You’ve probably come across posts on Facebook lately asking you to like your handsome brothers fighting for Ukraine or your grandmother who tied up the tankand baked the world’s largest cake — and if you are familiar with the art of neural networks, you can easily recognize a fake. However, tens of thousands of users actually believe in the photo’s reality, promoting previously unknown groups with their likes.
Here is one example, and as you can see, Facebook did not in any way label it as false or AI-generated:
In another post, there is a mark on the photo, but not from Facebook and is so small that it hardly catches anyone’s eye:
For people familiar with the art of neural networks, the images are obviously generated. However, for thousands of users Facebook users who have liked or reposted the post, the photos seem to be real
Another example — a post about a robot that allegedly gathered thousands of spectators for its first concert. According to the text, he «uses an Apple M3 Pro, given the recent partnership between Apple and OpenAI», and invited Mark Zuckerberg and Leonardo DiCaprio» to his performance. The images themselves, despite being extremely photorealistic, also show signs of artificial intelligence (for those who are not satisfied with the frankly meaningless text and the fact that such technical innovations do not exist in the world, unfortunately): in all four photos, the robot has different hairstyles, clothes, microphone, and even shoulder sizes, etc Despite the fact that in the end Facebook has managed to mark the post as falseIn two weeks, it garnered hundreds of likes and comments and thousands of reposts (and several dubious Russian outlets reprinted it as news)
In February, Meta announced that it would start adding «Made with AI» labels to photos uploaded to Facebook, Instagram, and Threads, but as we can see from the previous examples, something went wrong.
More: tools Meta, ignoring the generated content, mistakenly label real photos as AI. For example, this photo of the Indian Premier League cricket tournament on Instagram:
Also Meta recently tagged the photo of a basketball game taken by former White House photographer Pete Souza, as being created by AI (by the way, the marks are visible only in mobile versions of the apps). Souza himself said that he tried to remove the tag but failed.
According to PetaPixel, Instagram’s algorithms canimages as generated when photographers use generative AI tools such as Adobe’s Generative Fill to remove even the smallest objects. The publication tested this for itself by using Photoshop’s Generative Fill tool to remove a spot from an image, which Meta then labeled as AI-generated on Instagram. Surprisingly, the «Made with AI» label was not added when PetaPixel loaded the file back into Photoshop and then saved it after copying and pasting it into a draft document.
It seems that the AI identification process on Instagram and Facebook works in different ways, but both are not perfect. Given that Meta has long had problems with content moderation that they have not solved, a similar situation may arise with artificial intelligence. But don’t rush to remove your parents from Facebook (other social networks are unlikely to perform better), just take the time to learn how to recognize fake or artificial content