Casey Morris, a lawyer in Northern Virginia, recently started checking Facebook again after a long break. Among the posts from her friends and her family, she noticed her strange tendencies.
“The caption says, 'Close your eyes 70% and see the magic.' And without squinting at all, it's very obviously an image of Jesus, but it's a little girl with vegetables and a tractor and It’s made up of some kind of distorted things,” she said.
That wasn't the only strange thing on Morris' feed. Similar photos with the same captions appeared repeatedly. So did a variety of more emotionally exploitative posts depicting disabled mothers and children in the mud, or smiling amputees, with captions asking for birthday wishes.
“Facebook has become a very strange, very creepy place for me,” Morris said.
To Morris, the subject matter, stylistic cues, and strange errors quickly made it clear that these images were fakes, the products of artificial intelligence.
It wasn't posted by anyone she knows or follows. Instead, Facebook suggests she might be interested in them – and they seem to be really popular.
“They have received thousands of reactions and thousands of comments. [from] “There are people out there who think they're being real, so they say happy birthday or say religious things in their comments,” she said.
Morris isn't the only Facebook user whose feed has started filling up with AI-generated spam. Reporters from the tech website 404 Media have tracked a spike in posts on Facebook, which is owned by Meta, in recent months that are clearly AI-generated. AI-generated images like this are also starting to appear on other social media sites such as Threads and LinkedIn, also owned by Meta.
spam and fraud
On Facebook, the platform's own algorithms often seem to encourage AI posts.
Researchers from Georgetown University and Stanford University examined more than 100 Facebook pages that routinely post AI content, sometimes dozens of times a day, and found that many were involved in scams and spam. It turns out.
Josh Goldstein, a researcher at Georgetown University and co-author, said, “AI can create anything you can imagine, from log cabins to grandmas with birthday cakes to children drawing pictures that seem too real.'' “I saw an image that was generated by This is the result of a preprint study that has not yet been peer-reviewed.
Goldstein and his co-authors also found that Facebook actively recommended some of this AI content to users' feeds. This can create a cycle where your posts get more engagement and are recommended to even more users. Some of the individual posts on the pages they analyzed amassed hundreds of thousands or even millions of interactions.
“These weren't just sporadic images that only a few people were interacting with. They were really getting a lot of attention,” Goldstein said.
Their analysis revealed that some of these pages are typical spam posts with links to websites that can collect advertising revenue. There are also scammers promoting AI-generated products that don't actually exist.
But many of the pages have no clear financial motive, according to Goldstein. It seems like they're just trying to attract viewers for unknown purposes.
“These are malicious pages that were trying to gain an audience and may later try to sell products, link to ad-filled websites, or even change the topic to be completely political. '' Goldstein said. “But I suspect that many of these pages were simply the creators who realized it was an effective tactic to get audience engagement.”
Clickbait has always been on social media. But in recent years, Facebook has doubled the amount of posts it recommends to users, trying to keep up with the changes in social media pioneered by TikTok. During a recent earnings call, Meta CEO Mark Zuckerberg told analysts that recommended posts now make up about 30% of users' feeds.
Transitioning from reality-based imagery to the macabre
At the same time, AI-generated content is now easier than ever for anyone to create. Together, these dynamics create a recipe for the proliferation of bizarre depictions of Jesus, disturbing birthday posts, and impossible architecture and crafts.
“It's like they're imitating all of the elements that made something go viral, but they're putting in the weirdest images I've ever seen,” says AI on Facebook. Brian Penny, a freelance writer who has been tracking the company for almost two years. He is part of a group dedicated to sharing and debunking AI images.
Penny has seen a shift from reality-based photos, such as last year's viral depiction of Pope Francis wearing an AI-generated fluffy coat, to something more macabre.
“We want our users to have a good experience, so we're working to reduce the spread of spammy or sensational content. That's why we're giving you control over what you see in your feed.” Meta a spokesperson told NPR in a statement.
The company will soon begin labeling AI-generated content created using industry-leading tools. TikTok last week began applying similar labels to some AI-generated posts on its platform.
On the other hand, the proliferation of AI spam has turned many people off.
Katrina McVeigh, of Grand Rapids, Michigan, said she had to dissuade her mother from buying obviously fake woodwork and other home decor items she saw on Facebook.
“She'll say, 'Wouldn't this be great for your daughter?'” McVeigh said. “And I think, 'But that's not reality.'”
Some Facebook users are so frustrated by spammy AI image recommendations that they're considering leaving the platform altogether.
“Do I have to look into all this to see that my cousin just went to the Sahara?” asked Los Angeles furniture designer Boris Zonka. “It's not worth it to me anymore.”
Many people NPR spoke to said they not only find AI spam on Facebook annoying, but also are concerned about the ubiquity of artificial images.
“It kind of reinforces people's distrust and makes it hard to see what's real,” said a North Carolina puppeteer who saw the AI image appear in a Facebook science group. Hobey Ford said, and claimed to be describing a new discovery. .
“And I think that's dangerous in our world right now,” he said.