False Heroes and a Disappearing Soldier

From a US president styling himself as a king, church leader or ripped football player in online posts, to a tech billionaire depicting himself as a Roman gladiator in a meme published on social media platform X late last year. The world of digital images sometimes comes out with strange things. This is now possible thanks to AI image generators such as Midjourney and Dall-E. “They’re a bit like wish fulfillment machines,” says Roland Meyer, “you can enter a prompt and it’s immediately turned into an image”. And so, for example, at the touch of a button, a 21st-century American can turn himself into a heroic-looking gladiator in traditional armor from ancient Rome.

Or at least how artificial intelligence imagines such a gladiator to look. But “imagine” is not the right word. Because AI image generators sift through and synthesize huge amounts of data. They throw together completely different eras, styles and aesthetics more or less randomly, taking historical clichés to produce a new image of a fictional past. “In the Elon Musk meme, elements of 19th century historical painting are fused with the aesthetics of blockbuster movies, via ancient Rome and computer games to produce a timeless image of hypermasculinity,” says Roland Meyer, who incorporates both technical and visual history aspects in his analyses.

Roland Meyer is a media and visual culture researcher and has been DIZH bridge professor for digital cultures and arts since last summer. In this role, he works at both the UZH Department of Film Studies and at the Zurich University of the Arts. In his research, Meyer has explored the challenges of digital facial recognition and its analogue history. He has also looked at what it means if photos of our faces are evaluated by algorithms online and linked to profiling data to be exploited or used for surveillance purposes.

Meyer is also interested in how our understanding of images in the online world is shifting – in light of digital platforms, social media and generative AI. Indeed, the way in which we perceive images, how we interpret them, and what we expect from them in digital media, is changing massively. In the online world, the boundaries between desires, emotions and reality are starting to blur.

This is partly to do with the way social media works. Nowadays, users of social media platforms are used to reacting immediately to image content, for example to like or share it. Images posted on platforms like Instagram are therefore designed to drive engagement by eliciting these reactions. “These days, online images should above all have an emotive effect and spark spontaneous interactions,” says Roland Meyer. The focus is no longer so much on whether photos show an authentic slice of the world or not. “Often it’s more important that they are emotionally real for the viewer and that they align with their perceived world view,” he says.  

Many of the images shared online of the devastating wildfires in Los Angeles at the start of the year and of the catastrophic flooding in Florida last fall were AI generated. “Some people who shared such images emphatically said that they didn’t care whether they were genuine photos or not – as the disaster and human suffering were very real,” says Meyer. What was important to them was that the photos expressed their feelings and anguish, which they wanted to convey to others, and not so much that they accurately depicted specific events.

The AI-generated images portraying Elon Musk as a Roman or Donald Trump as pope have little bearing on reality either. The fact that these engineered images exist and that it is precisely men like Musk and Trump who are sharing them is no coincidence, according to Roland Meyer. “The far right loves generative AI, and we see that all over the world,” he wrote in an article for the online blog Geschichte der Gegenwart. In Germany, the far-right AfD party also uses AI-generated images in its propaganda, showing strapping young men, blond women with pigtails, and fictional, nostalgic, mostly rural idyllic families.

Meyer believes that the connection between such clichés, AI-generated ideals and right-wing political positions is based on an inner logic. He theorizes that commercial AI tools for image generation are not politically neutral tools, but that they lend themselves very readily to right-wing world views. “Many commercial tools are good at reinforcing stereotypes, particularly gender-based ones, and for example generating images of hypermasculine men and superfeminine women,” he says, “and they use image patterns from the past to consolidate them in completely ahistorical artificial visual worlds.” This gives rise, among other things, to historical myths that depict an idealized view of the past that never even existed. “This is why AI image generation is currently used on a massive scale by right-wing parties for propaganda purposes,” says Roland Meyer.

However, the visual culture researcher also sees other reasons why the political right is so keen to generate and share AI memes: “They upset their opponents – AI imagery has great potential for provocation.” They are therefore part of a culture war that also revolves around the use of historical images and ideas of history. In the United States, this conflict is also playing out in another context. “Earlier this year, the US administration had online image archives purged to reflect the political agenda, and diversity-related content erased,” says Meyer. This is part of the Trump administration’s fight against diversity, equity and inclusion (DEI).

In March, it was announced that the Pentagon had flagged 26,000 photos for deletion on the websites of the US military, as Meyer noted in a guest article for the German newspaper Frankfurter Allgemeine Sonntagszeitung. The images to be erased include photos of people who don’t fit the ideal of a male, white, heterosexual soldier. Photos of black military personnel in the Second World War have therefore been deleted. But photos from the 21st century have also disappeared from the internet, such as pictures of the first female US infantry soldier, Christina Fuentes Montenegro. “This is an attempt to idealize, sanitize and reframe history based on current political interests,” says Meyer, “our whole visual heritage is at stake”.

The purging of archives also has an impact on AI image generation. This is because ultimately AI models are trained on archived image data, which they use to generate new images. “If you prompt an AI tool to generate an image of a US soldier, this will likely already correspond more or less to the Trumpian ideals,” says Roland Meyer, “if the internet continues to be cleansed based on white supremacist categories, it’s conceivable that this will further reinforce racist and sexist prejudice in the future.”

Roland Meyer hopes that his research will raise awareness of such problematic relationships in the world of digital images. Going forward he would like to carry out further research into how AI changes our view of archives and history. A key basis for this is the interdisciplinary exchange at the new Center for Arts and Cultural Theory (ZKK), which was set up in 2024 and includes researchers from UZH and Zurich University of the Arts.