'If You Can Keep It': A.I. And Our Democracy

0:07Most Americans now get at least some of their news on social media, but these platforms are increasingly being overrun by AI-powered bots. One estimate found that around 20 percent of accounts on social media are bots, while another study found that for controversial topics, nearly half of all posts could be bot generated. Now, those numbers are still hard to verify. But experts are sounding the alarm that AI bots used in large-scale social media campaigns could threaten our democracy and the upcoming presidential election. Now, we know that social media content can influence how we view and internalize politics, but what happens when bad actors fan the flames with AI-generated photos or bot campaigns and misinformation?

0:53I'm Jen White. And I'm Todd Zwilich. You're listening to the 1A Podcast for this week's installment of If You Can Keep It, How Social Media and AI Can affect our elections and our democracy. That's right after this short break.

1:10Let's get right into the conversation with our guests. Joining us here in studio is Imran Ahmed. He's the founder and CEO of the Center for Countering Digital Hate. That's a nonprofit that works to stop the spread of online hate and disinformation. Imran, welcome back to the program. Good to be here. Also with us is Tiffany Hsu, a technology reporter for The New York Times, covering the information ecosystem, foreign influence, and disinformation. Tiffany, thanks for joining us. Hello, thanks for having me. And Inga Trotik is a research professor focused on political impacts of emerging technologies at Florida International University.

1:46Inga, welcome to the program. Thank you for having me. Inga, last month, a group of tech and social science experts published a warning in the peer-reviewed journal Science about the potential harms of AI bot swarms. So let's just start with the definition first. What do we mean when we say AI bot? Well, a bot is always, like, it's short for robot. It's a software application which is programmed, and it usually performs, like, automated, like, definitely automated, but usually repetitive tasks. So AI can also do that. It's an artificial intelligence bot then. The alarm that was sounded in a recent study that you're quoting is that this group of really acclaimed and high-profile researchers have been saying it's agentic AI swarms.

2:35And agentic AI is scary because it really mimics humans and we don't have this stupid misinformation anymore. We have really sophisticated misinformation that will engage with humans around the world, according to their study, and, yeah, probably persuade some humans in ways that are deceptive then. So, Tiffany, in your reporting, how are you seeing AI fuel disinformation online? It's absolutely everywhere. In the last six weeks of news events, it's come up in nearly every single one with Maduro's capture in Venezuela, in the unrest in Minnesota, in the release of the Epstein files. AI is absolutely unavoidable.

3:22And by that, I mean AI-generated images, AI-generated text, real footage that people are saying is AI. It's really been a mess for the public's relationship with facts and truth. And Ingo, what is a bot swarm? I mean, I can guess it's a lot of bots moving all together. But in the online ecosystem or on a social media platform, what is a swarm and what's its effect? So a swarm is exactly what you're saying. It's a lot of bots. And usually they have a target and an aim, like Tiffany was just saying around many of the events. She's seeing so much AI misinformation, et cetera. You want to fill an information vacuum on a topic that is maybe unclear or where there isn't much information out there.

4:11So as much as you can put out there, the better. And if you can get, for instance, hashtags to be trending, then that also favors your content over other content with the social media algorithms. That's what you want. That's where you employ swarms. So sheer volume can surf the algorithm and push that content up if you get enough of the bot swarm doing that kind of content. Imran, how concerned are you about the way AI is being used now in this moment or could be used on social media? There's plenty is it disinformation coming from humans? We'll talk about it. AI-generated content online, though.

4:49I think we need to be very concerned for two reasons. The reason why the social media age of disinformation has been so unique, so threatening, is because of the economics of disinformation on social media. So for each additional message and each additional recipient, the cost to the sender is zero. And that in the age of communications is unprecedented. If you think about sending a letter, you have to buy a stamp, you have to buy an envelope, you have to, there's a whole bunch of cost that goes into it. But now you can send a billion messages to a billion people for zero cost apart from the production of it.

