Overview
The Earth system—the atmosphere, ocean, cryosphere, biosphere, and geosphere that cycle energy, water, nutrients, and other trace substances—is a large, complex, not linear, multiscale system in space and time that involves human and natural system interactions. Understanding and predicting such a system presents an enormous computing and analysis challenge. Machine learning (ML) and artificial intelligence (AI) offer opportunities to tackle these challenges. Researchers are actively exploring ways to use ML/AI approaches to advance scientific discovery, make computation faster and cheaper, provide new system understanding, and link disparate scientific communities.
The National Academies of Sciences, Engineering, and Medicine convened a workshop on February 7, 10, and 11, 2022, on the opportunities and challenges of using ML/AI to advance Earth system science, including their ethical development and use. The workshop convened experts in all components of the Earth system, ML/AI researchers, social and behavioral scientists, ethicists, and decision makers across sectors to explore how these approaches can contribute to improving understanding, analysis, modeling, prediction, and decision making.
Workshop discussions highlighted the benefits of bringing together Earth system science and ML/AI experts. These two communities have different philosophies, vocabularies, and practices, which could be bridged through more robust interactions. In addition, off-the-shelf ML/AI approaches are often not sufficient for Earth system science applications, and domain scientists could benefit from engaging in the design of ML/AI tools and contributing their scientific knowledge. Opportunities exist to create better educational pathways for students and researchers who understand Earth system science to innovate in ML/AI.
Participants explained the high stakes for doing ethical and responsible ML/AI for Earth system science. Ethical choices are being made in the deployment of ML/AI, and these choices could be better explained, recognized, and improved. Particular challenges for applications to Earth system science include biases in training data and AI models. Opportunities also exist to develop sustainable AI—elevating smaller models and accounting for carbon efficiency—and to deploy responsible AI for climate action by ensuring that AI-for-climate applications are pursued fairly, accountably, transparently, and equitably. Participants also discussed how the interpretability of ML/AI tools could be a design goal from the outset of their development, and explainable and interpretable AI could be used for knowledge discovery, to build trust, and to uncover problematic algorithms.
Current Earth system science workflows present barriers to collaboration and inclusion, limit knowledge transfer, and may not scale to future data needs. While many current applications of ML/AI can exacerbate challenges in the existing paradigm, participants identified opportunities to build new infrastructure based on open science and open data principles. In order to achieve a system of truly open knowledge, participants suggested efforts to encourage contributions to community code, build capacity to enable equitable access to science, and foster an inclusive open source community. Infrastructure to support open science for ML/AI and Earth system science could support multisector collaborations, spur innovation, and benefit the entire community that is interested in using these tools and data.
Educational opportunities for those entering the workforce and those already working who want to continue their education at the intersection of ML/AI and Earth system science were raised throughout the workshop. Employers in the private sector are looking for workers
with production level skills who can optimize, automate, and process large amounts of data. Participants highlighted the value of early data literacy, interdisciplinary skills, and the development of learning communities. Speakers recognized the importance of diversifying the Earth system science and ML/AI communities and identified room for improvement in ensuring that these spaces are inclusive. Continued investment in broadening the workforce could include deeper partnerships between academia and the private sector, paid internships, and the expansion of existing programs like the National Science Foundation’s Research Experiences for Undergraduates.
In order for scientific advances at the intersection of ML/AI and Earth system science to be useful for decision making, participants discussed opportunities for further research on how end users understand information provided by an ML/AI algorithm and how best to communicate uncertainty. Earth system science has large and rapidly growing datasets that have enabled the effective use of ML/AI, and an opportunity exists to progress from success with simple problems to the use of ML/AI tools to model and predict complex scenarios. ML/AI tools are promising for producing probabilistic forecasts and integrating physical models and data; however, there are barriers to entry to learn and modify ML/AI methods, and further research may be needed to appropriately incorporate the robustness of ML/AI tools into decision making.
