A few months before the American elections, the limitless potential of artificial intelligence is worrying: can these images created from scratch influence the choice of the next president of the United States? L’Oeil du 20 heures interviewed YouTubers and bloggers behind disinformation operations.
This summer, a video shared by Elon Musk was viewed 135 million times and caused a scandal. Thanks to artificial intelligence, the perfectly imitated voice of Kamala Harris states some enormous things: “If you criticize anything I say, you are both sexist and racist. Joe told me: rule number one, hide your total incompetence.” When he posted this video on his social network X, formerly Twitter, Elon Musk did not specify that it was a fake. As a comment, he wrote a simple “it’s incredible”.
It was a certain Mr Reagan who created this video, by modifying a real campaign clip of Kamala Harris. This YouTuber, a fervent supporter of Trump, multiplies what he presented as parodies. He is trying to influence the American campaign from Bali, Indonesia. Chris Kohls, his real name, agreed to answer our questions. His goal: to counter what he calls the propaganda of traditional media.
The most important thing for me is to tell the truth in my videos. And that’s something that is sorely lacking in the American media today.
Chris Kohls, pro-Trump YouTuberat the Eye of 8pm
For him, it is not a question of misleading voters: everyone would be well aware that the video is fake. “Anyone who complains that this video is indistinguishable from a real one and that people will get it wrong is either very stupid or has no faith in the critical thinking of voters,” he assures us.
But is it really that easy to identify that this video is doctored? We showed it to American voters. “Joe Biden finally revealed his senility during the debate. Thanks Joe,” we hear in the excerpt we play to them. A fairly elderly first lady, met in front of the White House, believes it: “So Kamala Harris had information about Joe Biden’s senility… but she had never revealed it before!” When we point out to her that it is a fake video, she explains that she did not notice. “Maybe it’s my age.”she adds.
Other American tourists in Washington don’t believe it: “Someone used his voice to do this,” or even :”I think it’s all wrong.” A pro-Trump activist we met at a support rally doesn’t know what to think when we show him the video:
I don’t quite understand what it is. I don’t know. I can’t tell if this video was made to support her, or on the contrary against her.
A pro-Trump activistat the Eye of 8 p.m.
The video is clearly confusing. And the problem is that this kind of editing is far from isolated. Fake images of fans of the star Taylor Swift showing their support for Donald Trump have thus gone around the world. The candidate himself relayed this unexpected rally. With a delighted comment: “I accept!” It is impossible to know whether he believed it or not.
From a small town in Southern California, a pro-Trump blogger is trying to influence another crucial issue in the election: the African-American vote. So far, the vast majority of black voters support Kamala Harris. In an attempt to reverse the trend, Sharika Soal created an image of Trump surrounded by black women . It has been viewed 4 million times on X. In front of her computer, she tells us: “I use artificial intelligence software to create images and illustrate my tweets.”
Sharika Soal is not a computer professional. She uses very simple software. She just has to describe the image she wants to invent and artificial intelligence takes care of the rest. “I’m going to write ‘Donald Trump with beautiful, elegant black women’… and see what we get!” After several attempts, she is satisfied: “This one is very good, Donald Trump looks very happy, he always smiles in the photos.”
She then shared the images with her 180,000 followers on social media. Without specifying that they were fake. “This really pisses off left-wing people, and I love pissing them off, so I’m going to keep going. This isn’t a real picture, because it’s just an illustration!” she says. She adds:
I hope it will influence people, and allow them to understand that black women support Republicans. It’s a reality, it’s not a lie.
Sharika Soal, pro-Trump bloggerat the Eye of 8pm
These images are mostly created by Republicans, but Democratic Internet users are also getting in on the act. Like this photo, which shows Donald Trump with a minor, alongside Jeffrey Epstein, the billionaire accused of sex crimes.
We wanted to test for ourselves the possibilities offered by the software available online. Some have put in safeguards, like ChatGPT. When we try to create an image of Trump, the software refuses: “I can’t create images of specific public figures, like Donald Trump.”
The software available on X, Grok, imposes far fewer limits. It is therefore possible to support all the rumors: Joe Biden dying, Kamala Harris addicted to the bottle, or making political figures kiss.
Faced with this flood of misleading photos, social networks have become alarmed. Last February, Facebook, Tiktok, Google and X even put their rivalries aside to commit to preventing AI from disrupting elections.
But is the promise really being kept? In Washington, Tim Harper is an independent expert at the Center for Democracy and Technology. He monitors the impact of digital technology on democracy. He shows us on Facebook an AI-generated image of Kamala Harris in a skimpy outfit. “This image does not have any indication to warn that it has been manipulated and is misleading, when this should be mentioned”, he explains.
Those spreading this false image claim that Kamala Harris is transgender. According to Tim Harper, the platforms are not giving themselves sufficient means to apply their own rules.
There are signs that social networks have been investing less in moderating their content over the past four years.
Tim Harper, researcher at the Center for Democracy and Technologyat the Eye of 8 p.m.
He adds: “For example, platforms have fired employees who were responsible for moderation.” Social network X seems to be the champion of double talk. Here is what its terms of use mention: “You are not allowed to share artificial or manipulated content that may deceive or mislead other people.” However, it is on X that we find the vast majority of fake images today. Contacted, the social network did not respond to us.
Usually, these manipulations are quickly identified by the media… because they have visible flaws. But what will happen as the elections approach, when there will be no time to deny them? Images of violence near a polling station or electoral fraud could thus lead to the belief that the election was stolen. In a country still traumatized by the attack on the Capitol four years ago, following false accusations of fraud, the threat is taken very seriously.
Among our sources:
Tech giants agree on AI
The Center for Democracy and Technology
Rules for using X