This week, New York mayoral candidate Andrew Cuomo scaled a skyscraper, drove a subway train, and moonlighted as a Broadway stagehand—or at least you might think he had if you didn’t read the fine print. Given that he launched his campaign with a man-on-the-street video in which he appeared to walk no more than a few city blocks, the 67-year-old’s sudden willingness to hang off the side of a building for the sake of an eye-catching campaign ad would have been a striking development indeed, had it not been, as the small type at the bottom of the 30-second clip revealed, “created with the assistance of artificial intelligence.”
Cuomo’s 33-year-old opponent Zohran Mamdani, whose campaign has thrived on energetic stunts like walking the entire length of Manhattan, was quick to razz Cuomo for his fakery. But as irresistible as it is to dunk on Cuomo as an out-of-touch boomer falling back on A.I. slop, the images in his campaign ad aren’t just lazy replacements for the real thing. They’re a dog whistle to Cuomo’s right-leaning supporters, the ones who see the possibility of a Muslim socialist running the biggest city in the country as an “existential threat.” In theory, generative A.I. can create imagery in any imaginable style, or at least any style that an actual human has previously imagined. But its default mode, or at least the most widely propagated one, is a kind of kitschy hyperrealism that has also become the dominant aesthetic of the MAGA movement. Only a few days before Cuomo’s ad dropped, Donald Trump posted a deepfake of himself talking about the nonexistent technology known as “medbeds” to his Truth Social account, which he has frequently used to recirculate A.I.-generated content. As the tech journalist Charlie Warzel put it last year, “The GOP is becoming the party of AI slop.”
That makes using generative A.I. imagery in documentary filmmaking especially fraught. Not only does the inclusion of realistic computer-generated visuals undermine the form’s inherent claim to truthfulness, but it subconsciously links the film and its maker to an extremist political movement, not to mention a tech industry that has spent the past nine months cozying up to if not outright allying itself with Trump’s administration.
Still, generative A.I. has been making its way into more and more ostensibly nonfiction films, with varying degrees of transparency. The Oscar-winning director Morgan Neville gave no indication that his 2021 documentary Roadrunner had used A.I. to simulate the voice of the late Anthony Bourdain, and actually bragged to one reporter that viewers would never know which lines of Bourdain’s voice-over were real and which were posthumous fakes. André Is an Idiot, a portrait of an advertising executive progressing through the final stages of colon cancer that premiered at Sundance in January, prominently flags its use of A.I. to create audio of its late subject reading his final diary entry aloud. And you had to stick around through the end credits of Jesse Short Bull and David France’s Free Leonard Peltier to discover that the movie used A.I. to enhance recordings of its ailing subject’s voice and plug holes in the historical record.
The fact that A.I. has been overwhelmingly employed in documentary filmmaking as a kind of archival stopgap, cleaning up previously unusable materials and generating new ones when old ones aren’t available, is what makes its deployment in Raoul Peck’s Orwell: 2+2=5 especially striking. The movie, which pairs an examination of the writer’s work with a biography focused on the writing of 1984, and uses both to train George Orwell’s insights on a world that keeps lurching closer to the dystopias he imagined, draws on a vast range of secondary materials. There are the expected versions of 1984 and Animal Farm, of course, and numerous excerpts from Orwell’s writing, read by the actor Damian Lewis. There’s news footage that brings Orwell’s story into the present, cutting from the fascist Spain of 1937 to Israeli bulldozers demolishing Palestinian settlements in 2024. And then there are the citations from fiction films: Out of Africa, to illustrate the colonial mindset in which the Bengal-born Orwell was raised, and from which he later forcibly dissociated himself; Ken Loach’s Land and Freedom, which follows a British worker who enlists with the Communists in the Spanish Civil War; even Red Sparrow, a trashy thriller in which Jennifer Lawrence plays a Russian ballerina pressed into service by a totalitarian spymaster.
With so many sources to draw upon, why create new ones? In a sense, Trump forced Peck’s hand. During the 2024 campaign, Trump supporters circulated pictures of the presidential candidate surrounded by beaming Black men and women, purporting to show his support in the Black community. The trouble was that none of those Black people actually existed. The images had been created using generative A.I., and although a BBC investigation was unable to trace them back to their source, it did conclude that they appeared to have been “made and shared by US voters themselves.”
