Two years ago, Noelle Martin discovered someone had made a “deepfake” video about her. Martin is a 26-year-old Australian law graduate who has lobbied governments and corporations to take action against the online harassment of women. Now, someone on the internet had decided to attack her via a technique that uses artificial intelligence to swap one person’s face onto another’s body.
Experts studying this phenomenon have found that well over 90 percent of deepfake videos involve faces swapped into pornographic scenes—the vast majority being women, most often celebrities but also politicians, activists or non-famous women. That’s what someone had done with Martin. The video, she figured, was an attempt to get her to stop her advocacy work by shaming her. “It was absolutely weaponized,” she told me recently. When she saw the video circulating online, she felt a stab of fury: “The audacity of these people to do that to me,” she said. She also couldn’t help wondering: Would people who saw it actually believe it was her?
Deepfake videos present an unsettling new phase in the evolution of media. Manipulating video used to be wildly expensive, the province of special-effects masters. But new AI technology has made it much easier. Indeed, one commonly used piece of software for doing it—which uses a “deep learning” form of artificial intelligence, hence the “deep” in deepfakes—was released anonymously online for free in 2018.
In December 2020, Sensity, a fraud-detection firm, found 85,047 deepfake videos circulating online, a number that had been roughly doubling every six months; there are now likely hundreds of thousands in existence. Some are harmless—Nicolas Cage’s face swapped into scenes from movies he wasn’t in, say. But “the majority of deepfakes created by experts are malicious,” says Giorgio Patrini, Sensity’s CEO.
Many observers worry that deepfakes could become a major threat in politics, used to humiliate political figures and advocates like Martin or even make them appear to say things they never said. “What it could do to diplomacy and democracy—we’re holding our breath,” says Danielle Citron, professor of law at the University of Virginia School of Law. It’s an unsettling moment, where our ability to discern what’s real feels newly imperiled.
In fact, these anxieties echo the earliest days of photography. Then, as now, through cutting-edge fakery, major public figures were counterfeited, and questions emerged about whether a powerful new technology made it impossible to trust what you saw.
In one sense, photo manipulation began as soon as photography did. Early image-capturing technologies were crude—images had no color, and slow shutter speeds washed out details, such that skies, for example, appeared “ghastly, lifeless,” one photographer complained.
So photographers from the get-go were working hard to alter images. They would paint on colors, or enhance details by drawing on an image with ink and paint. They quickly hit upon the technique of composites: To make gorgeous seascapes, the French photographer Gustave Le Gray would take photos of wave-swept oceans and splice in separate photos of clouds, even reusing the same clouds in different photos. In 1857, the photographer George Washington Wilson created faux group shots of high society by taking individual photos of subjects, cutting out their images and assembling them into a crowd, then photographing the resulting collage.
Viewers were fooled. “I had numerous inquiries as to when and where all these people had been collected and photographed,” Wilson’s gallerist said.
Photographers found the manipulations exciting, a new artistic technique. “I think they were just like anyone experimenting with a new art—there’s a certain amount of just, isn’t it cool we could do this?” says Peter Manseau, curator of American religious history at the National Museum of American History and an expert on early photo manipulation.
Commercial portrait firms employed legions of women as touch-up artists, softening wrinkles and reshaping features in the 19th century’s forerunners to Instagram filters. “Anyone who went into a portrait studio would most likely be asked, ‘Would you like us to touch this up for you, and, you know, make your nose smaller?’” says Mia Fineman, photography curator at the Metropolitan Museum of Art in New York City and author of Faking It: Manipulated Photography Before Photoshop.
Fakery soon entered politics, as photographers tried to generate patriotic or stirring imagery. To create a “photo” of Ulysses S. Grant with his troops, photographer Levin Corbin Handy pasted Grant’s head onto the body of another man, then pasted that composite onto a picture of Confederate prisoners of war. There were even meme-like parodies. When false rumors flew that Confederate President Jefferson Davis had sought to escape capture in 1865 by wearing a woman’s petticoats, photographers gleefully produced composite photos that plastered his head onto an image of a woman.
Did the public know these images were fake? Historians aren’t sure. Even if they did know, it’s not clear they cared. Photography was not yet seen as a true document of reality. “There was no understanding that an image should be objective,” Manseau says.
But photo manipulation caused particularly hot debate in one field: “spirit photography.”
Amid the spiritualism movement after the Civil War, many bereaved Americans became convinced they could communicate with departed loved ones. They held séances, urging the dead to rap on tables or to speak to them through mediums. Photographers claimed they could capture images of the dead. In the United States, the most famous spiritualist deepfaker was Boston’s William Mumler, who in 1862 began creating pictures that appeared to show live human subjects accompanied by translucent ghosts. Many of Mumler’s subjects excitedly proclaimed he’d photographed one of their dead relatives. “What joy to the troubled heart,” as Mumler wrote in a promotional pamphlet, “to know that our friends who have passed away can return.”
