sector, cement and steel production account for 15 percent of greenhouse gas emissions. This is because burning coal is used to produce the extreme heat needed to manufacture both, in addition to their chemical compounds giving off carbon dioxide. Every ton of cement or steel produced yields two tons of carbon dioxide into the atmosphere.

Generative AI, which encodes the “essence” of the input to produce a similar output, can be deployed to generate ideas for a new formula for concrete that produces less carbon or requires less cooling under the same parameters: strong, cheap, and which can take any shape.

“This is work on basically accelerating the discovery of building materials,” Loehr said. “And this is not just cement. You can do it for new battery materials [and] for new carbon absorption materials.”

Visualizing climate change impacts: Generative AI can also be employed in understanding and visualizing the societal impacts of climate change. Researchers are currently using AI to generate images of people’s homes being flooded so individuals can understand and prepare for the personal impacts of climate change.

AI’s own contribution to climate change: Loehr also spoke about the carbon footprint of AI. While there is limited data on emissions from AI currently, estimates from 2021 and 2022 say that AI is responsible for 0.15 percent of global greenhouse gas emissions. In the worst-case scenario, this is predicted to rise to 1.5 percent by 2027.

To keep perspective, Loehr said, this is still less than the emissions caused by cow burps (2 percent) and only one-tenth the amount of emissions from carbon and steel. Additionally, ChatGPT currently uses less water daily than hamburgers eaten daily in the US and less energy than daily US TV usage.

“It’s very small compared to society’s larger footprints,” Loehr said. “But again, locally, people are really mad about the data centers because they’re taking the local power and the local water.”

While the impact is currently small, Loehr said it is still important to mitigate the impacts of AI since they are rising rapidly.

Loehr concluded his talk by providing resources for ways to get involved with using AI against climate change. Websites like climatechange.ai provide Python tutorials for applying various AI methods to climate change, while Project Drawdown provides examples of non-AI solutions being used to combat climate change.

Referencing past societal problems that scientists solved, like acid rain and smallpox, Loehr said: “All of these things we solved when we got to a societal tipping point. And so, if we start talking about climate change and the things we can do, then society reaches a new point.”