5:36So that's the economics. The second thing is that we have platforms that are asymmetric in the way that they decide what content to elevate and spread to billions rather than just to the number of followers that you have. And the way that they work is they decide what to amplify based on something called engagement, which is essentially how many times do people react to it. And we know that actually reactions often to good information, your reaction isn't to leap out of your chair and say, wow, your reaction to negative information, to disinformation, to emotionally arousing lies, to hate, to something that induces a negative response is much more aggressive.

6:26So we have this negativity bias in our psychology. And so for those two reasons, disinformation has this unprecedented advantage in our information ecosystem. Now, add in AI. So I said that the cost before of sending disinformation was just that you had to produce the billion messages. I mean, you know, it takes a bit of time, takes a bit of skill as well to produce content that's emotionally effective and therefore likely to be amplified. Not with AI. AI reduces the cost of production to zero as well. And then what you have as well is in an AI system that's, you could write this quite easily, a script that says, I want you to produce a bit of content that will persuade this type of person.

7:11It then deploys it onto social media. It then records what the engagement levels are for, say, a hundred or a thousand different variations. And self-adjusts. And then self-adjusts. And then you have a perpetual, self-improving, disinformation machine. And, you know, disinformation is such an ugly word. It doesn't really mean it's lies. It's lies with the aim of persuading something unreal, untrue. And we have a system that is designed to fracture our society and allow the sort of the hyper-personalized persuasion of each individual using a mixture of lies and perhaps a bit of truth in there as well.

7:54But that changes our entire society. Well, we can talk about the technology, but we can't talk about that absent the human element. And Inga, what do we know about the actors behind these lies on social media? Who's employing this technology to shift political conversations? So we have various actors behind this. Some of them are very sophisticated, but in my research, that is usually the smaller part. So we have just so many people who are employed around the world, either they work really for financial reasons, producing these lies that Imran spoke about. But of course, we also have some ideological actors who are, you know, politicians who want to drive a certain agenda.

8:38And there's a whole network of brokers, intermediaries, like very often, for instance, politicians wouldn't directly employ, for instance, a bot farm, but instead they would go through intermediaries. And that is something that doesn't only, you know, affect the online space, it affects our democracy. And there are a lot of connecting issues around campaign financing, transparency online and offline that are part of this threat. If I can make a point there, I think it's unfair just to talk about bad actors alone. So what we're going to do is we're going to come up with a list of villains. I don't think you should have that power, Jen.

9:16I don't think Todd should. I don't think any of us should. Because what we have here is systems that are designed to be easily exploitable by anyone that wants to use them. Anyone that has the ruthlessness to use these ruthless systems which exploit our psychology in such powerful ways. And when you say systems, what systems are you talking about specifically? So both social media platforms and generative AI platforms. And one of the things that we do is we study whether or not these systems can be abused and how the underlying mathematics and the rule sets and the guardrails that are in those systems work.

9:57And what we find again and again and again is the guardrails are simply ineffective. I mean, all the mathematics I've already told you about, but if the guardrails are ineffective, then anyone could abuse them. Andrew emails a wise friend in his mid-30s said, Someday we will look at social media in the same way as cigarettes. Today gives me some hope. I quit social media despite the cost that choice had on my business. More from our guests and from you in just a moment.

10:30Let's get back into our conversation about AI and democracy with this message we got from one of you. This comes from Hannah in Wisconsin. I have been so fed up with social media and its inability to give us clear, unbiased news that I actually do not use social media pretty much at all anymore, especially not for the news. I rely on the radio and print media and reputable news sites because, especially with AI now, I really don't trust social media for my news anymore. And Ron, you were talking about the systems at work. What are some of the challenges in addressing some of those systems and trying to help them function more responsibly?

