University of Michigan researchers use AI to extract birdwatcher data

AI Extracts Ecological Data from Birdwatcher Comments

Researchers at the University of Michigan have demonstrated that large language models like ChatGPT and Claude can extract structured ecological data from millions of informal text comments on citizen science platforms such as eBird and iNaturalist.

Every day, hundreds of thousands of birdwatchers and nature enthusiasts upload their observations to these digital platforms. Beyond standardized metrics like species names and location, users routinely include short remarks in comment fields detailing unique animal behaviors, such as birds fighting over food, roosting together, or preying on other species. Historically, processing this massive volume of informal text manually proved impossible for scientists due to the sheer scale of the data.

Unlocking Hidden Text Behind the Data Bottleneck

The new study, published in the scientific journal Proceedings of the National Academy of Sciences (PNAS), utilized artificial intelligence to scan millions of user comments from eBird and iNaturalist. In this study, the research team tested the ability of LLMs to scan millions of user comment records from the eBird and iNaturalist platforms. The models were trained to identify, categorize, and pull out specific information regarding interspecies interactions—spanning competition, predation, and mutualistic symbiosis.

This automated approach processes thousands of raw comment entries in seconds, bypassing a data extraction bottleneck that previously required years of manual human labor.

Mapping Food Web Dynamics Across Vast Scales

Using this technology, ecologists have mapped species interaction networks across vast geographic scales and timeframes that were previously unmapped in comprehensive detail. Through the help of this technology, the scientists successfully mapped species interaction networks across an extremely wide geographic scale and time range that had never been comprehensively mapped before.

This automated extraction framework supplies researchers with fresh instruments to evaluate food web dynamics and ecosystem responses to environmental shifts or global warming. According to the University of Michigan research team, citizen science platforms provide exceptionally rich datasets, and artificial intelligence unlocks the potential of hidden textual information to answer large-scale ecological questions.

Amateur Observations Fueling Global Biodiversity Efforts

The integration of large language models into ecological research highlights how routine submissions from amateur birdwatchers and nature observers, when paired with automated text processing, yield critical scientific insights for worldwide biodiversity conservation efforts.