Assessing the impact of AI and ML on greenhouse gas emissions

Lynn Kaack and co-authors offer a framework for researchers and policymakers in the journal Nature Climate Change.

What are the greenhouse gas emissions impacts from the growing use of artificial intelligence (AI) and machine learning (ML)? A new paper co-authored by Lynn Kaack, Professor of Computer Science and Public Policy and faculty at the Hertie School’s Data Science Lab, and published in June 2022 in the journal Nature Climate Change, offers a framework to quantify and assess how these new technologies may alter greenhouse gas emissions – both positively and negatively ­– and what policy measures can help to align ML with climate change mitigation.

The authors divide this emissions impact into three categories: direct impacts from computing on end-user devices, servers and data centres; immediate impacts from AI applications used in the economy; and system-level impacts through structural change such as increased demand for products and services and lifestyle changes. They illustrate and analyse these impacts through examples, showing how they can be both positive and negative for the climate. And they offer a number of recommendations for policymaking and the private sector.

“Climate change should be a key consideration when developing and assessing AI technologies. We find that those impacts that are easiest to measure are not necessarily those with the largest impacts. So, evaluating the effect of AI on the climate holistically is important,” says Kaack.

The paper, “Aligning artificial intelligence with climate change mitigation”, was co-authored by Lynn Kaack, as well as Priya L. Donti and Emma Strubell of Carnegie Mellon University, George Kamiya of the International Energy Agency, Felix Creutzig of the Mercator Research Institute on Global Commons and Climate Change and the Technical University Berlin, and David Rolnick of Mila – Quebec AI Institute and McGill University. Their findings were also cited in the latest report of the Intergovernmental Panel on Climate Change (IPCC), and in a report for the Global Partnership on AI co-authored by Kaack.

Measuring and forecasting the impact of AI and ML on greenhouse gas emissions is particularly difficult because of the diverse mechanisms through which they take place, the authors say, but even without precise estimates a lot can be done to reduce it. “ML’s ultimate effect on the climate is far from predestined, and societal decisions will play a large role in shaping its overall impacts. This will require a holistic portfolio of approaches across policy, industry and academia to incentivize uses of ML that support climate change strategies while mitigating the impacts of use cases that may counteract climate change goals,“ they write.

The paper also provides a basis for further exploration. “Research areas such as life cycle analysis and industrial ecology can now build on the AI-specific considerations in our study,” says Kaack.

Find the paper in Nature Climate Change here.


The Hertie School is not responsible for any content linked or referred to from these pages. Views expressed by the author/interviewee may not necessarily reflect the views and values of the Hertie School.

More about our expert

  • Lynn Kaack, Assistant Professor of Computer Science and Public Policy