A field guide for the Wild West of AI-assisted environmental science

Researchers at UC Santa Barbara's National Center for Ecological Analysis and Synthesis (NCEAS) have developed a field guide to help environmental data scientists navigate the challenges posed by AI-assisted coding. The guide, published in PLOS Computational Biology, emerged from a need within the NCEAS community to establish guidelines for responsible use of generative AI in code development.

The six-person team that created the Wildfire Resilience Index, an open-access tool for measuring community and landscape wildfire preparedness, encountered rapid changes in AI technology. By the time their project was released, new AI coding tools had emerged, rendering the team's earlier learning irrelevant. This experience was not unique to the WRI team; dozens of research teams at NCEAS faced similar challenges.

To address these issues, NCEAS convened a community of researchers, software developers, data analysts, and professors to discuss responsible AI use. The result was "Ten simple rules for effective use of generative AI for code development in environmental science," which outlines guidelines for preparing projects, selecting appropriate AI tools, verifying generated code, and documenting the AI usage process.

The authors emphasize that appropriate AI use is a skill that has not yet been taught, and that their field requires specific guidance for effective AI integration.

The researchers also highlight the unequal benefits of AI, noting that male researchers report larger productivity gains than their female counterparts. They point to a UN report indicating that while roughly two-thirds of people in high-income countries use GenAI tools, usage in many low-income countries remains low, as low as 5%. The researchers caution that GenAI tools may become inaccessible to underfunded institutions and researchers in low-income countries once they shift toward paid tiers.

Additionally, the environmental impact of GenAI is uncertain due to the rapidly changing landscape and the private nature of much of the data needed for estimates. Data centers are projected to consume a significant portion of U.S. electricity and water by 2030 and 2028, respectively. The researchers stress the need for a task-specific ethical assessment when deciding to use AI, rather than providing a blanket endorsement.