Executive summary
Artificial intelligence (AI) is already driving scientific breakthroughs in a variety of research fields, ranging from the life sciences to mathematics. This raises a critical question: can AI be applied both responsibly and effectively to address complex and interconnected sustainability challenges? These challenges include climate change, biodiversity loss, ocean acidification, and other transformations of our living planet. Each of these poses serious risks to societal stability, human health and well-being, and the ability of present and future generations to thrive within planetary boundaries. Here, science plays a fundamental role in two ways. First, by helping accelerate changes and innovations that move us closer to a just and safe future for all. Second, by making sure that AI is developed and used in ways that neither exacerbates inequalities, nor increases planetary pressures.
This report explores the potential and limitations of using AI as a research method in eight issue areas: Preparing for a Future of Interconnected Shocks, Understanding a Complex Earth System, Stewarding Our Blue Planet, Securing Freshwater for All, Enhancing Nature’s Contributions to People, Prospering on an Urban Planet, Improving Sustainability Science Communication, and Collective Decisions for a Planet Under Pressure. Together, these areas reflect a range of complex research challenges where people, nature, and climate interplay. By taking such an integrated approach, our work stands out within the growing landscape of “AI for Science,” and brings together “AI for Climate” and “AI for Nature” initiatives.
Our results build on iterated expert dialogues and assessments, a systematic AI-supported literature overview including over 8,500 academic publications, and expert deep-dives into each specific issue area. In conclusion, we show that: (1) AI offers vast potential to accelerate progress across the sustainability sciences. (2) AI can sharpen our decision-making and clarify complex environmental challenges for researchers and the public alike. (3) However, realizing this promise requires careful navigation of the risks, including AI’s own environmental footprint, inherent biases, and the challenge of unequal access. (4) Despite these hurdles, responsible and ethical applications of AI in sustainability research are not just a possibility—they are an urgent necessity. (5) Pioneering these uses can unlock the breakthroughs we need to build a more sustainable future.
Advancing from the potential of AI to its responsible application in research that benefits both people and the planet will require a balance between urgency and innovation, and a commitment to ethical and inclusive practices. Our concluding recommendations—addressed to researchers, policymakers, funding agencies, and philanthropic organizations—outline practical steps for achieving this balance. Leveraging powerful compute (i.e., computational power) and new methodologies, advancements in AI enable the discovery of patterns and relationships that were previously too complex, or too interdisciplinary, to tackle. As experience has shown, AI has the potential to open up new, exciting areas of inquiry for sustainability research. It’s an opportunity the world cannot afford to miss. It is now up to sustainability researchers, companies, technology entrepreneurs, and decision-makers to carefully shape AI development in the service of a just and safe future.
Use of AI in Writing this Report
L.W-E. used Grammarly for editing the Freshwater chapter. N.K. has used ChatGPT to rephrase and shorten parts of the draft version of the Freshwater chapter. E.Z. used Gemini 2.5 Flash for grammar and syntax checks of all writing contributions to the report chapters; GitHub Copilot assisted with script programming to improve the visual appeal of the plots in the literature review and conclusion chapters; and used DeepSeek-R1-7b for analysis of the collected literature, as described in the corresponding chapter. A.V. used Gemini for search of relevant literature (together with Scopus) and editing. L. D. used ChatGPT (4o) for spell checking and rewording individual sentences, upon which he reviewed and edited the content. W.B. used Grammarly for language editing. M.R. used ChatGPT to refine the academic language. M.S. used Copilot in Microsoft 365 to revise some of the sentences in the full report to improve readability. C.S. used Gemini to provide input to the summary of the full report, and worked with V.G. and other chapter authors with the report summary, conclusions, and recommendations.
Funding
Workshop, travel costs, research assistance time, and design and lay-out of this report, has received funding from Google DeepMind. V.G. would like to acknowledge funding from the Marianne and Marcus Wallenberg Foundation, Google.org, and the Beijer Institute of Ecological Economics (Royal
Swedish Academy of Sciences). M.S. acknowledges financial support from Stockholm Resilience Centre, the Beijer Institute of Ecological Economics, and Google.org. L.W-E. acknowledges funding from Formas (2022-02089, 2023-00321), and the IKEA Foundation. J.R. acknowledges support from Vetenskapsrådet grant 2022-04122. I.F. acknowledges funding from the IKEA Foundation. A.V. acknowledges funding from the European Research Council (ERC), Grant agreement No 101124903 – TwinPolitics – ERC-2024-CoG. M.M. acknowledges funding from ML4Earth by DLR. R.L. acknowledges funding from Formas (2022-02089). L.D. thanks the Erling-Persson Family Foundation for funding. M.G. acknowledges founding from the Volkswagen Foundation under the Freigeist program. W.B. acknowledges financial support received from the Cooperative AI Foundation. T.M. acknowledges financial support from Google.org. J.H-S. acknowledges funding from Marcus and Marianne Wallenberg foundation 2018-0093, and Vetenskapsrådet 2021- 03892. M.R. gratefully acknowledges support from the project “KI und Citizen Science gestütztes Monitoring von zertifizierten Biodiversitätsprojekten (KICS-Zert)” (16LW0441), funded by BMFTR, Germany. A.S. received support from William H. Miller III, the Princeton University Dean for Research, the High Meadows Environmental Institute, and the Army Research Office under grant number W911NF2410126.
Competing interests
D.P. and C.S. are both employed by Google DeepMind. F.T. has received funding from Google Research between 2020 and 2024 through the programmes: Latin America Research Awards (LARA), AI for Social Good, and the Award for Inclusion Research (AIR). T.M., A.M., V.G. and M.S. receive financial support from Google.org. V.G. has also engaged as consultant for Klarna and Milkywire for the program “AI for Climate Resilience”. The workshop and travel costs, research assistance time, and design and lay-out of this report, has received funding from Google DeepMind.
Thank You
Emelie Elfvengren (Stockholm Resilience Centre, Stockholm University), has provided helpful research assistance in the compilation of this report. Fredrik Moberg (Stockholm Resilience Centre, Stockholm University, and Albaeco) has proofread and commented on an early version of the full report. Elinor Kruse (Google DeepMind) contributed to reviewing the Taxonomy chapter and the final version of the report. Marcus Lundstedt (Stockholm Resilience Centre, Stockholm University) has provided helpful advice in discussions about the report title, key messages, and Executive Summary. Professional proofreading has been done by The Content Creation Company (CCC), UK.
The views and conclusions expressed in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of their respective organizations, the Army Research Office, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation herein.
.png)
