Meta’s artificial intelligence (AI) content detection tool on Instagram has been spotted with genuine images that have a ‘Made by AI’ label. One such incorrectly labelled post was from the official Instagram account of Kolkata Knight Riders, which recently won the Indian Premier League (IPL) 2024 cricket tournament. One of the images posted by the account, in which the team is holding the trophy, was labelled as AI-generated by the platform. Multiple photographers on the social media platform have faced the same issue.
In February, Meta said it was working on introducing an AI-generated content detection feature that would protect users from misinformation and flag instances of deepfakes (AI-generated, or digitally altered, images and videos made to resemble another person, location, or event). The feature recently went live on Instagram, and it appears to be labeling real photos as AI-generated content as well. At the moment, these labels are only visible in the iOS and Android apps and not on the web.
Photo of Kolkata Knight Rider, tagged as ‘Created by AI’
Photo credit: Instagram/kkriders
While KKR’s photo is one of the most notable examples of this error, but there have been many other such mislabeling mentioned by users. One of them is former White House photographer Pete Souza who posted a photo of an old basketball game.
After the wrong label was added, edited the caption to write, “I don’t understand why Instagram is using the “made with AI” on my post. There is no AI in my photos.” He also emphasized that he couldn’t uncheck the label because the platform kept adding it.
Frustrated users have also begun flooding Threads, Meta’s text-based social media platform, tagging Instagram head Adam Mosseri to bring attention to the issue. One user said“Not a single photographer or artist on Facebook and Instagram has any idea what triggers the “Made with AI” label. Even though Mosseri made it clear that they only read C2PA labels, no one knows how to avoid it.”
Previously, Nick Clegg, President of Global Affairs at Meta, had said that the company is working with “industry partners to adopt common technical standards that indicate when a piece of content has been created using AI.” He also claimed that the detection tool can correctly label images from Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock.
However, the implementation of the feature seems to be flawed. In a reportPetaPixel found that removing even a small spot in an image using Adobe’s AI-powered Generative Fill would give images the “Made by AI” tag. However, non-AI tools like the Spot Healing Brush tool or Clone Stamp tool did not add the tag, despite the same results.
The publication also found that when an image previously labeled as AI-generated was reloaded and saved in Photoshop after being copied and pasted into a black document, the AI label did not appear.
Meta spokesperson Kate McLaughlin told The Verge said the company is now taking recent user feedback into account and evaluating its approach. “We rely on industry-standard indicators that other companies include in content in their tools, so we’re actively working with these companies to improve the process so that our labeling approach aligns with our intent,” McLaughlin was quoted as saying by the publication.
