Several photographers have recently reported that the social media site Meta has been unfairly labeling their photos as “Made by AI.” These incidents have occurred in various contexts, such as photographs taken at a basketball match and a cricket tournament in the Indian Premier League. Interestingly, these labels appear when viewing photos in mobile applications but not in the web version.
Former White House photographer Pete Souza, who took the photo at the basketball game, tried to remove the mark but was unsuccessful. He speculated that the Meta algorithm flagged his photo because he had pre-cropped it using Adobe software and saved it in JPEG format. False positives seem to occur when photographers use generative AI tools like Adobe Generative Fill in their processing, even for minor adjustments.
Investigative Experiments and Findings
PetaPixel journalists confirmed this hypothesis by conducting an experiment. They used Generative Fill to remove a small speck from a photo, and the “Created by AI” mark appeared on Instagram. Continuing their experiment, they opened the photo in Photoshop, copied and pasted it into a document with a black background, saved it, and published it again on the social network. This time, the label did not appear.
Meta has responded to the feedback and plans to re-evaluate its approach to ensure that the labels accurately reflect the AI’s involvement in creating the image. “We rely on the industry standard indicators that other companies add to content with their tools, so we actively work with companies to improve our process and ensure that our approach to labeling is consistent with our intent,” said Meta spokesperson Kate McLaughlin in a statement to The Verge. Meta announced its plans to add such notes in April, while Adobe uses its own Content Credentials method for tagging AI content, reminds NIXSolutions.
Meta’s commitment to addressing these concerns is evident in their willingness to listen to user feedback. As the situation develops, we’ll keep you updated on any changes or improvements Meta implements to their labeling process. This ongoing dialogue between Meta and its users aims to refine the accuracy and transparency of AI content identification, aligning with industry standards and user expectations.