Researchers, students and practitioners recently gathered at Columbia University for the Emerging Frontiers Symposium: AI, Genomics & Community Science in Biogeography, a full-day event exploring how rapidly evolving tools and new data sources are transforming biodiversity research and conservation. The symposium was hosted by the Columbia MS in Biodiversity Data Analytics in partnership with the American Museum of Natural History and the Department of Ecology, Evolution and Environmental Biology at Columbia.
The symposium brought together experts working across artificial intelligence, genomics, species distribution modeling, community science and conservation. It focused on advances within each field and what becomes possible when they intersect.
“Artificial intelligence, genomics, community science applications — they’re all frontiers,” said Viorel D. Popescu, senior lecturer and director of Columbia University School of Professional Studies M.S. in Biodiversity Data Analytics program. “And they’re not separate. We need to integrate them.”
Popescu also framed the symposium as part of growing momentum in biodiversity and AI research at Columbia. Just four days earlier, Columbia unveiled the National Synthesis Center for Organismal Resilience (NSCORE), a $20 million NSF-funded institute designed to integrate organismal biology, artificial intelligence, and data science to understand how nature works. Meanwhile, the M.S. in Biodiversity Data Analytics program welcomed its first students this fall.
Together, these initiatives reflect a broader effort to connect cutting-edge analytical tools with one of the most consequential challenges facing scientists and society: understanding how biodiversity is changing, and using that knowledge to make better, sustainable decisions about the natural world.
The symposium brought together experts working across artificial intelligence, genomics, species distribution modeling, community science and conservation. It focused on advances within each field and what becomes possible when they intersect.
Learning From the Past as AI Advances
The first session focused on artificial intelligence and biogeography, with Lauren Gillespie, assistant professor at the University of Michigan, and Robert Anderson, professor at the City University of New York.
Gillespie traced the progression from traditional machine learning to deep learning and increasingly large foundation models. These approaches can draw on enormous amounts of biodiversity, climate and remote-sensing data, creating new possibilities for predicting where species occur and how ecosystems may change. Her research also illustrated the limitations of those tools. Biodiversity data can be biased: observations can cluster around easily accessible locations, favor particular species or reflect socioeconomic differences in who can access the technology used to collect them. Even contributors' experience on platforms such as iNaturalist or eBird can affect the quality of the data available to model biodiversity.
Anderson connected the current AI boom to an earlier stage in the development of species distribution modeling. Two decades ago, researchers faced their own proliferation of new techniques before the field developed stronger benchmarks, standards, and shared practices.
“History offers a lesson for the present,” Anderson suggested. Increasingly powerful tools still require biological expertise, careful evaluation and cross-disciplinary collaboration. Researchers need to understand not only what an algorithm can do, but also the assumptions behind the data and the biological problem the model is meant to solve.
That need for scrutiny returned during the morning Q&A, as Gillespie discussed the rapid proliferation of geospatial foundation models.
“Just because we can, it doesn’t mean we should,” she said, emphasizing the importance of asking when new tools can be trusted and whether people mistake ease of use for reliability.
Looking Below the Species Level
The symposium next turned to genomics and the assumptions researchers make when predicting how species will respond to environmental change.
Tereza Jezkova, associate professor at Miami University, challenged two common simplifications in species distribution modeling: treating a species as a single, homogeneous unit and assuming that its ecological niche remains static over time.
Drawing on her research with desert horned lizards, Jezkova showed that populations within a single species can be genetically distinct and adapted to very different environments. Lizards living in warmer desert regions, for example, can differ substantially from populations living in colder areas in their activity patterns, heat tolerance and genetic responses to environmental stress.
Those differences can become especially important when models are used to forecast the effects of climate change. A model that treats the species as a whole may suggest that warmer temperatures remain suitable, while a model that considers individual genetic lineages can reveal that some populations are poorly equipped for those changes.
The broader lesson, Jezkova explained, is that populations can adapt and ecological niches can evolve. Genomics provides researchers with another source of evidence for identifying those differences and evaluating predictions about both past and future distributions.
Diego Alvarado-Serrano, associate professor at Ohio University, expanded the discussion by asking how researchers can account for the dynamic nature of populations themselves. As environments change, species may move, adapt or become extinct. Which path they take can depend on genetic diversity, gene flow between populations, the rate of environmental change and other factors.
That complexity places limits on what predictive models can promise. Rather than treating forecasts as certainties, Alvarado-Serrano argued for thinking in terms of likely scenarios.
“We are not able to predict the future,” he said. “We are able to predict likely scenarios.”
Expanding the Meaning of Biodiversity Data
The final morning session widened the conversation beyond computational and genetic data.
Mary E. Blair, associate director of the Center for Biodiversity and Conservation at the American Museum of Natural History, discussed the growing importance of community-led biodiversity monitoring. Platforms such as eBird and iNaturalist have greatly expanded the amount of observational data available to researchers, while community-operated camera traps, acoustic monitoring, environmental DNA and Indigenous and local knowledge offer additional ways to understand how ecosystems change over time.
For Blair, however, communities should not simply be treated as sources of information. Community expertise can help researchers identify gaps and biases, shape research questions, and make results more useful for the people who ultimately rely on them.
That also raises questions about who controls biodiversity data and how they are used. Blair highlighted efforts to develop stronger approaches to Indigenous data sovereignty, including principles centered on collective benefit, community authority, transparency, responsibility and ethical use.
Popescu then returned to the question from the perspective of a practitioner. While much of the scientific conversation is focused on making data “AI ready,” he argued that biodiversity researchers should also ask how to make their data and work “decision ready.”
Species distribution models already inform conservation planning, policy, corporate sustainability, and other real-world decisions. As the models become increasingly sophisticated, researchers also need ways to translate that complexity into credible information that decision-makers can understand and use.
Choosing a Lane and Building Bridges
By the end of the morning, the symposium’s three themes had converged around a larger question: How can researchers incorporate AI, genomics, community science, and other forms of expertise without expecting any one person to master them all?
The answer repeatedly came back to collaboration.
“We as individuals can’t do everything, but we can choose our lane and walk a path together,” Blair said.
The symposium was designed to put that principle into practice. The afternoon agenda had participants divide into incubator groups and identify tangible ways to continue the conversations, including potential papers, research collaborations, grant proposals, working groups, student groups and future events. The groups then reconvened for a debrief and closing remarks.
Across the morning’s presentations, one theme remained consistent: emerging technologies alone will not determine the future of biodiversity research. Their value will depend on how researchers connect them with biological knowledge, community expertise, careful evaluation and the real-world decisions those tools are ultimately intended to support.
About the Program
The Master of Science in Biodiversity Data Analytics program at Columbia University equips a new generation of leaders with the data literacy, analytical tools, and interdisciplinary expertise to design evidence-based solutions that benefit both people and nature.
Designed for both working professionals and early-career changemakers, this online program allows students to learn from anywhere while gaining the skills to collect, analyze, and translate biodiversity data into meaningful action. Coursework prepares students to take on nature data applications across industries, from ESG finance to urban planning to environmental consulting, and culminates in a hands-on capstone with industry partners that provides practical experience, valuable networks, and the tools to make an immediate impact.
The application deadline for the M.S. in Biodiversity Data Analytics program is June 1. Learn more about the program here.
