Trump’s AI pics of Harris & Swift raise concerns about political media manipulation

pic.twitter.com/H0ExcNXBdl

— Donald J. Trump (@realDonaldTrump) August 18, 2024

Trump has referred to Harris as ‘Comrade Kamala’ and falsely labeled her a communist. The Republican nominee posted the image to Truth Social, his own social platform on Saturday, before reposting it to X, formerly Twitter, on Sunday. He rejoined X last week after a long hiatus, taking to the platform to campaign and join owner Elon Musk for a rambling, two-hour long interview.

The post came on the tails of broader personal and political criticisms levied at Harris by Trump, who has said that her policies would produce disastrous economic outcomes. Trump has specifically targeted Harris’ proposal to address price gouging, suggesting she aims to impose socialist-style price controls.

The image was posted within hours of other manipulated media shared by Trump – this time falsely signaling campaign support from Taylor Swift fans. Those images, some of which were AI-generated, depict young women wearing ‘Swifties for Trump’ tee-shirts, with one claiming that Swift fans shifted their support to Trump after the pop star canceled a concert in Austria over security concerns. Another image, a parodic recreation of the iconic ‘We Want You’ Uncle Sam poster, suggests that Swift herself is encouraging fans to vote for Trump.

Nonprofit group True Media, which tracks the spread of election-related deepfakes, has confirmed that most of the visuals have been manipulated. There is also no evidence of a real ‘Swifties for Trump’ movement (though there is a ‘Swifties for Harris’ Facebook group with over 5,000 members).

One image included in the post is authentic, however; it depicts Jenna Piwowarczyk, who donned a homemade ‘Swifties for Trump’ shirt at a June rally in Wisconsin and has since begun selling her shirts online.

Political speech or parody?

Neither the Swifties images nor the communist Harris convention images proved to be especially sophisticated fakes that could easily trick millions, raising questions about how user-generated content – and the platforms that host it – toe the line between protected speech, parody and maliciously deceitful (and potentially policy-violating) media.

“There’s a difference between what I would understand to be satire – theoretically, Republicans waving around taxidermied bear arms because of the Second Amendment – compared to [something less obviously satirical], like a picture of Taylor Swift saying, ‘I support Trump,’” says Henry Ajder, a leading expert on on AI and synthetic media and the founder of AI advisory firm Latent Space.

The doctored Harris and ‘Swifties for Trump’ media shared by Trump this weekend, Ajder suggests are “particularly tricky” because they blur the lines between political and satirical speech. “The interpretation of what satire and parody is, is [loose here]; this seems like pretty low-effort if that's what [Trump’s campaign] is trying to claim it is,” he says.

Platforms also label and police manipulated media differently, setting their own standards for content moderation. X, for example, where the communist Harris rally image was reshared, forbids “misleading media,” which it defines as “synthetic, manipulated or out-of-context media that may deceive or confuse people and lead to harm.” But it also protects parodic or satirical content to some degree.

So with an image like the Harris one, which is not likely to trick discerning audiences, a potential content policy breach on X may be unclear. “To what extent can we say, ‘This is clearly fake?’ To whom is it clearly fake?,” says Ajder. “To the white-collar worker in New York compared to the blue-collar worker in Montana or … [someone] international? It’s really tough to understand what is sufficiently deceptive or sufficiently realistic to require a label [or other policing].”

Neither social platforms and publishers nor AI developers at this point have standardized practices or requirements around disclosure of manipulated media. This means that while some AI media generators may produce watermarks, and some platforms, like Instagram, include default labels indicating when media has been modified or created with AI, audiences lack consistent methods for discerning fact from fiction.

Use of AI-generated and synthetic media has already proved a recurring issue in the 2024 election cycle, sparking discussions about the role of artificial intelligence in politics. Earlier in the year, Trump supporters circulated AI-generated images of the former President surrounded by Black voters. In another instance, voters in New Hampshire received robocalls from an AI-generated recreation of the voice of Joe Biden, which urged them not to vote in the primary election.

And it’s not just a US problem. Elections across the globe – from Argentina’s 2023 general election to elections in the UK and the Netherlands this summer – have seen an influx of politically-motivated deepfakes and manipulated media.

