AI-enhanced bird photos threaten iNaturalist research data

A false sighting of a red-winged blackbird in central Brazil turned out to be an epaulet oriole whose photo had been run through an AI 'make it look better' tool, and the enhancement quietly grafted features from the North American species onto the South American one. The Guardian reports that this is the kind of glitch scientists are now finding across birdwatching forums, and the reason they are worried has less to do with hoaxes than with the everyday photo-improvement habit birders picked up from their phones.

The warning comes from Alexander Lees, an ecologist at Manchester Metropolitan University, in a commentary that identified hundreds of fake or altered images already sitting inside species-recording databases. His concern, as reported by PetaPixel summarising the same piece, is not that people are faking rarities on purpose. It is that ordinary photographers use AI to clean up a shot, and the model reaches for training-data features from whichever bird it thinks the picture should look like. 'The idea that we could maybe use those photos to help us understand where species are in space and time is very difficult,' Lees said.

Why that matters beyond birding: platforms like iNaturalist and the Macaulay Library are load-bearing inputs into real ecological research on species range and habitat. iNaturalist's own numbers, as carried by the Taipei Times, give a sense of the detection gap. Only around 1,400 of the platform's more than 610 million images have been flagged for AI use so far. Tony Iwane, iNaturalist's director of community support, put the stakes plainly in the same piece: 'The more we know about where species are, the better informed we can be as conservationists. But the information needs to be accurate.'

The honest caveat is that the reporting does not tell you how many of those altered images actually made it into a published study or triggered a rarity alert before being caught, or how the split between outright generation and routine enhancement breaks down. It also does not say whether Macaulay Library or eBird are planning policy changes of their own. Take the numbers as reported, not as a full accounting.

The forward-looking thread worth pulling is provenance. Any dataset built from user-uploaded photos is now competing with a default consumer workflow that silently rewrites pixels, and the platforms that ship verifiable capture metadata, raw-file requirements for rare-species records, or their own AI-detection classifiers are the ones that will still be trusted as scientific inputs a couple of years from now.