Meta is Incorrectly Marking Real Photos as ‘Made by AI’

Multiple photographers are raising concerns that Meta is erroneously labeling genuine photographs as "Made by AI," reports TechCrunch. This issue has sparked widespread frustration among photographers who feel their authentic work is being misrepresented by Meta's AI detection algorithm.

Examples of Mislabeling

Several photographers have shared instances where their real photos have been wrongly tagged. Notably, former White House photographer Pete Souza discovered that a photo he took of a basketball game was labeled as AI-generated by Meta. Similarly, an Instagram photo of the Kolkata Knight Riders celebrating their Indian Premier League Cricket tournament victory was incorrectly tagged with the same label. Interestingly, these labels appear only on mobile devices, not on the web.

Image showing how Meta incorrectly labels genuine photos as AI-generated.

Souza attempted to remove the label but was unsuccessful. He speculates that using Adobe's cropping tool and flattening images before saving them as JPEGs may be triggering Meta's algorithm to apply the AI label.

Issues with Minor Edits

Meta's algorithm also seems to incorrectly label real photos that have undergone minimal edits using generative AI tools. According to PetaPixel, even slight modifications, like using Adobe's Generative Fill to remove a small object, result in the photo being marked as AI-generated. However, when PetaPixel saved the edited file after copying and pasting it into a black document, Meta did not apply the label.

Photographers are expressing their frustration over such minor edits being tagged as AI-generated. Noah Kalina, a photographer, voiced his concerns on Threads, stating, "If ‘retouched’ photos are ‘Made with AI’ then that term effectively has no meaning. They might as well auto-tag every photograph ‘Not a True Representation of Reality’ if they are serious about protecting people."

Meta's Response

In response to the backlash, Meta spokesperson Kate McLaughlin stated that the company is "taking into account recent feedback" and is evaluating its approach to ensure that the labels accurately reflect the amount of AI used in an image. She added, "We rely on industry standard indicators that other companies include in content from their tools, so we’re actively working with these companies to improve the process so our labeling approach matches our intent."

Background on Meta's AI Labeling

In February, Meta announced its intention to add "Made with AI" labels to photos across Facebook, Instagram, and Threads ahead of this year's election season. The goal was to clearly indicate AI-generated content, particularly from tools by Google, OpenAI, Microsoft, Adobe, Midjourney, and Shutterstock.

Meta has not disclosed the specific criteria that trigger the "Made with AI" label. However, many of the companies involved are incorporating metadata into image files to signify the use of AI tools. Adobe, for example, introduced its Content Credentials system last year to add information about an image's origins into the metadata, a practice Meta uses to identify AI-generated photos.

Conclusion

As Meta continues to refine its AI detection algorithm, it is crucial for the company to address these mislabeling issues. Photographers rely on accurate representation of their work, and the current system's flaws are undermining their efforts. Meta's collaboration with other companies to improve the process is a positive step, but until a more precise method is implemented, photographers may continue to face challenges with the mislabeling of their genuine photos.