Researchers warn AI could be undermining wildlife photography

Wildlife photographers are being urged to think carefully before using generative AI to enhance their images, after researchers warned that AI-edited and AI-generated photographs could compromise valuable biodiversity data.

In a commentary published on science journal Nature, an international team of researchers says popular citizen science platforms, including iNaturalist and the Macaulay Library, are already seeing submissions of AI-generated and heavily AI-edited wildlife images that have the potential to mislead both scientists and automated identification systems.

These platforms collectively host hundreds of millions of photographs submitted by the public, and the images are increasingly used by researchers to track where species occur, monitor changes in behaviour and habitat, and study the impacts of climate change.

According to researchers, the growing availability of AI image generation and editing tools introduces two key risks: completely fabricated wildlife photographs submitted as genuine observations, and authentic photographs that have been altered enough for AI to unintentionally change important identifying features.

One example highlighted in the paper involved a photograph of an epaulet oriole taken in Brazil. After the image was processed using Google's Gemini AI image editor with the simple prompt to "make this look better", the software altered the bird so extensively that it closely resembled a red-winged blackbird – a North American species that would represent a highly unusual sighting in Brazil.

The researchers recreated the edit themselves, demonstrating how easily an AI enhancement could transform one species into another without any intention to deceive.

Lead author Dr Alexander Lees, an ecologist at Manchester Metropolitan University who authored the journal article, said the issue is becoming increasingly common as AI-powered editing tools are integrated into everyday photo workflows.

“My experience of looking at Facebook these days is that a huge volume of wildlife photos now are simply AI-generated imagery,” he said.

“The idea that we could maybe use those photos to help us understand where species are in space and time is very difficult.”

Even minor AI edits designed to improve an image can alter scientifically important details such as plumage, markings or colouration. Those subtle changes may affect species identification and reduce the value of photographs submitted to citizen science databases.

There are also longer-term implications. AI-generated or heavily manipulated photographs could eventually be incorporated into the training datasets used by computer vision systems that help identify wildlife, potentially reducing the accuracy of those tools over time.

According to researchers, citizen science platforms are already responding. iNaturalist has introduced options that allow users to flag observations containing fully AI-generated images or photographs that have been manipulated to inaccurately represent the organism or scene. Flagged observations are downgraded so they are excluded from research datasets shared with organisations such as the Global Biodiversity Information Facility.

Although only a small fraction of the hundreds of millions of images hosted by iNaturalist have been flagged so far, the authors say the true scale of the problem remains unknown, as many manipulated images may go unnoticed.