Dope Francis, Disaster Photography and Generative Artificial Intelligence
Earlier this year, Pope Francis caused a global stir as images of him “swagged out” in a knee-length silken whiten designer puffer jacket circulated wildly on social media. The image sparked humorous commentary about the pontiff’s new cool, chic look. The problem with the viral image, though, was that it was fake – dreamed up by a 31-year-old construction worker from Chicago (@skyferrori) and executed through Midjourney, a generative artificial intelligence (GAI) programme that can create images based on text descriptions provided by users.
Web culture expert, Ryan Broderick, described this viral event as perhaps the first “mass-level AI misinformation case” but it certainly isn’t the only. Since then there have been more AI images associated with news stories, such as the arrest of Donald Trump and of Emmanuel Macron running from a protest. These images, while photorealistic, portray questionable scenarios and are, therefore, quickly called out as fake. More recently, however, Australian media outlet Crikey reported the use of GAI images in news stories about the ongoing Israeli military assault on Gaza. These images, created by independent content creators, were available for purchase via the U.S. software company, Adobe’s, stock images platform. Crikey noted that although most of these images had been marked as “generated by AI” on Adobe Stock, news stories that used these images didn’t indicate the same.
Following up on the story, The Washington Post also reported the existence of fifteen thousand fake images associated with “Ukraine war”, hundreds of AI images of Black Lives Matter (BLM) protests that never really happened, and dozens of similar images of the Maui wildfires. While the incident of the “Dope Francis” is relatively harmless and the source of much amusement, the circulation of GAI images of a news event is far more dangerous.
For GAI programmes – like Midjourney or Adobe’s Firefly – to produce “new” images, they must first be trained on existing images and language models so that they can associate visual data with specific textual data. So, when a user inputs, for example, “Maui wildfire”, the GAI programme synthesises information it has been “taught” (that is, has been fed into it) from masses of already existing images tagged as “Maui” (or, more broadly perhaps, Hawai’i, Pacific islands, etc..) and “wildfires” to produce new visuals. GAI images are thus regurgitations – new arrangements of an old form.
The danger of GAI images, of course, is that they can be used to misrepresent events – like the image of Macron running from a crowd or the “fake” BLM protest. The former instance constitutes a clear case of misinformation – representing something that never happened. In the case of the latter, one might argue that, given several BLM marches have in fact occurred, the AI image is not necessarily misinformation but merely a simulation. But such simulations, too, are dangerous because they project a dominant, often one dimensional, version of a story. Images of one conflict become interchangeable with another – GAI images of both, Gaza and Ukraine, show, for example, children running from bombs, cityscapes blown to rubble – images that are, in turn, reminiscent of scenes from Afghanistan or Vietnam. The children and landscapes might differ but the overall nature of the scene remains the same.
Images of conflict, or disaster photography, in general, have tended to focus on visuals of devastation and abjection in order to create a sense of shock or empathy within the public eye. But, as some critics have argued, these images can be dehumanising – fixing victims and survivors of calamities as broken and in need of rescue. This is of significant concern for people of colour and people in the global south who are, by and large, already imagined to be “disempowered” or “damaged”, constantly mired in violence and catastrophe. Since GAI programmes are trained on already existing images that portray this sense of brokenness and terror, they continue to spit out more of the same stereotypes.
It is worth noting that the rationale underlying the design and operation of GAI programmes is ease, efficiency and profitability (see for example recent debates on the use of GAI in Hollywood). Media outlets benefit from GAI as it reduces the cost of photojournalism while also keeping photojournalists safe from the threat of violence. Similarly, creators of GAI images are interested in monetising the prowess of new technology. On the one hand, GAI creators benefit from the work of photojournalists, the formers’ creations being a simulation of the latters’. On the other, the work of photojournalists is not merely to document events for entertainment or profit. Good journalistic work follows, ideally, from ethical and moral deliberations of what this work means and the responsibilities in undertaking it. GAI creators remain unencumbered by such considerations.
Contrary to popular visualisations of war, professional and amateur photojournalists in Gaza, for example, have been disseminating images of life and livingness amidst the ongoing violence – break dance classes, people playing with children, people with their cats or playing in the rain. This is an equally poignant reality of living through war, one that chooses to recall life instead of death. These images remind of the complexity of life – what persists and what is truly at stake – rather than repeating what we already know, through news and entertainment, of what war supposedly looks like.
The challenge posed by GAI, then, is not only about the threat of misinformation or the loss of jobs, as it is commonly understood to be. Rather, the displacement of human labour by AI dissolves the possibility of a compassionate and complex relationship between journalists and their subjects, and by extension, the audience. The basest form of photojournalism treats its subjects as spectacle – objects of fancy to be consumed by viewers – and GAI images risk repeating this cruelty. While AI will have a continuing role in journalism, it is important that its use does not degrade the ethical responsibilities of the profession, and further erode the dignity and agency of those facing catastrophe. This necessitates that we develop a new understanding of the role of AI in society. It demands a shift in the principles that guide the development and design of AI – so that this is driven by the need to facilitate ethical practice rather than by motives of efficiency and profit that ultimately simulate or circumvent moral deliberations.