In Orwell: 2+2=5, those images play in a montage after footage of the real Trump claiming he’s done more for Black Americans than any president since Abraham Lincoln, and they’re overlaid with a wobbly Delta blues song whose singer keeps warbling the word Bla-a-ck as a vaguely Trumplike voice says things like “Your strength is unmatched” and “Hungry for gumbo and for cornbread.” As a caption at the end of the sequence reveals, all of it—the images, the song, the spoken words—was generated by A.I.
Given that Orwell wrote about how he feared that “the very concept of objective truth is fading out of this world,” Peck knew he had to deal with this latest and most insidious subversion. “But I also knew I had to be totally transparent about it,” he explained this week at the offices of his distributor, Neon. (The movie is now playing in select theaters and will expand to more cities on Friday.) “I couldn’t fake it or not say that I’m using it, and I don’t use it just for the sake of using it. I knew it was part of the whole deconstruction of language and image and reality.”
Peck says that a Belgian filmmaker friend who has worked extensively with A.I. helped generate the film’s images. (As a marker of how controversial the use of A.I. is in filmmaking right now, Peck declines to provides the filmmaker’s name, and they don’t appear to be credited.) But Peck also stresses that he chose carefully from a wide range of available options, and that the final creative control rested with him. “It’s not an instrument at this stage that I could use and say ‘This is cinema,’ ” he explains. “If you don’t have a handle on what you’re creating, it’s a lottery. We should control our instrument, and not the other way around.” In other words, the uncanny grotesquerie of the movie’s A.I. images is a feature, not a bug. They look as bad as the world feels.
The 72-year-old Peck, who briefly served as Haiti’s minister of culture before he made his international breakthrough as a filmmaker, concentrates in both fiction and documentary on the history of systemic oppression and the struggle for liberation, whether he’s channeling James Baldwin in the Oscar-nominated I Am Not Your Negro or casting Idris Elba as a survivor of the Rwandan genocide in Sometimes in April. But like his HBO miniseries Exterminate All the Brutes, Orwell: 2+2=5 works toward a more transhistorical understanding of how totalitarianism works, and why it persists. At times, the movie sprawls into vague generalization and condenses other points so much that they risk being misconstrued. In a series of captions identifying contemporary examples of Orwellian Newspeak, the movie cites “Antisemitism 2024,” which it defines as “a weaponized term to silence critics of Israeli occupation.” Peck says he spent a good deal of effort getting the wording right—the 2024 was a late clarification—but even so, he says, a Jewish friend told him she felt a “kind of malaise” when the audience at the film’s Cannes premiere saw those words and burst into applause.
That sweeping approach also allows the movie to draw provocative connections across decades and cultures, and to cut through the techno-utopian rhetoric of the present to find the old-fashioned authoritarianism beneath. In one clip, the CEO of the now-defunct facial-recognition startup AnyVision asks an unseen interviewer, “What do you prefer, being safer when you wake up in the morning and go to the streets, or not being analyzed by A.I. systems? I want to be safer.” Peck could have drawn on any number of fictional or real-life sources to illustrate the dangers of the surveillance state. But instead, he once again uses generative A.I. to create a series of eerie tableaux, images of happy but blank-faced men and women whose faces eventually distort into a wordless scream. And underneath that, there’s a peppy A.I. pop song whose lyrics blithely mix authoritarian come-ons and intimate promises. “I am watching you, ’cause I know what’s good for you … You want to be safe, and I’m here for you.”
Mainstream documentaries love to slap a pop song over their closing credits, leaving the audience with a vague feeling of uplift and perhaps giving the film another shot at an Oscar. But Orwell: 2+2=5 brings back that A.I.-created banger, letting it play out at length. Before long, the A.I. is effectively singing its own praises, talking about how it can keep you safe and occasionally dropping the words I love you into the middle of an unrelated thought. But it’s also a chilling callback to the final line of Orwell’s most famous novel, in which an all-powerful government demands not just to be obeyed or respected, but to be loved.