AI Is Fueling a New Kind of Political Propaganda

[00:00:00] Earlier this year, a video of a Legofied Donald Trump spread online. In this cartoonish video, Lego Trump launches a missile toward Iran, which is then followed by a wave of Iranian counterstrikes. Then a group of ultra-Orthodox Jews, also in Lego form, is shown fleeing an Iranian strike on Tel Aviv,

[00:00:26] while the country's Prime Minister, Lego Benjamin Netanyahu, runs to an underground bunker as American Lego soldiers are shown being brought home in coffins draped by American flags. To be clear, the video clip has nothing to do with the actual Lego group, the Danish toy maker that makes the popular plastic bricks children have been playing with since 1958.

[00:00:51] It was created by an Iranian entity called Explosive Media, which has said it produces content without outside influence. But the clip was circulated by the country's state broadcasters under the control of its military. And it didn't stop there. The videos were shared so widely, they even reached my American feet.

[00:01:17] As you can probably guess, the Lego videos were created using artificial intelligence tools. Since then, as America's midterm elections near, there's been a wave of AI-generated political content popping up on social media feeds in the U.S. as well. Today, we'll look at what happens to our sense of reality when political figures and bad actors use AI to fill our feeds with targeted, biased, and politically motivated content.

[00:01:47] I'm Nicole Nguyen, personal tech columnist at The Wall Street Journal. It's Saturday, August 8th. This is the second episode of our special series, AI and the Blurring of Reality, where we're examining how the rapid rise in generative AI could be fundamentally altering how we perceive our world. AI slop is a global phenomenon, and it's a local one too, one that landed this primary season.

[00:02:13] If you're in the U.S., you've probably come across a deepfake, an AI-made image or video featuring an often hyper-realistic impersonation. One notable deepfake showed James Tallarico, the Democratic Senate candidate from Texas, singing a rendition of My Favorite Things, praising transgender children. Boys in white dresses with blue satin sashes, Girls dust with hormones till they grow mustaches,

[00:02:42] Changing the gender of all your offspring. The video wasn't real. It was created by the Trump-linked political action committee Citizens for Sanity, in support of Tallarico's opponent, Republican Ken Paxton. Over in California, a fan-made video depicted Los Angeles mayoral hopeful Spencer Pratt as a heroic Batman. I'm just one man. I'm just one man.

[00:03:10] Battling a villainous, Thanos-like version of his political opponent, the incumbent mayor Karen Bass. Maybe Spencer Pratt should have played Iron Man, because in this fight, Thanos won. And if you haven't seen any of the superhero movies in the last decade, all you need to know is Karen Bass advanced to the November runoff.

[00:03:38] These are just a few examples of a growing wave of AI-generated clips in politics. Not all of them clearly fake, like that showdown between Batman and Thanos. So I study adversarial abuse, understanding how bad actors manipulate technology and social technology systems. Renee DiResta is an associate professor at Georgetown University. You may recall hearing from her in last week's episode.

[00:04:05] She says the flood of AI content in our feeds is causing people to be unsure of what's real and what is not. That can result in a perpetually skeptical mindset and make people especially wary of videos that contradict their existing beliefs. All of this creates an unprecedented opportunity for propaganda makers. You've worked so much with propaganda. Is this technology making propaganda more effective? Effective is an interesting question.

[00:04:35] It didn't invent propaganda, obviously. It just industrializes it. It makes it cheaper. It makes it faster to produce. And more sophisticated. For instance, it can be more directed to a specific audience. You can produce propaganda that has people who look like your target audience in it, whatever your target audience is, without having to go out and recruit for a speaker.

[00:05:02] In 2023, then-New York City Mayor Eric Adams used AI to send robocalls to constituents in languages other than English, including Mandarin and Yiddish. In the last Republican presidential primary, Florida Governor Ron DeSantis released deepfakes depicting Trump hugging and kissing Dr. Anthony Fauci.