11:18We have two big problems. One is transparency. And the second is mechanisms for accountability. So transparency, these platforms are designed to be black boxes. And any attempt to try and understand how they work at anything but the most superficial level is met with intense resistance. Two very good examples are that when we, for example, at CCDH studied the platform X when Elon Musk took it over, he actually sued us for the act of doing research. Now, a court in California tossed that case out. It cost us a lot of money, but it still tossed the case out. And the second example, again, using X, but look, this is applicable to all platforms, is that the European Commission recently fined X$140 million.

12:08and the reason why was because actually in the European Union now and in the United Kingdom, there are laws on transparency of algorithms, of how platforms take decisions that they take on what content to take down, what content to leave up. So those rules that are in place, they are failing to abide by them and that's because they know that if we knew how we are being manipulated, A, the manipulation would be less effective, But also, we'd be outraged. And so are the advertisers. So with this lack of transparency, how does an organization like yours discern what's happening behind the scenes?

12:47Our burn rates, so we spend around$6 million a year employing data scientists, researchers, experts in all sorts of aspects of how social media platforms work. We do limited studies. imagine the algorithm as a cave system we are a torchlight we're a flashlight uh we just we're shining a beam and we get some understanding of the overall topography and what transparency laws are they're the floodlights you you pull the switch and boom you can see exactly how the system works and so what we do is we use the flashlight to say look there's something here guys in order to persuade lawmakers. And I think lawmakers everywhere in the world should be doing this.

13:35Transparency is not censorship. It's the exact opposite of censorship. It permits informed discourse. And then it can lead to accountability. And that's where you get the real magic is when those two things come together. Now, you know, at the moment, we know that because of odd aspects of US laws, a law passed actually almost exactly 30 years ago before social media and AI systems were even a sort of a figment of someone's imagination. It's something called Section 230 of the Communications Decency at 1996. There is no real mechanism for holding these platforms accountable. So when they do cause systemic harm, when their algorithms do cause problems, when, you know, and we study everything from eating disorder content and how that affects our kids.

14:22I'm a father myself. It's something I'm really passionate about through to how it can cause extremism, terror attacks and everything else. In those extraordinary circumstances, can we find a way to make sure they pay? So there is an incentive for them to actually build better systems that give us a healthier information diet than the ultra-processed sludge that is being shoveled down our information throats at the moment. And I'm sure in the future on 1A we'll be talking about whether big tech or social media companies have indeed become utilities and whether they should be regulated like the phone company or the electric company.

15:00In fact, if that's how they function. Let's get back to some of the specifics in the news and some of the information that's being manipulated in real time as we consume it. Tiffany, you recently reported on disinformation spreading in Minnesota during the crackdown by ICE agents there. How are you seeing AI and misinformation lies, as Imran points out, influence political narratives around what we saw in Minnesota? Let's just say I've been very, very busy since the start of the year. There have been just countless examples, in part because AI is increasingly accessible and it's very easy to use and often it's free.

15:41So one of the prevailing examples that went viral was an image of Alex Petty in the in the scrum. When he was being pushed down by the ICE agents, there were many, many videos that were taken by bystanders. But from those videos, there was a still shot that was then enhanced using AI. But the issue with enhancing, quote unquote, images using that technology is that often it'll hallucinate details. And so what happened is people began to see what some thought was a gun in his hand. Other people said that it was a cell phone in his hand. And so you ended up with this interesting phenomenon where people were looking at the same pixels and not being able to agree with what their eyes were telling them.

16:35And this phenomenon has extended through a number of other news events, like I said earlier. But another issue that we're seeing that is really concerning is that the administration is getting in on this. You know, the Trump administration is one that has always kind of embraced memes as a form of communication with its target audience. In this administration, they're taking a step beyond that. They're actually altering photographs. There was a protester who was arrested last month, and a photo of her arrest that was circulated by the White House showed her with darkened skin sobbing. And that's not the photo that was originally taken.

