Who would have thought we’ll see Mark Zuckerberg walking the runway in flamboyant clothes and not in his basic grey shirt, or see images of Pope Francis parading the streets wearing a Balenciaga-style white puffer jacket? We can now visualise these unlikely scenarios thanks to AI-generated images through easily accessible image-generating tools.
The incredibly realistic images of the Pope, Zuckerberg, Donald Trump, among others went viral on social media in the past couple of days. While netizens were fascinated by these images, they also expressed worries about how this could spread misinformation.
Here's a look at some fake AI-generated images that went viral recently and how to tell them apart:
Trump’s Arrest Fake Images Leave Netizens Stumped
Much before Donald Trump’s indictment on March 31, fabricated images of his arrest went viral on social media, opening a pandora’s box of numerous other viral AI-generated images.
Twitter user @EliotHiggins said that he used Midjourney v5 to create the viral image. "Donald Trump falling over while getting arrested. Fibonacci Spiral. News footage," was the prompt that was used to generate the fake image.
Higgins, in a tweet, claimed that Midjourney banned him from using the service after Trump's images went viral.
No, Putin Did Not Bow Down To Xi Jinping
Amid diplomatic talks between Chinese President Xi Jinping and Russian President Vladimir Putin in Moscow, a viral image showed Putin bowing down to Xi Jinping and kissing his hand.
Tech reporter, Amanda Florian explained in a Twitter thread how she deduced that the image could be a result of an AI program. At first, Amanda used reverse image search but couldn't find the source of the image. On running the image through an AI image detector created by Hugging Face, the image detector found the photo to be artificial.
"Xi's hair has some irregular noise and blurriness, which is usually a good indication that an image is GAN-generated," Amanda added. Other signs, according to the reporter, included a slightly hooded ear on Xi and some irregular bumps on Putin's ear.
When Pope Francis Puffer Jacket Moment Started An AI Wave
Perhaps the most-talked-about AI image so far, Pope Francis' fake image in a puffer jacket stole the limelight by making millions believe that he truly stepped out in a never seen avatar. According to TIME, a closer look at the image revealed that it is not a real image. "In his right hand is what appears to be a blurry coffee cup, but his fingers are closed around thin air rather than the cup itself. His eyelid somehow merges into his glasses, which in turn flow into their own shadow," the report stated.
Seeing the virality of the fake image, many other prompts were also shared by netizens. "The topic of the modern Pope generated by AI (#midjourney) is getting out of hand for us😂," a Twitter user wrote.
Midjourney Turns Matchmaker For Elon Musk
Twitter CEO was also made part of this "AI image gala" by Twitteratti when an image of Elon Musk with General Motors CEO Mary Barra went viral. Brian Lovett created the image using Midjourney. "This is an AI generated image. There are currently no reports of any communications between Mary Barra and Elon Musk. Mary Barra is married to Anthony Barra," the context under the image on Twitter read.
Elon Musk also responded to the image, stating, “Also … I would never wear that outfit(sic)."
AI Adds Colour To Zuckerberg's Wardrobe
Meta CEO Zuckerberg is known to wear plain, monochrome outfits. But a Twitter user @itsandrewgao used AI to create images of what it would look like if Zuckerberg walked the runway for luxury fashion brand Louis Vuitton. The image showed the Meta CEO in a shiny outfit on a runway.
Others followed suit. @LinusEkenstam posted images of Zuckerberg in bright yellow clothes saying, "Mark Zuckerberg doing rounds on the runway, pure fashionista."
Others created similar images of Twitter CEO Elon Musk using the same prompt.
Identifying AI-Generated Images
Midjourney, one of the AI tools that were used to generate the viral images, has a Code of Conduct, that asks its users to not "be a jerk". "Don't use our tools to make images that could inflame, upset, or cause drama. That includes gore and adult content. Be respectful to other people and the team." However, the scope for error still exists. So what can social media users do to ensure they're able to tell fake images from real ones and not further a fake narrative?
Speaking to BOOM, Sam Gregory, human rights advocate, technologist, and executive director at witness.org, said that there was a need to slow down on clicking the reposting or retweeting button when the intent and context of a piece of media was unclear. "The SIFT media literacy acronym is a useful way to remember this: 'Stop, Investigate the source, Find trusted coverage and Trace the original,' rather than let our emotions or first impressions lead us," he added.
Finding the source of the image is of utmost importance and can be done within minutes using a reverse image search or Google lens. If an image has been manipulated or edited, this practice can help one find the original image, other versions, or perspectives and also help in tracing the original context.
"We should remember that the bulk of deceptive images is still ‘shallowfakes’ - mis-contextualized or edited images, not deepfakes or AI-generated images. So our approach to suspicious images should aim to deal with both," Gregory added.
On the current state of generative AI, Gregory explained that in most of the images, there seems to be some sort of distortion, waxy skin, unrealistic shadows, and eye reflections that wouldn't make sense in the real world. Though AI tools at present do not flawlessly create text or logos within images, it is likely that AI will improve swiftly and may soon surpass all these limitations.
Emphasising the need for verification, Gregory said, "It’s better to encourage both traditional media literacy and also more robust AI media literacy so that people can develop a broader understanding of these systems instead of a false confidence that individuals can detect deepfakes."
Gregory doesn’t recommend relying on online detection tools that claim to be able to show if an image is synthetic or authentic, as most of them don’t work well and they can create false positives and negatives.
Growing Concerns About The Spread Of Fake News
While many people were able to identify loopholes in the fabricated images, some passed them on, believing them to be true. Considering the threats involved, there are growing concerns that as AI technology advances, the misinformation web will soon be wide enough, making it difficult to be brought under control.
Eliot Higgins, who created the fake images of Donald Trump's arrest, highlighted the vast dangers that AI and fake news peddlers can pose. He shared images of an AI-generated attack that could be shared, under the false pretext of "claiming a nuclear weapon has been used" to incite people in Ukraine and Russia.
Amanda Florian, a journalist highlighted the importance of investigating and fact-checking everything that the internet offers, commenting, "AI is incredible, but it can clearly be used in manipulative & harmful ways. I believe proper disclosure is needed if you're sharing these kinds of photos. Even if it's a joke or meme—disclose that (!!)(sic)"
Gregory said, "We need to push the responsibility onto the companies developing tech, the platforms and distributors, who should watermark the content and develop provenance and detection technology that can help us see that images were synthesized, manipulated and shared and ensure those traces remain with images as they circulate."
On the endless possibilities and potential of AI, a Twitter user asked, "How can any photographic evidence ever proven to be real or fake? Surely this will only get better?" Brian Lovett, who created the fake image of Elon Musk and Mary Barra, holding hands on the street, replied, "Within a few months it will be indistinguishable from reality."
Blaming a sudden influx of new users, Midjourney— the tool used for creating most of the viral AI images, has halted free trials of its service now, citing “extraordinary demand and trial abuse”.