[00:05:22] The following year, OpenAI reported that Russian, Chinese, and Iranian groups used the company's models to create fake content, including comments on social media sites in various languages. And just this March, there was a new study from USC that found human posters aren't even necessary. An army of AI agents can organize and coordinate propaganda campaigns on social media all by themselves. One of the lead scientists said,

[00:05:51] Our paper shows that this is not a future threat. It's already technically possible. You can make that avatar somebody who is speaking to them who looks like them. It can be more emotional, more visually persuasive. You can use an LLM to refine the language, to ask it, Hey, put this into the vernacular of the group that I want to connect with. You can do translation quite quickly, real-time translation in some instances.

[00:06:19] So, again, it's not that it is new. It's just that it makes it more efficient. Although it's easier to make, DiResta says ultimately propaganda is still limited by the way it's distributed. So you can make very, very excellent content. But the question is, can you land it in a place where people are going to share it? And I think one of the most interesting examples we've seen of this in the AI realm has really been the Lego videos coming out of Iran.

[00:06:48] Like the one you heard at the top of the show. DiResta says the reason the Lego clips are being spread so widely is because they're compelling. It is visually arresting. It is novel, right? It is very rare to see a government or a government-linked actor doing something quite so clever and so culturally relevant.

[00:07:10] They are touching on content in the U.S. political domestic discourse, Epstein, things like that, that is directly in the moment. Plus, the videos are capitalizing on the toys' popularity. And the message behind the Legos is pretty straightforward. They're very clearly anti-the U.S. political party in power, the political regime. And what they are doing is they are producing compelling content that is actually quite overt.

[00:07:39] Oftentimes, when we think of AI, we think of deception. We imagine being fooled by AI deepfakes. But DiResta says the Lego videos stand out because they are not that. We all know that Legos aren't real. When we see those videos, we're not being taken in. There's no active deception there. The bigger problem emerges when the political deepfakes look more realistic, especially as they come across our social media feeds.

[00:08:05] The Trump administration now routinely posts fake content. More on that after the break. Social media has become a staple of the American news diet.

[00:08:32] According to the Pew Research Center, over half of U.S. adults surveyed last year said that they get news from social media, a stat that's been roughly stable over the last few years. That figure goes up for Americans without a college degree, and not surprisingly, for younger adults as well. Pew has also found that more people are turning to social media for breaking news. So, as more of us get our news online, experts say generative AI content is remaking our geopolitical world.

[00:09:02] And fast, as bad actors have access to tools to make more AI content at a rapid clip. We spoke about this with Hani Farid, the Dartmouth professor and digital forensics expert. You may remember him from episode one. What are some notable examples of the way politically themed AI manipulated imagery has shifted that political conversation in a meaningful way? Yeah.

[00:09:30] So, on the politics side of things, we've seen quite dramatic impact of generative AI. So, for example, from the U.S. invasion to Venezuela, to the violence in Minneapolis, to the conflict in Iran and Gaza. As soon as these world events happen,

[00:09:55] we are inundated with generative AI, deep fakes, fake content purporting to be one thing or another. An explosion here, a person being arrested here, violence here. And we've seen it across the boards, where people either for ideological purposes, state-sponsored actors trying to sow civil unrest, people simply trying to monetize, are flooding the social media airwaves, if you will, with fake content. But we've also seen it coming from official channels.

[00:10:25] Farid is referring to the fake content the official White House social media accounts post to X and Truth Social. Like this one from last month. The symptoms can be relentless. Fortunately, I'm Dr. Trump is shown wearing a white lab coat, offering a cure for TDS, or Trump derangement syndrome. The video shows celebrities, including Rosie O'Donnell and Robert De Niro, outspoken critics of Trump, saying they're looking for cures.

