Québec bird observers averaged seven correct answers in a synthetic-image exercise
La Presse and Québec Oiseaux showed roughly ten team members ten wildlife images, six generated with AI. Participants correctly classified an average of seven, and only one identified every image correctly. This small, non-scientific exercise does not measure the public’s ability to detect synthetic pictures. Interviews accompanying it describe concern about fabricated animal images entering research datasets, while stressing that a detection tool’s negative result does not prove a photograph is genuine.
Artificial Intelligence··Night
The score is a small-group result
In Québec Oiseaux’s small classification exercise with La Presse, the group averaged seven correct choices among ten pictures. The bird-observation organization supplied roughly ten team members. Of the ten wildlife images they assessed, six were synthetic. One participant alone sorted the entire set correctly. That person described intermediate experience observing wildlife and said the choices involved repeated changes of mind. This was a non-scientific exercise; its scores cannot establish how well the wider public or professional researchers detect generated imagery.[1]
Research collections are part of the concern
The concern extends to where wildlife pictures end up. In the interviews accompanying the exercise, Montréal Insectarium director Maxim Larrivée warned about synthetic images entering collections used to train research models. Jean-Simon Bégin, who photographs wildlife, said Facebook posts showing invented animals were drawing thousands of likes. Recognition difficulties sit alongside concern about the authenticity of pictures available for research; the exercise did not measure contamination in a scientific dataset.[1]
A detector’s silence is not authentication
Detection software also has limits. La Presse notes that a tool failing to flag a fabricated image does not prove the photograph is authentic. Larrivée advocated clearly labeling generated images. Both points concern how pictures are identified and presented, rather than a claim that automated screening can settle every doubtful wildlife photograph.[1]
