The AI program Midjourney has the ability to not only generate images, but also analyze them and provide descriptive prompts that could have been used to create them. We chose some of the most well-known photographs from the past century and had Midjourney generate prompts for them. These prompts were then fed back into Midjourney, resulting in the images you see in this story which were created in a matter of seconds.
However, can you distinguish between what is genuine and what is false? Test your knowledge with our quiz before proceeding with the reading.
Name it The Story of Two Selfies.
Soon after two Indian wrestlers were detained in New Delhi for protesting against purported sexual misconduct by the head of the country’s wrestling governing body, two nearly identical pictures of the pair started to circulate on the internet.
The two women were seen inside a police van with their team members and officers. However, in one photo they appeared unhappy while in the other, they were enthusiastically smiling – as if the arrest was just a performance.
The image of the wrestlers smiling spread quickly on social media for hours, shared by those in favor of the federation president. However, journalists and fact-checkers, as well as the two women depicted in the photo, called it out as fake. It was not until later that a comparison of their smiles to previous photos revealed that they had been digitally altered. This was most likely done using readily available software like FaceApp, which utilizes AI to edit images.
Stories like this one point to a rapidly approaching future in which nothing can be trusted to be as it seems. AI-generated images, video and audio are already being deployed in election campaigns. These include fake pictures of former U.S. President Donald Trump hugging and kissing the country’s top COVID adviser Anthony Fauci; a video in Poland mixing real footage of right-wing Prime Minister Mateusz Morawiecki with AI-generated clips of his voice; and a deepfake “recording” of the British Labour Party leader Keir Starmer throwing a fit.
Although these instances were swiftly revealed or disproven, they will not be the last. Future occurrences will be even more difficult to detect as artificial intelligence technology advances. This will lead to an increase in political misinformation and hinder the ability to verify everything from political blunders to atrocities committed in distant conflicts. The term “photographic evidence” may soon become obsolete in our modern era.
Prompt generated by Midjourney: che guevara in a red vest, hat and jacket, in the style of strong facial expression, keith carter, associated press photo, olive cotton, hans memling, movie still, blink-and-you-miss-it detail –ar 99:128.
The image has been illustrated by Ellen Boonen and Midjourney.
The coming flood
The potential danger posed by AI is not completely novel. Altered images have been around since the early days of photography. The earliest known instance dates back to 1840, when Hippolyte Bayard, a French inventor, produced a photograph claiming to be the first ever taken. As a form of protest against not receiving credit for his achievement, he created a picture depicting his own supposed death, titled “Self-Portrait as a Drowned Man.”
Some examples of fraud that occurred before the digital age include photographs from 1917 featuring small fairies with wings that tricked the creator of Sherlock Holmes, Arthur Conan Doyle, as well as the Soviet leader Joseph Stalin’s habit of removing disgraced officials from black-and-white photos.
In recent times, tools such as Adobe Photoshop, which has become a common verb, and filters on social media platforms have made visual manipulation accessible to everyone. What AI is accomplishing is elevating it to a new standard.
New advancements in image creation, such as Midjourney, DALL-E 3, and Stable Diffusion, offer three significant contributions. First, they make image generation more accessible by eliminating the need to learn complicated software like Photoshop. Instead, users can simply write a few lines and wait a short amount of time. Second, these tools produce high-quality images using AI technology, even for unrealistic subjects. And finally, with the potential for mass production, we can expect a flood of fake images created at an industrial level.
Most experts in disinformation are not overly concerned about the possibility of AI-generated images deceiving large portions of the population. While some people were tricked by a photo of the pope wearing a bulky white jacket or a blurry image of Julian Assange in his prison cell, these were not highly impactful. In May, a tweet featuring billowing black smoke coming from the Pentagon gained some traction, but it was soon proven to be false.
Prompt generated by Midjourney: a group of children run towards the burning sand, in the style of distorted bodies, dao trong le, demonic photograph, war scenes, controversial, chalk, candid shots of famous figures –ar 32:19.
The image was created by combining a photo and illustration, credited to Ellen Boonen and Midjourney.
According to Carl Miller, a researcher at the Centre for the Analysis of Social Media, the use of AI significantly reduces the barriers to entry and enables more convincing and innovative forms of deception. However, this ultimately leads to an increase in spam. Miller also notes that the public already has a heightened level of skepticism towards impactful images that originate from untrusted or unfamiliar sources.
The true threat may actually be the opposite of the issue at hand: a society that is becoming more and more hesitant to trust their own perceptions. The images circulating in March of ex-president Donald Trump being taken into custody were obviously fabricated. However, what about the photographs of Yevgeny Prigozhin, a Russian mercenary leader, sporting various absurd wigs and false beards? (As it turns out, they are probably genuine).
