THE posters that dotted the streets of Buenos Aires had a certain Soviet flavor.
There was one of Argentina’s presidential candidates, Sergio Massa, dressed in a T-shirt with what looked like military medals, pointing to a blue sky. He was surrounded by hundreds of elderly people – in drab clothes, with serious and often disfigured faces – who looked towards him with hope.
The style was not a mistake. The illustrator was given clear instructions.
“Illustration of a Soviet political propaganda poster by Gustav Klutsis depicting a leader, Masssa, standing firmly,” said a suggestion that Massa’s campaign was feeding an artificial intelligence program to produce the image. “Symbols of unity and power fill the space,” the message continued. “The image exudes authority and determination.”
Javier Milei, the other candidate in Sunday’s run-off, reacted by sharing what appear to be artificial intelligence images depicting Massa as a Chinese communist leader and himself as a cute cartoon lion. They have been viewed more than 30 million times.
Argentina’s elections quickly became a test case for artificial intelligence in campaigning, with the two candidates and their supporters using the technology to manipulate existing images and videos and create others from scratch.
AI made candidates say things they hadn’t said and put them in popular movies and memes. It has created election posters and sparked debates over whether the real videos are actually real.
AI’s prominent role in Argentina’s election campaign and the political debate it has sparked underlines the growing prevalence of the technology and shows that, with its expanding power and falling costs, it is now likely to become a factor in many democratic elections around the world.
Experts liken this moment to the dawn of social media, a technology that offers tantalizing new tools for politics – and unanticipated threats.
Massa’s campaign has created an AI system that can create images and videos of many of the election’s key players — candidates, running mate, political allies — doing a wide variety of things.
The campaign used artificial intelligence to portray Massa, Argentina’s serious, center-left economy minister, as strong, fearless and charismatic, including videos showing him as a soldier at warTO Ghost Buster AND Indiana Jonesas well as posters that evoke Barack Obama’s 2008 “Hope” poster. and a cover of The New Yorker.
The campaign also used the system to portray his opponent, Milei – a far-right libertarian economist and TV personality known for his outbursts – as unstable, placing him in films like “Clockwork Orange” AND “Fear and loathing in Las Vegas.”
Much of the content was clearly false. But some creations have followed the line of misinformation. Massa’s campaign produced a “deepfake” video in which Mr. Milei explains how a market for human organs would worksomething he said fits philosophically with his libertarian views.
“Imagine having children and thinking that each of them is a long-term investment. Not in the traditional sense, but thinking about the economic potential of their organs,” says the manipulated image of Mr. Milei the fabricated videopublished by the Massa campaign on its Instagram account for AI content, called “AI for the Homeland”.
The caption of the post reads: “We asked an artificial intelligence to help Javier explain the business of selling organs and this happened.”
In an interview, Massa said he was shocked the first time he saw what artificial intelligence could do. “I wasn’t mentally prepared for the world I’m going to live in,” he said. “It’s a huge challenge. “We are on a horse that we have to ride but we don’t know his tricks yet.”
The New York Times then showed him the deepfake of his campaign created by Mr. Milei and human organs. He looked troubled. “I don’t agree with this usage,” he said.
His spokesperson later stressed that the post was a joke and clearly labeled as generated by artificial intelligence. His campaign said in a statement that the use of artificial intelligence is to entertain and make political points, not to deceive.
Researchers have long worried about the impact of artificial intelligence on elections. Technology can deceive and confuse voters, casting doubt on what is real, adding to the misinformation that can be spread by social networks.
For years, these fears had been largely speculative because the technology to produce such fakes was too complicated, expensive and unsophisticated.
“Now we’ve seen this absolute explosion of incredibly accessible and increasingly powerful democratized toolsets, and that calculus has fundamentally changed,” said Henry Ajder, an England-based expert who has advised governments on content generated by artificial intelligence.
This year, a Toronto mayoral candidate used grim AI-generated images of homeless people to telegraph what Toronto would turn into if he were not elected. In the United States, the Republican Party released an AI-created video showing China invading Taiwan and other dystopian scenes to depict what would happen if President Biden wins a second term.
And the campaign of Governor Ron DeSantis of Florida shared a video showing AI-generated images of Donald J. Trump embracing Dr. Anthony S. Fauci, who has become an enemy of the American right for his role leading the national response to the pandemic.
So far, AI-generated content shared by campaigns in Argentina has either been labeled as AI-generated or is so clearly fabricated that it is unlikely to fool even the most gullible voters. Instead, technology has enhanced the ability to create viral content that previously would have taken teams of graphic designers days or weeks to complete.
Meta, the company that owns Facebook and Instagram, said this week that it will require political ads to disclose whether they use artificial intelligence. Other unpaid posts on sites using artificial intelligence, even if related to politics, would not be required to contain any disclosures. The U.S. Federal Election Commission is, too considering whether to regulate the use of artificial intelligence in political ads.
The Institute for Strategic Dialogue, a London-based research group that studies Internet platforms, signed a letter urging such regulation. Isabelle Frances-Wright, the group’s head of technology and society, said the extensive use of artificial intelligence in Argentina’s elections was worrying.
“I definitely think it’s a slippery slope,” he said. “A year from now, what already seems very realistic will seem even more so.”
Massa’s campaign said it decided to use artificial intelligence in an effort to show that Peronism, the 78-year-old political movement behind Massa, can appeal to young voters by mixing Massa’s image with pop culture and memes.
To do this, campaign engineers and artists fed photos of Argentina’s various political actors into open-source software called Stable Diffusion to train its AI system so it could create fake images of those real people. Now they can quickly produce an image or video of more than a dozen major political actors in Argentina doing almost anything they ask.
During the campaign, Massa’s communications team informed the artists working with the campaign’s AI about what messages or emotions they want the images to convey, such as national unity, family values and fear. The artists then brainstormed ideas for featuring Massa or Milei, as well as other political figures, in content that references films, memes, art styles, or historical moments.
For Halloween, Massa’s campaign asked its AI to create a series of cartoonish images Mr. Milei and his allies as zombies. The campaign also used artificial intelligence to create a dramatic cinematic trailerwith Buenos Aires, the capital of Argentina, in flames, Milei as an evil villain in a straightjacket and Massa as the hero who will save the country.
The AI images were also shown in the real world. The Soviet posters were one of dozens of drawings that Massa’s campaign and his supporters printed to display in public spaces across Argentina.
Some images were generated by the campaign’s AI, while others were created by supporters using AI, including one of the best-known, an image of Mr. Massa on horseback in the style of José de San Martín, a hero of Argentine independence.
“Massa was too rigid,” said Octavio Tome, a community organizer who helped create the image. “We show a boss-like Massa, and he’s very Argentinian.”
The rise of artificial intelligence in Argentine elections has also led some voters to question what is real. After a video circulated last week of Massa appearing exhausted after a campaign event, his critics accused him of taking drugs. His supporters quickly responded, claiming that the video was in fact a deepfake.
His campaign confirmed, however, that the video was indeed real.
Massa said people are already using artificial intelligence to try to hide past mistakes or scandals. “It’s very easy to hide behind AI when something you said came out that you didn’t want,” Massa said in the interview.
Early in the race, Patricia Bullrich, a candidate who failed to qualify for the runoff, tried to explain away leaked audio recordings of her economic advisor offering a woman a job in exchange for sex, saying the recordings were been manufactured. “They can falsify rumors, alter videos,” she said.
Were the recordings real or fake? It is not clear.