Hurricane Helene made landfall in the United States from Florida on September 26 and hit many states and places in the southern United States. It has killed more than 230 people so far. Helene has become the hurricane with the highest death toll in the United States since Hurricane Katrina in 2005. At the same time, an AI-generated photo also caused an uproar on social media, even triggering a "red-blue showdown" between the two parties before the election next month.
What picture could be more serious than the impact of a hurricane? Amy Kremer, member of the Republican National Committee for Hurricane Rescue or African Vulture and co-founder of Women for Trump, tweeted on
Such photos that focus on a single victim, in which an innocent girl has to escape a disaster, are very easy to arouse people's sympathy, such as the "Hungry Sudan", where the skinny little girl is backed by a vulture staring closely at her, the "Napalm Girl" who evaded incendiary bombs during the Vietnam War, aroused the anti-war sentiments of countless young people, and the "big-eyed girl" studying also made the Hope Project deeply rooted in the hearts of the people. Therefore, this photo of "girl hugging puppy during hurricane" has been widely shared on social media.
But after looking carefully after the emotion, the little girl's arms and legs were extremely smooth and inconsistent with her skin color. Her hair also had a very strange texture, and her thumb joints also had strange bulges. Yes, this photo is an AI-generated image.
This AI trace is thicker, but it does not prevent it from gaining millions of views｜Picture source: X
After being pointed out as a "fake picture" by someone in the comment area, Kremer directly posted a message saying, "Yeah, I don't know where this photo comes from. To be honest, it doesn't matter where it comes from. It has been burned into my mind forever. Some people are more experienced than this photo shows." "It's a symbol of the trauma and pain that people are going through right now." In other words, she knew the image was fake, but she didn't care because her purpose was not to sympathize with the "AI girl," but to attack the ruling party's failure to respond to Hurricane Helene. Because similar AI photos have also been forwarded by far-right politician Laura Loomer, Republican KOLs, and Trump supporters Buzz Patterson and Juanita Broaddrick, with the captions "Our government has failed us again" or "The rulers have abandoned them." Each of these tweets has been shared in the millions.
Spreading is just to shirk responsibility｜Image source: X
AI fake pictures have become a political tool for mutual restraint between parties. Along with this photo, various conspiracy theories and fake news were spread, such as "The Federal Emergency Management Agency is preparing to confiscate the property of disaster victims" and "The authorities are controlling the weather in Republican-supporting states (the hurricane-stricken states are mostly Republican-supporting (red states)). Correspondingly, a picture of Trump braving floods to help residents was also circulated on Facebook, which was of course also synthesized by AI. The post was shared more than 160,000 times in two days.
Image source: PolitiFact
Trump himself also recently posted "SwiftiesforTrump" (Taylor fans who support Trump) on his X after Taylor Swift publicly expressed support for Democratic presidential candidate Harris. The attached pictures were all fake pictures generated by AI.
Image source:X
This is not the first time that Trump has posted fake AI pictures. As a candidate supported by Musk, it is impossible for Trump to not have the ability to distinguish AI content. After all, not long ago he also released a DeepFakeAI synthetic video of him and Musk dancing together. However, on fragmented social media platforms, politicians do not care whether the information is true or false. Whether it is a real photo or an AI-generated photo, as long as it has dissemination effect, building momentum is the primary purpose of political propaganda on social platforms. Perceptual impressions replace objective facts. The communication principle of "spreading rumors only opens your mouth, and refuting them will break your legs" is reverberating more violently in the AI era. Who is creating AI garbage? But tracing back to the source, who generated these fake AI content, and how can it be widely spread on various social media, and how can it be deceived one by one? This picture with the text "Handmade, thank you all for liking" received 870,000 likes and 35,000 comments on Facebook. However, it is a fake picture generated by AI. The original picture fed to the AI was of a wood carving artist, Michael Jones, and his real wood carving dog. Jones’ original post only received 1,063 likes and 110 comments, which was far from the fake AI post.
