AI-generated virtual influencers like Mia Zelu are a huge problem
The leggy blonde appears to lead a life of luxury, sitting in the crowd at Wimbledon one week and in the audience at Coldplay the next. But she’s not real and she’s fooled the world.
Business
Don't miss out on the headlines from Business. Followed categories will be added to My News.
The Pimm’s looked real as the setting sun over a summer’s day in southwest London backlit the sliced cucumber, orange and strawberry that garnished the cocktail.
In another shot, a woman played with a tendril of her blonde hair while seated beside an elderly gentleman – who, like Hugh Grant, appeared to be caught napping – in the stands at Wimbledon.
Meet Mia Zelu, the AI influencer who posted these images under the caption “still not over the event … but the party’s a whole other game”.
In case you missed it – and it was hard with all those little details that created a degree of verisimilitude – Zelu is fake.
She’s an AI creation, a bot. Still, she has amassed 165,000 followers from just 55 posts on Instagram, fooling the world.
It may appear harmless fun on social media but, according to global photography agency Getty Images, it’s a big problem.
Here’s why.
The explosion of AI has made it easier than ever to deceive people online. With a few simple verbal prompts, anyone can create photorealistic images and videos – even with sound. Bots that appear human risk spreading falsehood and misinformation, potentially undermining democratic elections.
More troubling, the technology is being used to create pornographic and abusive material. Graphic fake images of Taylor Swift were shared across X in January last year.
In other cases, it has just wasted people’s time as social media influencers, well, grow in influence. Just look at an elderly Malaysian couple who travelled 4½ hours across the country to go on a cable ride they saw online, only to find it didn’t exist. It was generated by artificial intelligence.
In regard to Zelu, Getty Asia-Pacific head of creative Kate Rourke said thousands of people mistook her for being real, despite her bio saying she was an “AI influencer”.
Ms Rourke says this highlights the need to clearly label AI-generated content, warning that Zelu “isn’t just a one-off trend”.
“It echoes findings from our VisualGSP research and image testing in Australia and New Zealand,” Ms Rourke said.
“Our findings showed that while 69 per cent of people could spot at least one AI image, a staggering 95 per cent also mistakenly thought real images were AI-generated.
“It’s no surprise that 68 per cent of people told us that, in general, they can’t tell if an image is AI-generated. This highlights the need for clear labelling for AI-generated content. In fact, 90 per cent of consumers agree that building trust and maintaining brand integrity (is critical) in the age of AI.”
Ms Rourke said Getty’s study showed how the current social media visual trend of a hyper-polished aesthetic converges with the advancements in AI.
“People are used to seeing curated, almost perfect images from human influencers, achieved through extensive editing, filters and professional photography. Zelu, being entirely AI-generated, naturally embodies this idealised flawless skin, perfect lighting and picture-perfect poses,” she said.
“AI-generated content is currently ‘hyper-polished’, allowing it to seamlessly blend into the existing visual standard that many people have come to expect. The fact that thousands of fans mistook her for real, despite her AI disclosure, points to how our eyes have become conditioned to this hyper-polished aesthetic on social media.
“Ironically, AI’s ability to mirror the hyper-polished aesthetic seen on social media has sparked counter-trends in the visual world, with more people wanting to see more candid and unfiltered visuals, the ‘anti-polish’ or ‘grit over gloss’.”
Some companies already label AI content. Samsung will automatically watermark photos that have been edited using AI – even something innocuous such as removing a lens flare dot – as AI content.
Canva has made its debiasing technology freely available to other developers in order to help stop artificially generated imagery being used for political deepfakes, hate speech and other abusive material.
Ms Rourke said what the issue comes down to is building trust and doing things ethically.
“The burden of discerning real from synthetic content shouldn’t fall solely on the user,” she said. Creators of AI models and adopters of AI bear a significant responsibility to encourage transparent, ethical and responsible use of generative AI.”
More Coverage
Originally published as AI-generated virtual influencers like Mia Zelu are a huge problem
Popular suburban pub for sale
Pub group GM Hotels is continuing its sell-off in SA, with a landmark Adelaide north-eastern suburbs establishment expected to attract interest from local and interstate groups.
‘Tone deaf’: Aussie brands on tightrope with kiss-cam scandal memes
Aussie and international brands have rushed to piggyback the Coldplay kiss-cam scandal, creating a flood of memes riffing off the viral moment - but can this sort of marketing backfire? Read what the experts say and take our poll
Pub baron Jon Adgemis’ $1.5bn debt and zero savings
The former owner of a string of pubs across Sydney and Melbourne owes close to $1.5bn to creditors, according to bankruptcy trustees, amid warnings his fast cars, fancy clothes and $3.79 in the bank won’t be enough to settle his debts.
Regulators need to follow the commission trail to protect Australians
Politicians have conceived bodies and acts of Parliament to minimise consumer investment losses. All contributed to a better investment community, but none of them were able to stop consumers being defrauded.
This chart made Buffett millions, it’s screaming Wall St is cracking
Billionaire Warren Buffett’s go-to market signal has hit the most extreme levels in history, flashing bright red a market meltdown is looming. It’s telling Aussies to look local, writes ASX Trader.
Tech retailer probed for insolvent trading amid $30m collapse
The administrators of Australian computer and video game retailer Mwave have revealed the business may have traded while insolvent, as details of the company’s $30m collapse are exposed.