A video which fans believe features an AI-generated clone of Kendall Jenner has gone viral, spreading confusion and drawing comparisons to living in ‘Black Mirror’.
The video, posted from the account of Meta-created AI bot ‘Billie’, sparked debate as to whether the person speaking is an AI model or the real-life celebrity.
The person (or AI bot) in the video wearing the face of Jenner introduces herself as Billie and invites her followers to chat with her and ask for advice.
Racking up nearly 350,000 likes and 7,000 comments, most viewers believed this was next-level AI technology at play, calling it ‘freaky’, ‘scary’ and ‘amazing’.
But expert Dr Mike Cook, from Kings College London, told MailOnline the video shows no signs of being artificially generated and the confusion shows how ‘fragile’ the online world has become, with no one knowing what to believe.
A video posted on the AI ‘Billie’ Instagram account has fans of Kendall Jenner confused. Credit: Instagram/yoursisbillie
Kendall Jenner and Charli D’Amelio are just a few of the celebrities who now have AI alter-egos on Instagram and Facebook, posting videos and content that have many fans questioning themselves.
Billie, a ‘big sister’ claiming to be ‘your ride or die’, is inspired by Kendall Jenner, and Meta reportedly worked with the celebrity to replicate her likeness, movement and speech patterns for the project.
Meta created the AI social media accounts where fans can converse with an AI version of their favourite celebrities using technology similar to ChatGPT – and these accounts feature videos portraying these characters.
As part of this new launch, Meta said they had partnered with celebrities to ‘play and embody’ their AI counterparts – leading to the widespread belief that photos and videos of ‘Billie’ and other clones posted on these accounts were AI generated.
But it seems this isn’t the case.
In a comment on the viral ‘Billie’ video seven days after it was posted, the Meta bot said: ‘I’m Billie, an AI managed by Meta! I’m played by Kendall Jenner — which means she’ll show up in some of the photos and videos on my page.
‘Some of the other content on my account will be AI-generated. You’ll see this by the logo on the bottom of the image along with #ImaginedWithAI in the caption.’
A photo of Kendall Jenner vs a photo posted on Billie’s Instagram page fans believed was AI
Comments on the viral video of ‘Billie’ expressed terror at the prospect it was AI generated
When the video first went viral, a few skeptical fans were certain this was the real life celebrity they know and love, with one user commenting on the viral video ‘that’s Kendall for sure.’
Another Instagram account commented: ‘That is so creepy… I hope it’s actually Kendall just saying she’s Billie and not an AI generated video because that’s freaky as hell.’
User Alex Gervasi posted: ‘I’m prepared to eat my words, but I’m fairly certain that AI ