Images of children dressed as cabbages, amateur artwork and homemade cakes are racking up millions of views on Facebook, but experts warn the content might not be as innocent as it seems.
Among the most bizarre creations are images depicting Jesus made from prawns or other shellfish, which bemused users have dubbed “Shrimp Jesus.”
Other images purport to be amateur art, including elaborate rock and sand sculptures or breathtaking oil paintings that wouldn’t look out of place in the Louvre.
But while the weird and wonderful creations might appear to result from many hours of meticulous concentration, they were created in seconds with a computer.
Researchers at Stanford Internet Observatory in California say the wave of AI-generated content appears to be an attempt to make money by attracting clicks and shares from social media users.
They analysed 120 Facebook pages that each posted at least 50 AI-generated images and collectively received hundreds of millions of engagements.
The researchers found that the spam pages were designed as clickbait to encourage users to visit suspicious websites or reveal their personal information.
Hacked pages were among those posting AI content, the study found.
Co-author Renee DiResta told AAP that other pages appeared to be building large follower counts to increase their resale value.
Many Facebook users are unaware the images are not authentic.
“Amen and congratulations young man,” says one comment under an AI-generated image of a child sitting on the lap of an abstract sculpture of Jesus made from potatoes.
Another comment on the same post reads: “So amazing. God bless you.”
Ms. DiResta says religious themes appear popular with AI spammers and scammers because they appeal to social media users worldwide.
The posts also use emotive captions to encourage likes and shares.
“Made it with my own hands, but no one appreciates this,” reads a typical caption.
Another common caption is, “This is my first cake! I will be glad for your marks.”
Dan Halpin, the chief executive of cyber investigations firm Cybertrace, says the intention of the posts isn’t always straightforward.
“Such tactics are typically part of broader strategies aimed at either monetisation through clicks and affiliate marketing, or audience building for potential future use, potentially for scam targeting,” he said.
AAP attempted to contact four of Facebook’s most prolific AI-generated content pages but did not receive any responses.
None of the pages have told users the images they’re posting aren’t real.
Some images include flaws that reveal their digital origins, such as malformed hands or indecipherable words on birthday cakes.
Jeannie Paterson from the University of Melbourne’s Centre for AI and Digital Ethics says that rapid advances in generative AI have made it tougher to determine if an image is authentic.
“On a casual viewing, many are indistinguishable,” Professor Paterson said.
“AI-generated images have improved dramatically and seem likely to continue to do so.”
Prof. Paterson says publishers have an ethical obligation to disclose when images are AI-generated.
She says there may also be a legal obligation to do so, such as when content could mislead consumers.
Meta has plans to label video, audio, and image posts on Facebook as “Made with AI” when it detects them.
Mr. Halpin says people should ignore AI-generated image spam or report it to Facebook.
“Sharing or interacting with AI-generated spam might accidentally help these dodgy networks and could even put your personal data at risk,” he said.
“Reporting these posts to Facebook can help clean up the site. It’s generally a good idea to steer clear of any content that looks suspicious or doesn’t have a clear, proper purpose.”