17:24The original photo had her very taciturn and looking quite calm while she was handcuffed, and the photo the White House put out was not that at all. Exactly. And when asked about, you know, this altered photo, the White House basically said justice will continue and the memes will continue. Well, I want to talk about a few more specifics from the Trump administration. And Inga, Imran, I'd love to hear you get in on this. I mean, we have example. Tiffany just put a finger on it. We have example after example of the Trump administration and the president himself as disinformation sources. The killing of Renee Good and Alex Petty are examples that that Tiffany just got to.

18:01Just this weekend, the government's claims about a Venezuelan man shot by immigration agents in Minneapolis collapsed in court. There were plenty of lies and disinformation online around that. Trump's years-long campaign to lie about 2020 and elections is another example, and we could spend the hour on more. Inga, how vital is social media and the social media mode of communication to the administration's incentives to essentially act as a disinformation source on a daily basis? The social media is important for every politician around the world, given the importance that it has gotten over the last years, given how many Americans get their news off social media, how many young people get their news off TikTok.

18:44So any politician would be well advised to put some resources into how to engage with the social media space. And we've already earlier heard about regulation or lacking regulation or how hard it is to enforce regulation. And regulation is a big part of what we need to create a positive online information environment, if this is still something we're aiming to achieve. The other part is norms. And this is something that we've seen a huge norm shift in the US with the Trump administration, how to employ, for instance, AI and generative AI also in their social media communication. It is something that they love to do as part of their strategy.

19:24And that is something where, as a researcher, I'm just seeing the gloves are coming off. Politicians have been much more hesitant, even in the lead up to the 2024 elections, politicians were still more hesitant using Gen AI across the spectrum. The political consultants that I've interviewed as part of a study in 2024 were saying there's much more we can do with AI and Gen AI. and there's nothing to really regulate us or rein us in in that sense, but we're not doing it because we think the voters are going to push back. We are thinking that these are just norms that we don't want to break. We've been seeing norms unraveling in the US over the last months heavily and that gives me big concerns for the next elections to come.

20:06It's worth saying that given that these platforms are so easy to manipulate and they have few guardrails, I think the last time I was in this studio, I was talking about our report on fake image factories. So we tested a variety of AI image tools. We found that they generated photorealistic convincing election disinformation in 41 % of tests, voting-related disinformation in 59 % of tests, and misleading election content was produced by MidJourney, the leading image generator, in 65 % of tests. And what we said was we actually need a non-proliferation treaty in politics. We need people to agree that we will not use these because when it can be abused, it will be.

20:54And if it's so easy, in fact, if it's almost an incentive to use it, that abuse is so easy, is it genuinely an abuse or is it just what they intended their platform to be used for? I do think we have to, you know if you owned a candy store you didn't put anyone on the front desk you didn't have any cameras would you be surprised if you had no candy at the end of the day? Of course you wouldn't it's your fault you know, it's your fault for failing to actually put into place the guardrails to make sure that these systems aren't abused and they have been and they increasingly will be and what worries me is of course that we've already seen the degradation of our politics We've seen the normalization of performative politics.

21:40And that's, again, in part driven by our information ecosystem because an algorithmic environment that favors extreme emotionally effective disinformation rather than good information shifts power towards anyone that can maximize attention. And we've seen that sort of slowly destroy our politics, turn it away from seriousness into performance. But now we see an incentive to lie in the most egregious way possible to literally rewrite reality using generative AI platforms. One of you emails, it's easy to navigate. I go to the source and I only follow reporters I know and trust. Another of you asks, does 1A print or admit to prior AI deceptions?

22:25Now, we do have a policy on AI and just when we get it wrong, we report on AI deceptions that reach the public. but we only rely on reputable reporters and outlets. Tiffany, I want to return to how the administration is using memes to communicate with the public. And what did you see during the release of the Epstein files, those files that we've seen so far? So the DOJ had an interesting note in its release about the Epstein files where it said, just be aware that these files should not be treated as just evidence in and of themselves, right? There are unsubstantiated tips and there's a lot of fake imagery and fake emails in there as well.