[00:10:53] I really thought I was a lost cause. This was going to affect me for the rest of my life. Or another video, which shows House Minority Leader Hakeem Jeffries wearing a sombrero and mustache next to Senate Minority Leader Chuck Schumer. There's no way to sugarcoat it. Nobody likes Democrats anymore. We have no voters left. That video was reposted on President Trump's X account and was viewed over 42 million times.

[00:11:22] The AI videos that the White House posts span a wide spectrum. Some are clearly satirical and fake. But it's also posted images that could be mistaken as authentic. The Trump administration now routinely posts fake content of people who are being arrested. They change the face of somebody to make it look like they're crying, when in fact they were quite stoic. Fareed is talking about a digitally altered photo

[00:11:49] of an attorney arrested for protesting at a Minneapolis church. The White House's official X account posted the image showing the lawyer's face contorted with tears. But in the original photo, she was actually calm and expressionless. In an interview with the CBC, the attorney said of the experience, quote, They couldn't break me, and so they altered an image showing me broken. A few years ago, it might have taken a master of Photoshop hours to convincingly change that photo.

[00:12:19] But now, it's a 10-second adjustment using many widely available AI tools. At the time, the White House shrugged off the altered photo as a meme. And Deputy Communications Director, Kaylin Doerr posted on X that, quote, Enforcement of the law will continue. The memes will continue. So it's being used in these sort of clearly propaganda-like ways, but then in ways that are much more deceptive. And we're seeing that not just in the U.S.,

[00:12:48] we're seeing that around the world. When it comes to seeing truth in online media, it's not just about believing what you see, but also questioning what's real. What a bunch of malarkey. In 2024, a voice that sounded like then-President Joe Biden told New Hampshire constituents not to vote in the state's Democratic primary. Voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again.

[00:13:17] Your vote makes a difference. It was actually a fake robocall. It's uncertain if the deepfake audio changed voter behavior. But zooming out, Farid says a greater concern is our new perpetual state of doubt. I mean, how often have you heard, is that AI? Right? Because that's the whole other side of this coin, is that we are seeing horrific images coming out of the Middle East. We are seeing horrific abuses and police brutality and human rights violations,

[00:13:46] and everybody suddenly, well, how do I know if that's real? And so there's this double-edged sword that we are now dealing with, which is people are polluting the information ecosystem with fake content, and then denying the real content because the fake content itself exists. And suddenly, we're all sitting around being like, well, I don't know what's going on anymore in the world, and that's very messy. What does it mean to a society and a democracy when we don't have a shared sense of reality?

[00:14:15] Which is honestly sort of where we are now, and that is being exaggerated by the role of AI. I would say these are existential threats to society and democracy. We are no longer arguing about policy and the role of government and taxes and international aid. We're arguing whether one plus one is equal to two or is applesauce. I mean, that's the tenor of the conversation. And that is dangerous for a democracy where we can have disagreements. Disagreements are good. I like disagreements. I like having discussions.

[00:14:44] But I can't argue about whether one plus one is two or not. In other words, this isn't just about individual fake videos. It's about a systemic erosion of reality. When we can no longer trust our eyes and ears, even basic truths become debatable. We are moving from a world where our perception of reality is being fundamentally restructured by the very tools we once thought we controlled. So how do we stay ahead of the game?

[00:15:15] How do we fight back? What you're asking me to tell you is how do I defeat my adversary while my adversary is listening to this conversation? And that doesn't seem very advisable on my part, does it? Next week, we'll tell you about some ways you can identify AI-generated content. At least, the ones we can talk about. And that's it for the second installment of Tech News Briefing's special series AI and the Blurring of Reality.

[00:15:44] Today's show was produced by Julie Chang and mixed by Katie Ferguson. I'm your host, Nicole Nguyen. Michael LaValle wrote our theme music. Our supervising producer is Katie Ferguson. Our development producer is Aisha Al-Muslim. We had additional support from Wilson Rothman. And Chris Zinsley is the deputy editor of audio for The Wall Street Journal. We'll be back Monday morning with your TNB Tech Minute. Thanks for listening.