Large language models unlock the ecology of species interactions

Species interactions can determine species population sizes, geographic ranges, evolutionary trajectories, and responses to environmental change. Yet, despite their importance to many fundamental and applied questions, information on species interactions is often lacking due to constraints in data collection. Billions of text comments that have been submitted by millions of citizen scientists around the world have the potential to fill these gaps. Comments can be used to identify biotic interactions using advanced large language models (LLMs), providing a novel source of interaction data that is unusually high in spatiotemporal coverage, breadth, and resolution. This novel approach opens new avenues to evaluate species interactions on a broader scale, and to characterize and conserve biodiversity under pressing global change. Highlights- Although species interactions are central to biodiversity dynamics, progress in resolving their fundamental properties and forecasting their shifts under global change has been hindered by persistent data limitations - Citizen science platforms contain billions of observer text comments that often contain valuable information about species interactions, but the unstructured format of the information and the size of the datasets make these comments difficult to use - Large language models (LLMs) provide an unparalleled opportunity to collect and analyze species interactions from such comments - Using two case studies, we present a workflow that leverages LLMs to automatically collect species interaction observations from citizen science comments in multiple languages around the world - Such a novel source of data greatly expands the data coverage and resolution of species interactions across space and time and can help to answer both long-standing ecological questions and new, pressing questions about ecological responses to global change