At the same time, similar pictures of different races, genders, and dog breeds are also circulating on Facebook. They all come from Jones’ picture. This means that anyone can use AI to create hundreds of fake and real images every hour, and then publish them immediately on social platforms. There are always a few that can gain good traffic and exposure. Among them, Facebook is the hardest hit area.
At the end of last year, a reporter from the technology website 404Media discovered that the number of posts on Meta’s Facebook that were obviously AI-generated content had surged, and it had also spread to Meta’s Threads.
Facebook’s “fake at first glance” image data is surprisingly good｜Image source: Facebook
Why is Facebook the hardest hit? The reason is simple: the algorithmization of the platform has caught up with the aging of users. In a recent earnings call, Meta CEO Zuckerberg told analysts that in order to keep up with changes in platforms such as TikTok, Facebook has doubled the number of posts recommended to user algorithms, with recommended posts now accounting for about 30% of user homepages. However, according to data from the OBERLO survey agency, Facebook users over the age of 25 account for 77.4%, of which users over the age of 35 account for 46.6%. For a large number of middle-aged and elderly people who were born in the pre-Internet era and may only start using social media and smartphones in middle age, it is difficult to understand what AI is, let alone identify AI content.
The paper "How Spam and Scammers Use AI-Generated Images to Grow Audiences on Facebook" published by the Stanford Internet Lab in August also found in tests that the images they generated with AI received hundreds of millions of impressions in total. AI-generated graphic posts are usually headline-grabbing and eye-catching pictures. For example, a child with a broken leg holding a sign saying "Happy Birthday to Me" can receive 70,000 likes and 3,000 comments. User comments can also show that they were unaware that the image was generated by AI.
Moreover, algorithm recommendation is like an upward spiral. Researchers have found that after clicking on a few AI-generated content posts, even if there are no followers or likes, more and more AI posts will be pushed to your homepage in the future. Some live users have commented on Reddit that "Facebook has become an endless MidjourneyAI photo wall."
A glance at AI can harvest a lot of traffic｜Image source: 404
In addition to deceiving traffic, AI fake pictures also want to defraud money. For example, like the wood-carved dog shown above, the post may include a link to "buy the same type of wood products." In fact, these products do not exist, or the audience may be directed to other advertising websites, such as the most popular wood-carved dog post above. When you click on the comment area, you will find a sales link for pet supplies pinned to the top. What’s even more paradoxical is that many photos may have been generated using MetaAI. One of the purposes of MetaAI is to make the photos look real. However, these photos deceived their users, and the contradiction has turned into a traffic cycle. AI-generated spam content affects not only the "aging" Facebook, but even Quora, the "American version of Zhihu" that was once famous for its high-quality text content, has to face the current situation of AI-generated content diluting the quality of community content. What’s even worse is that many AI-generated pictures are now firmly tied to reality. For example, Hurricane Helene has not yet subsided, but after the fake AI pictures are discovered, every user will question the authenticity of every photo, even if it is a real photo based on reality. After being diluted by a large amount of fake content, they will face the dilemma of the "real and fake Monkey King" proving their innocence. The Palestinian-Israeli conflict, the U.S. election, the Brazilian floods, Hurricane Helene...these are real natural and man-made disasters that occurred in real life. In a content community that is fragmented, brief, graphic, and text-based, and focuses on high timeliness, the real-life pictures of war reporters may not receive as many views as the fake pictures generated by AI. These contents are often targeted at unsuspecting older people, who contribute their own likes and retweets out of compassion, but unknowingly become the accomplices of the algorithm. The nourishment of these AIs is based on original but invisible artists. As a social platform with important responsibilities, such as Meta, it is caught in the self-contradictory war of AI: users use Meta's AI tools to produce false information and spam information and publish it to the Meta platform, while Meta's AI team needs to use technology to identify which pictures are real and which ones are generated by AI. The irony is intriguing - the more advanced the AI generation technology team is, the harder it will be for the review and identification team. At a time when spam has flooded the Internet, the birth of AI generation tools has undoubtedly once again accelerated the progress of Internet spam. If this challenge cannot be solved, it will not just be users and platforms, but everyone who will be "trapped in the system".