The sliding scale of realism & deceit

To be sure, some attempts to influence elections using manipulated media are more advanced than others. Argentina’s Bolsonaro, portrayed as a “chainsaw-wielding maniac,” Ajder says, is likely to be less convincing than more advanced edits.

Of course, questions of labeling and policing manipulated content online become more pressing the when the media is more sophisticated. And with the advent of advanced generative AI models like those commercialized by OpenAI and Google, the problem is already rearing its ugly head.

“Realism has really improved of recent,” says Ajder, who points to improvements in lighting, skin tones and the depiction of extremities like fingers, that have been made by many AI image- and video-generating tools in the last six to nine months.

The notion that highly realistic portrayals create more immediate threats is echoed by Dr Shawn DuBravac, a futurist, tech writer and the former chief economist for the Consumer Technology Association. As he puts it: “The real danger in synthetic media and deepfakes doesn’t lie in the obviously manipulated images. The true risk is in synthetic media that is indistinguishable from reality. Synthetic media can and will erode trust in the platforms where they are shared.”

Heightened realism in synthetic and manipulated media ups the stakes for both platforms and consumers. As the technology advances and more realistic fakes flood our newsfeeds, Ajder says, “these kinds of images depend even more on the context in which they’re shown to give audiences a sense of whether they’re authentic or not.”

One silver lining to the widespread dissemination of manipulated media is that it could incentivize platforms to proactively solve the problem with new measures around transparency. As a result, these platforms could help stem the spread of misinformation and win user trust.

But acting on content moderation and labeling policies will always be easier said than done. While Meta, TikTok, YouTube and X all have rules for moderating synthetic and manipulated media, Ajder says, “having a policy is one thing – enforcing a policy is another.”

Consistency in enforcement is likely to require more advanced human- and machine-driven moderation capable of identifying manipulated media and applying principles in standardized ways.

Ad dollars caught in the crossfire

Monitoring synthetic and manipulated media will need to become a priority for platforms that hope to maintain trust as well as ad spend. Just ask X, whose lax content moderation practices under Musk’s leadership led to a mass exodus of ad dollars, effectively gutting the company’s revenues. (During the first year of Musk’s ownership of X, the platform lost around 55% of its ad revenue year-over-year every month, according to data provided to Reuters.)

For advertisers – many of whom are already especially cautious about brand safety in the midst of contentious elections coverage and potential misinformation on social media sites – the risks only rise in the face of manipulated media.

“There are growing risks of being inadvertently associated with misleading or inflammatory content,” DuBravac says. “This will likely lead to increased scrutiny of ad placements and a reassessment of partnerships.”

And even seemingly obvious fakes, like the Uncle Sam-styled Swift in Trump’s post on Truth Social, may ostracize some brands, say Ajder. “For advertisers, you would not want to be associated with content or positioned near content which could potentially be seen by some of your audiences to be associated with their idol being puppeteered for a political purpose.”

For both advertisers and everyday users attempting to navigate a high-stakes election cycle increasingly inundated with manipulated content, DuBravac offers some simple advice: “Stay calm and carry on critically.” What that means in practice, he says, is approaching political content with a scrupulous eye, seeking to identify original sources of inflammatory content and cross-referencing credible sources to establish veracity.

It’s advice echoed by Dr Laurel Cook, a West Virginia University associate professor of marketing who has studied deepfakes. “When it comes to the use of advanced technology and its potentially dangerous outcomes, like deepfakes and tech-altered content,” she says, “nothing can beat healthy consumption practices and discernment.” She suggests studying common ‘tells’ in manipulated media, like unnatural facial movements and inconsistent lighting, and urges users and advertisers to consult multiple sources for accurate information.

“By adopting these practices, as consumers or as brand managers, we are attenuating the effects of deceptive tactics and contributing to a more informed and resilient online community,” she says.

As we inch closer to the US presidential election in November, it’s unlikely that we’ll see a decline in AI-generated and manipulated media, Ajder says. Outside of imagery, he expects to witness an uptick in video- and audio-based fakes. And while most of this content is likely to be obviously satirical, some of it will surely be more subversive and more dangerous.

“The big question – and the jury is out on it,” he says, “is, to what extent will that have a meaningful impact on voter turnout, on actual Senate seats or on the actual presidential election itself?”

For more, sign up for The Drum’s daily newsletter here.