Many infectious diseases begin with a spillover: a pathogen circulating in animals finds its way into humans. Changes in land use, agriculture and climate can increase contact between people and wildlife, while global connectivity can help infections spread once they emerge. A recent Nature feature on AI and zoonotic disease surveillance asks whether artificial intelligence could help us identify some of these threats earlier.
One possibility starts with virus discovery. In 2024, researchers developed LucaProt, a deep-learning model designed to detect highly divergent RNA viruses in genetic data. Applied to 10,487 metatranscriptomic datasets, it identified 161,979 potential RNA virus species, substantially expanding the known RNA virosphere. Discovering a virus does not mean that it threatens humans. Other machine-learning approaches are therefore being developed to help prioritize animal viruses for further investigation according to their potential to infect people.
AI can also help researchers make sense of information once disease is circulating. Systems described in the Nature article combine sources ranging from public-health reports and news to ecological, climate and travel data, looking for patterns that could provide earlier warning of emerging outbreaks.
This approach is also being explored through a One Health perspective. Around Uganda’s Bwindi Impenetrable National Park, researchers have combined disease testing in livestock with existing gorilla-monitoring data. These data are being used to train predictive models intended as early-warning systems for diseases that could affect people, livestock and gorillas.
At present, many of these applications remain research tools, and their usefulness depends on the quality and geographical coverage of the data available. Large gaps in pathogen sampling can distort global risk assessments, while local knowledge can be lost if communities are excluded from how surveillance systems are designed.
The opportunity, then, is not to replace epidemiologists or traditional surveillance with algorithms. It is to use AI alongside better field data, local expertise and human judgement to detect signals that might otherwise be missed. Therefore, AI has the potential to be a tool for learning more about emerging threats before they become emergencies.
Coordinator at la Verneda-Sant Martí Learning Community and adjunct professor at the University of Barcelona