Researchers Warn of AI-Generated Fakes Infiltrating Citizen Science Biodiversity Records
Key Takeaways
- ▸Generative AI tools like Google Gemini can create convincing fake species photos that deceive expert observers on citizen science platforms
- ▸At least two documented cases of AI-generated fakes have already appeared on iNaturalist and birding platforms in real-world use
- ▸The threat extends beyond platform integrity to scientific research itself, as peer-reviewed papers increasingly rely on citizen science data without robust verification
Summary
A new research paper published in Nature Ecology & Evolution reveals that generative AI tools, including Google Gemini, are being used to create convincing fake species photographs that are infiltrating citizen science platforms like iNaturalist and eBird. Researchers including Kris Anderson and Alexander Lees have documented multiple instances of AI-generated creature images and manipulated photos of rare birds that could deceive both citizen scientists and potentially contaminate peer-reviewed research.
The problem has already begun manifesting in practice. Anderson discovered a fabricated dead leaf mantis on iNaturalist while Lees uncovered a fake willow tit photo on a Scottish birding site—both created so convincingly that they passed initial scrutiny. The researchers warn that as generative AI continues to improve, the volume and sophistication of fake observations could surge, posing significant risks to environmental assessments, scientific papers, and the credibility of biodiversity research that increasingly relies on citizen-contributed data.
The findings underscore a growing vulnerability in scientific infrastructure that has become dependent on crowdsourced observations. The ease with which modern AI tools can fabricate or manipulate images of organisms—particularly rare species that few people have seen—creates a verification crisis for platforms and researchers who lack adequate detection mechanisms. Experts are calling for improved screening protocols and verification processes across citizen science platforms before AI-generated fraudulent observations become commonplace in the scientific record.
- Researchers are documenting the problem early as a vulnerability warning before AI-generated biodiversity observations become widespread and difficult to detect
- Citizen science platforms lack adequate screening mechanisms to filter AI-generated content at scale
Editorial Opinion
This research exposes a critical blind spot in the scientific community's embrace of generative AI and crowdsourced data. While citizen science has democratized biodiversity research and generated invaluable datasets, the discovery of AI-generated fakes reveals that these platforms were built with an implicit assumption of human-generated content and human oversight. As AI tools continue to improve in fidelity, the research community must act now to develop robust verification standards—not after fraudulent data has contaminated environmental policies and conservation decisions.
