Tech Desk 
Jun 22, 2024
Meta plans to introduce a new labelling system for photos generated using artificial intelligence (AI) tools on its social platforms like Facebook, Instagram, and Threads. The company has frequently mentioned that these photos will bear a "Made with AI" label.
However, there have been instances where Meta incorrectly applied this label to photos that were not generated using AI tools. For example, a photo of the Kolkata Knight Riders winning the Indian Premier League Cricket tournament was mistakenly labelled. Such errors have raised questions about the accuracy of Meta's labelling system.
One notable case involved former White House photographer Pete Souza, who discovered that one of his photos was tagged with the "Made with AI" label. Souza speculated that changes in Adobe's cropping tool, which require flattening an image before saving as a JPEG, might have triggered Meta's algorithm to apply the label. He expressed frustration, noting that he had unchecked the option to include the label but it was added anyway.
Read more: Instagram live-stream gets private — New feature allows close friends-only broadcasts
Meta claims to use metadata from photos to determine whether the "Made with AI" label should be applied. The company aims to implement industry-leading tools that can detect invisible markers indicating AI generation, based on standards like C2PA and IPTC.
This approach includes labelling images from major tech companies such as Google, OpenAI, Microsoft, Adobe, Midjourney, and Shutterstock as they add metadata indicating AI usage.
Despite these efforts, Meta has not clarified exactly when its algorithm automatically applies the "Made with AI" label. Currently, there are no distinct labels to differentiate whether a photo was enhanced using AI tools or generated entirely by AI. This lack of clarity makes it difficult for users to understand the extent to which AI was involved in the creation of a photograph.
While some photos on Meta's platforms are indeed AI-generated, they have not always been correctly labelled by the company's algorithm, leading to confusion among users and photographers alike.