Benjamin Strick, director of investigations at the Centre for Information Resilience and an experienced practitioner of open-source intelligence (OSINT), expressed concern about the prevalence of fake news and discussions surrounding AI. He believes that this has led to a widespread skepticism towards information, particularly in authoritarian regimes where inconvenient truths are dismissed as lies or Western propaganda.
In June, there was an occurrence where pro-Kremlin social media accounts spread misinformation about the supposed death of Ukraine’s military intelligence chief, Kyrylo Budanov, dismissing a video of him as a “deepfake.” (Budanov is actually alive and well and has even given TV interviews.) As the conflict between Israel and Hamas intensifies, be prepared for more deceptive tactics of spreading false information, making it increasingly difficult to determine what is true.
Strick fears a potential future where authoritarian leaders manipulate and exploit the concept of the “liar’s dividend.” Coined by American scholars Robert Chesney and Danielle Citron in 2018, this term refers to the tactic of dismissing any visual evidence of human rights violations or atrocities as fake. According to Strick, this allows dictators to easily claim that all evidence has been fabricated by artificial intelligence.
In order to combat the issues arising from AI-produced images, various companies such as Intel, OpenAI (the creators of DALL-E 3 and ChatGPT), have created tools utilizing AI to identify these types of media. Some developers have even taken measures to mark their images with concealed or obvious watermarks as a way to indicate that they are artificially generated. However, these detection methods are not foolproof and as technology continues to advance, efforts to enhance these tools may become an ongoing challenge.
Tech corporations and media organizations have also joined forces through initiatives like the Coalition for Content Provenance and Authenticity (C2PA), in which organizations such as Microsoft, Adobe, the BBC and the New York Times are attempting to find ways to tag authentic images with encrypted information including its history, authorship and any modifications that might have been made to them.
Prompt generated by Midjourney: a group of men are sitting on a ledge and looking down, in the style of working class subjects, iconic imagery, anthony thieme, constructed photography, gerald harvey jones, steel/iron frame construction, captivating documentary photos –ar 64:4.
This image was created by Ellen Boonen and Midjourney.
Eric Horvitz, Microsoft’s chief scientific officer and leader of Project Origin, one of the coalition partners, posed the inquiry: “Is it possible to develop technologies that will synchronize every photon detected by a camera with what is being presented to consumers, while also recording any modifications made in the process?”
According to a recent report from the Royal Society, the leading science academy in the U.K., there is a potential risk that reputable media organizations using these systems could be seen as forming “knowledge cartels” that hold a monopoly on truth. This could have negative consequences in today’s highly divided political climate.
Carl Miller stated that he does not believe a scenario where everyone will have faith in the BBC is feasible. Instead, he predicts that we would experience a fragmented reality, consisting of various versions of truth.
Horvitz is optimistic that authentication tools will eventually be accessible to smaller groups and individuals, as technology and photography companies begin incorporating them into their software and devices. However, this process will likely take some time and there is no guarantee that it will be universally available.
In 2018, the BBC utilized videos shared on Twitter to uncover the soldiers responsible for the unlawful murder of four individuals in Cameroon. Presently, a significant amount of in-depth coverage of the conflict in Ukraine heavily relies on a vast collection of photos and videos uploaded to the internet by various sources, ranging from Ukrainian adolescents to Russian mercenaries. However, even if these sources had advanced verification technology at their disposal, would they be willing to disclose their identities and attach them to incriminating evidence of human rights violations or acts of war? (C2PA is working on developing anonymous authentication, but this may bring about additional complications.)
According to Strick, the abundance of artificial intelligence images means that the future of evidence will become less reliant on digital sources. In order to confirm information found on the internet, it will be necessary to gather evidence from physical sources. He believes that this will return journalism to its original purpose and that open-source intelligence (OSINT) should be combined with investigative reporting and strong local journalism.
A person positioned on a platform with wires connected, resembling modern art from the Middle East and North Africa, confessional, photography using Lady Grey film, art that involves performance, from 1970 to the present, incorporating fabric, with a speckled appearance – AR 95:128.
An image created by combining a photograph and illustration, credited to Ellen Boonen and Midjourney.
Meanwhile, there are indications that trust may already be declining. Greg Foster-Rice, an associate professor of photography at Columbia College Chicago and a member of the university’s AI task force, has observed an unexpected trend among his Gen-Z students in recent months.
The trend is shifting away from digital photography and towards traditional analog methods, such as using film and developing pictures in dark rooms. According to Foster-Rice, there is a growing interest in the retro style, possibly because it is perceived as more genuine and connected to reality.