An image of Pope Francis went viral in late March, which is nothing new for the iconic figure. But this image differed from the other photos you can find online about the Pope. In the photo, the Pope is wearing an ankle-length, puffy white jacket from the high fashion brand Balenciaga. And on his feet, very stylish and modern white sneakers. The image was posted to Reddit, and then it went viral, with scores of viewers believing that the 86-year-old leader of the Catholic church was in the midst of fashion evolution. 

What was different about this viral image of the Pope is that a 31-year-old man from Chicago created the image using AI. Pablo Xavier used the AI image generator Midjourney to create the photo of the dripped-out Pope. Xavier used the prompt “Catholic Pope Francis. Balenciaga puffy coat. Streets of Paris” to generate the image that tricked millions of people into believing the Pope had been photographed in Balenciaga. Xavier told the Chicago Tribune that he had been tripping on mushrooms when he decided to play around on Midjourney. He started using the platform after his brother died to create images of his brother as a therapeutic hobby. He posted the photos of the Pope to the Facebook group AI Art Universe, and then on Reddit, they went mega-viral. 

Just days before Xavier’s photos of the Pope went viral, AI-generated images of former president Donald Trump being arrested went viral when news broke that he had been indicted by the state of New York for payments he made to adult film actress Stormy Daniels. The images fooled millions on social media into believing Trump had been arrested when he had not. The journalist who generated the photos of Trump getting arrested was banned from Midjourney. One reporter called Xavier’s images of the Pope and the images that went viral of Donald Trump the first real mass-level AI misinformation case. The images fooled many on social media, but for some experts, the photos were easy to spot as AI-generated. 

Take the image of the Pope in the Balenciaga puffy jacket. If you look closely, the crucifix hanging from his neck only has one strand of the chain meant to wrap around his neck. His eyes and eyeglasses seemingly merge together. His fingers don’t quite reach the coffee cup he is supposedly carrying. One of the telltale signs of an AI-generated image is distorted or unrealistic hands. AI image generators are notoriously bad at creating images of hands. But, a recent update by Midjourney might make it harder for people to use hands as a calling card for AI-generated fakes. Artist Del Walker took to Twitter to display the capability of the v5 language model that Midjourney upgraded to and to caution users to be vigilant when viewing viral images on platforms like Twitter. Experts also pointed out that the fabric in the images was distorted and that the faces of the subjects tended to have a waxy sheen to them. 

The rapid spread of the photo of the Pope led some to raise alarm bells about the speed of AI adoption. Other than a few small telltale signs, the images were challenging to discern as fake, which worried some experts in the field about the potential for deep fakes to become increasingly convincing. As a result, some experts in technology urged companies to slow the adoption of AI technologies in what they refer to as the race to AI. Leaders like Elon Musk and Apple’s Steve Wozniak signed the petition in response to the release of ChatGPT-4 by OpenAI. ChatGPT-4 is OpenAI’s most advanced language model, yet that is capable of answering more complicated questions, scoring higher on exams, and completing higher reasoning tasks. According to OpenAI, GPT-4 is 82% less likely to respond to requests from users for banned content and 40% more likely to produce accurate, factual results than GPT-3.5. Despite reassurances from OpenAI, the petition gained signatures from thousands of industry leaders, warning companies against the unchecked adoption of AI tools because of the questionable ethics behind driving toward a future where AI replaces human-led jobs and asking companies to pause the development of AI tools with anything smarter than GPT-4 for six months.

Critics of the petition were quick to point out that there is by no means consensus amongst the tech and AI community about how to proceed as more complex learning models become available. One thing is clear, the call to adopt responsible AI solutions is stronger than ever, and as the field advances, the importance of using trusted AI products becomes clearer than ever as well. Educating everyday users about the hallmarks of AI-generated images or even text is an important new area of study. It will require thought leadership from those in the AI space to make sure that deep fakes, like the Pope’s undeniably cool fashion pivot, don’t fool so many people into believing something that isn’t true is, in fact, true.