23:13And that kind of has thrown a wrench into the public's understanding of what's in this tranche of documentation, because you have people who are treating the Epstein files released as a sort of game. They're going in and they're picking and choosing pieces of information that they're finding, imagery that they're finding, and they're mashing it together into conspiracy theories and rumors. And a lot of the foundation of what's forming those conspiracy theories is either not true or it's misleading. And then on top of that, outside of what's in that tranche of files, social media is full of things that people have created using artificial intelligence as if it was in that tranche, but it's not.

24:02And so you have the government's official release of documents, which is not reliable on its own. And then layered on top of that, you have folks on social media who are further muddling the information environment by creating things that never existed. And so We're seeing accusations against people like Zoran Mandami, the New York mayor, that when he was a kid, he and his mom hung out with Epstein. That's a rumor that stems from an artificially generated image. John in Maryland emails, I'm a health care worker. More and more, I have to try to dispel wrong and often harmful information that patients find through AI, influencers and even more disturbing from U.S.

24:47government websites. Politics and social media have erased people's trust in those of us who are educated professionals. And Susanna in South Carolina emails, I am definitely worried about AI-generated misinformation influencing elections. That's why I follow experts like Brian Krebs, who is a security reporter, and 404 Media, which was founded by tech journalists, so that I'm aware of nefarious bot campaigns. Reading their content makes me more skeptical, which I think is good. So to come, what we can learn about the effect of misinformation on democracy and elections that's just ahead.

25:24Imran, in December, the Trump administration placed you and five other people combating online misinformation on a visa ban. A judge blocked that order from the administration, blocked them from detaining or deporting you. What was the reason behind that move against you and how would you characterize the administration's attitude toward free speech and online disinformation and lies right now? Yeah. So on December 23rd actually, two days before Christmas, someone texted me a tweet in which the State Department announced that it was issuing sanctions against five European individuals for what it said was censorship.

26:08And in particular, they were aggrieved at the fine that was placed on Elon Musk's platform, the one formerly known as Twitter, for failing to abide by the European Commission's transparency regulations. Now, let's leave aside for the moment the insanity of saying that a fine for failing to abide by transparency rules is an act of censorship, which is clearly ridiculous. the idea of then seeking to solve censorship by telling someone that you are deporting them for their advocacy is again bananas because that is actually the definition of censorship itself and you know this happened before Christmas I'm married my my wife is you know she's not just American.

27:00She's from Oklahoma, so she's Merican. We have a daughter. We're having our second daughter in a few weeks. And I've built a family life here. I have paid my taxes. I've been a good law-abiding citizen. Where do you think this order came from? Why did the Trump administration choose you? Let me make it simple for you. It's big money, it's big tech, and it's utterly corrupting impact on government that's been going on for some years now. And it's not just the United States where they seek to change the way that government works. They seek to influence government in the most autocratic, in the most tyrannical, in the most anti-constitutional ways.

27:43The idea that Elon Musk can say, I hate the way that this man speaks, and so therefore I want to punish him. Now, that might sound like I'm, you know, that might sound like a conspiracy theory, apart from the fact that he's already tried to sue us for our speech. And a court found that he was trying to violate our First Amendment's rights and awarded us an anti-slap ruling, which is a kind of ruling which says, basically, this is a classic example of a rich man trying to use lawfare to hurt someone less powerful, less wealthy than them. And I think we're seeing the same thing happen here again.

28:21And it is extraordinary given the impact that social media platforms, the negative impact that social media platforms are already having on our democracy at a systemic level and the way that they're designed, in the way that they corrupt our information ecosystem. But now they're taking it that one step further and they're seeking to actually be allowed to carry out unconstitutional acts. We got this email from Christopher who says, I'm a commissioner in Bay City, Michigan. I'm concerned about how social media warps the reality of politicians and decision makers. Before even considering the AI bot swarms, we have the mundane reality that local officials are influenced by so much misinformation and disinformation pushed by non-constituents.