If you've spent any time on Facebook within the last six months, you’ll have noticed photorealistic images which might be too good to be true: children with paintings that seem like the work of skilled artists or majestic log cabin interiors that resemble the The stuff is Airbnb dreams.
Others, equivalent to depictions of Jesus constituted of crustaceans, are simply bizarre.
Like the AI image of the Pope in a puffer jacket These AI-generated images, which went viral in May 2023, have gotten increasingly common and popular on social media platforms. Even though lots of them border on the surreal, they are sometimes used to lure regular users into engagement.
Our team of researchers from the Stanford Internet Observatory and the Center for Security and Emerging Technology at Georgetown University examined over 100 Facebook pages which posted large amounts of AI-generated content. We published the ends in March 2024 pre-printed paperThis signifies that the outcomes haven’t yet been peer reviewed.
We examined image patterns, discovered evidence of coordination between some sites, and tried to discover the posters' likely targets.
Site operators apparently posted pictures of AI-generated babies, kitchens or birthday cakes for various reasons.
There were content creators who innocently tried to grow their following with synthetic content; Scammers use pages stolen from small businesses to advertise products that don't appear to exist; and spammers share AI-generated images of animals while directing users to advertising-filled web sites, allowing owners to earn promoting revenue without creating quality content.
Our results suggest that these AI-generated images attract users – and Facebook's suggestion algorithm may promote these posts organically.
Generative AI meets fraud and spam
Internet spammers and scammers are nothing recent.
They have been using them for greater than twenty years unsolicited mass email Promote pyramid schemes. You aimed Seniors while posing as a Medicare representative or computer technician.
On social media, profiteers have used clickbait articles to direct users to ad-laden web sites. Remember the 2016 US presidential election, when Macedonian teenagers shared sensational political memes on Facebook and others Advertising revenue collected after users visit the URLs they posted. The teenagers didn't care who won the election. They just wanted it earn money.
In the early 2010s, spammers grabbed people's attention with ads that promised that anyone with “an odd trick.”
AI-generated content has develop into one other “weird trick.”
It is visually appealing and cheap to provide, allowing scammers and spammers to generate large volumes of engaging posts. Some of the sites we monitored had dozens of unique images uploaded day by day. In doing so, they followed Metas own advice for page builders. The company suggests that posting continuously helps creators get the form of algorithmic coverage that ends in their content appearing in what was formerly referred to as the “feed.”news.”
Much of the content continues to be clickbait in some ways: Shrimp Jesus makes people stop of their tracks and encourage sharing simply because it's so bizarre.
Many users respond by liking the post or leaving a comment. This signals to algorithmic curators that perhaps the content must be pushed into much more people's feeds.
Some of the more established spammers we've observed have likely recognized this and improved their engagement by switching from publishing URLs to publishing AI-generated images. They then commented on the post with the AI-generated images with the URLs of the ad-supported content farms that users should click on.
But more extraordinary YouTubers also used interaction with AI-generated images without obviously violating the platform's guidelines.
Rate “my” work!
When we looked up the post headings on CrowdTangle – a social media monitoring platform by Meta and set for sunset in August – we discovered it was them “Copypasta” subtitleswhich suggests they were repeated in all posts.
Some of the copypasta captions enticed interaction by directly asking users to, for instance, rate a “painting” by a first-time artist – even when the image was generated by AI – or wish an older person a comfortable birthday. Facebook users often responded to AI-generated images with encouraging comments and congratulations
Algorithms push AI-generated content
Our research noticeably modified our own Facebook feeds: Within days of visiting the pages—and without commenting, liking, or following the fabric—Facebook's algorithm beneficial tons of other AI-generated content.
Interestingly, the incontrovertible fact that we viewed groups of AI-generated miniature cow pages, for instance, didn’t end in a short-term increase in recommendations for pages focused on real miniature cows, full-sized cows, or other cattle. Rather, the algorithm beneficial pages on a variety of topics and topics, but that they had one thing in common: they contained AI-generated images.
In 2022 the technology website Verge detailed an internal Facebook memo about proposed changes to the corporate's algorithm.
According to the memo, the algorithm would develop into a “discovery engine,” allowing users to be exposed to posts from people and pages they haven’t explicitly looked for, much like TikTok’s “For You” page.
We have Facebook's own “Widely followed content reports“, which lists the preferred content, domains, links, pages and posts on the platform per quarter.
It showed that the proportion of content that users saw from Pages and other people they don't follow increased steadily between 2021 and 2023. Changes to the algorithm have created more room for AI-generated content to be beneficial organically without prior interaction – which is probably our explanation and people of other users.
“This post was dropped at you by AI”
Because Meta doesn’t currently label AI-generated content by default, we now have sometimes observed users warning others about scams or spamming AI content with infographics.
However, Meta appears to pay attention to the potential problems if AI-generated content slips unnoticed into the data environment. The company published several Announcements about learn how to handle AI-generated content.
In May 2024, Facebook will begin applying the “Made with AI” label to content that it reliably recognizes as synthetic.
But the devil is in the main points. How accurate will the detection models be? What AI-generated content will slip through? What content is flagged inappropriately? And what is going to the general public do? Production of such labels?
While our work focused on Facebook spam and fraud, there are broader implications.
reporters have written about it AI-generated videos are geared toward children on YouTube and influencers on TikTok using generative AI make a profit.
Social media platforms must grapple with learn how to handle AI-generated content; It's entirely possible that user engagement will decline when online worlds are stuffed with artificially generated posts, images, and videos.
Shrimp Jesus might be an obvious fake. But the challenge is judging what’s real just heats up.
