The celebrity-studded event of the Metropolitan Museum of Art’s annual gala, better known as the Met Gala, was lacking two of its biggest stars this year. Notably absent were Katy Perry and Rihanna, much to the disappointment of their fervent followers. Yet, despite their absence, the pair curiously appeared atop the Met Gala’s grand staircase, not in person, but through artificial intelligence-generated deepfake images that circulated across social platforms, duping some fans, including Perry’s mother, into thinking they had graced fashion’s biggest night.
The AI-crafted images of these luminaries surfaced online early Tuesday and spread like wildfire. Observant social media users and certain community note platforms quickly pointed out discrepancies that revealed the images to be, well, fakes. The viral image of Perry bedecked in a flower-strewn gown seemed genuine but for one telltale sign: the carpet adorning the museum’s famed staircase mirrored that of the 2018 event, a far cry from this year’s verdant, foliage-lined fabric.
Yet, the artificial intelligence’s tantalizing trickery was not lost on Perry’s own mother. Perry shared two of the false images on her Instagram along with a screenshot of a text, presumably from her mother, commending her on her stunning “appearance.” Perry humorously responded to her mother’s faux paus by warning, “lol mom the AI got to you too, BEWARE!” In the same Instagram post, Perry confirmed she didn’t attend the event due to “work.”
Simultaneously, an equally unsubstantiated image of Rihanna, resplendent in an elaborate white gown adorned with embroidered fauna, also took social media by storm. Despite Vogue’s earlier confirmation of her absence, followers were momentarily deluded by the deceptive image. Pervasive rumors of Rihanna nursing the flu did nothing to dissuade the false image’s propagation.
As intriguing as the origins of these artificial images are, identifying the source proves tricky. The uncannily realistic Met Gala backdrop suggests a highly advanced AI trained on countless images of prior events. Last year, severe legal action was taken by Getty, the Met Gala’s official photographer, against a London-based AI image generator, Stability AI, for illegally replicating millions of Getty’s images. Getty launched its own AI image generator trained on its collection, curbing attempts at generating controversial content.
This is certainly not the first time generative AI has fashioned bogus content. Deepfakes, from well-known personalities like Pope Francis to Taylor Swift, have seen a considerable online presence. Each occurrence underscores concerns about this technology being exploited for disinformation, identity theft, propaganda, and even manipulation of election outcomes.
Professor Cayce Myers of Virginia Tech observes that with the rapid advancement of AI technologies, the idea that “seeing is believing” is being swiftly overturned. The fact that even Perry’s mother was tricked by this, demonstrates the sophistication of this technology. While creating images of celebrities in fictional luxury gowns may seem trivial, the potential misuse of this technology is startling. The exploitation ranges from generating sexually explicit and abusive fake images to victims non-consensually appearing in deepfakes, with research showing the most harm being caused to women and children.
Additionally, in significant election years worldwide, experts are continually warning about the severe geopolitical repercussions that fraudulent, AI-generated content could unleash. As technologies rapidly evolve, the challenge lies in managing the benefits of generative AI while safeguarding consumers and society. An appropriate infrastructure demands corporate accountability, universal industry standards, and effective government regulations.
In a world where tech companies dictate the regulations around AI and its risks, governments are struggling to keep pace. Notwithstanding, the European Union has made strides by imposing the world’s first comprehensive AI rules this past December. However, their implications won’t become apparent until two years after the final approval.