But debate raged. Skeptical photographers suspected Mumler’s pictures were mere double exposures—two negatives exposed onto a single photo sheet, with the “ghost” exposed only partially, to make it translucent. Yet when a few skeptics accompanied him into the darkroom, they couldn’t deduce how he was pulling it off. Even so, in 1869, New York’s city marshal charged Mumler with fraud after a reporter lodged a complaint at City Hall, and the ensuing trial made explosive headlines: “The Science of the World Against Spiritualist Theory,” the New York Herald proclaimed. The city even brought in showman P.T. Barnum to testify against Mumler; Barnum showed the court a faked spirit photograph he’d staged of himself, to demonstrate how readily such fakery could be done. Still, after more than a month of trial, the judge let Mumler go free, saying the prosecution hadn’t proved that “trick and deception has been practiced by the prisoner.”
Manseau—who wrote The Apparitionists, a 2017 book about Mumler’s trial—can’t be sure how many people believed that spirit photos were real. He thinks many took them seriously, but not literally: The photos gave comfort, and that was enough. Post-trial, Mumler still took the occasional spirit photograph. His most famous was one of Mary Todd Lincoln next to a translucent image of her assassinated husband.
“It was a real consolation to her to have this image,” Manseau notes, though it’s unclear whether Mary Todd truly believed it was Lincoln’s ghost.
Arthur Conan Doyle, the famous creator of Sherlock Holmes, became an adherent of spirit photography, despite having made his fortune writing tales about ruthlessly empirical deductions. In 1920, he was outright duped by a pair of girls in Cottingley, England, who faked a set of five photos that purported to show cavorting fairies. Conan Doyle published the pictures in the Strand magazine, and in a 1921 book, The Coming of the Fairies, he rhapsodized about the images: “What joy is in the complete abandon of their little graceful figures.”
Still, the public was becoming more familiar with the tricks of composite photography. “Poor Sherlock Holmes—Hopelessly Crazy?” ran one 1922 headline about Conan Doyle. “He did get pilloried, and it didn’t help his career,” says Andrew Lycett, author of The Man Who Created Sherlock Holmes. “But he didn’t really care.” The author went to his grave believing those photos of fairies and spirits were real.
How many of us will be gulled in the same way today by deepfake videos?
On the one hand, experts say, we might be less easily duped because we’re more savvy about image manipulation, regularly using smartphone and computer apps to tweak our own pictures and videos. But video deepfakes are still novel, and we haven’t yet learned to suss out whether a face has been swapped.
To protect those targeted by hostile deepfakes—again, for now, mostly women—legal scholars like Mary Anne Franks at the University of Miami are proposing laws to criminalize “digital forgeries,” or deepfakes that would appear authentic to a reasonable person. Non-malevolent uses, like satire or comedy, would remain legal, Franks says.
Granted, she adds, such laws against forgeries are only “a blunt tool.” Indeed, the people who make pornographic deepfakes often aren’t trying to fool anyone. Many openly revel in the fact that they’re using a fake to humiliate a female celebrity.
Deepfakes are not yet common in politics, possibly because they still require more technical skill than merchants of political misinfo typically possess. But in three to five years, says Hany Farid, an expert on digital images at the University of California at Berkeley, you’ll be able to create realistic deepfakes on your iPhone; rudimentary ones are already possible.
Deepfakes may also become a new canvas for artists. Stephanie Lepp, one video artist, recently created “Deep Reckonings”: surprisingly realistic videos of public figures regretting their actions, including an uncannily believable Mark Zuckerberg apologizing for Facebook’s alleged promotion of “hateful propagandists” and “ethnic violence.” When Lepp posted them, she explicitly marked the videos as deepfakes, but viewers said it was still thought-provoking to see these figures wrestle with their public impact, however fictitiously.
For Lepp, deepfakes are a tool to help imagine a different, better world. They can evoke “that pathway to the future we aspire to,” she tells me. Sometimes only a fake can express our truest desires.
A cropped history of visual hoaxes
By Ted Scheinman
Hippolyte Bayard claimed to invent photography, but after fellow Frenchman Louis Daguerre snagged a patent, in 1839, Bayard showed his anguish by staging the first known fake photograph: Bayard’s body, supposedly after his suicide by drowning. In fact, he lived until 1887.
In 1933, Betty Archer convinced English publisher William Heinemann to release a book of remarkable photographs supposedly taken by her husband during World War I—and earned $20,000. Not until 1984 did Smithsonian archivists discover that Archer’s husband, a former U.S. Air Force pilot who made props and scenery for movies, had staged the action using model planes.
This widely distributed 1934 “evidence” of the Loch Ness Monster was created as a prank by Robert Kenneth Wilson and friends during a fishing trip in Scotland. In 1975, the son of one of the conspirators admitted the photograph was no more than a toy set adrift in the water.
Ghost in the Machine
Hubert Provand claimed in 1936 to have photographed “the Brown Lady,” a spirit said to haunt Raynham Hall in Norfolk, England. Many viewers saw it as proof that ghosts are real. In 1984, it was revealed as a mere composite of two photographs.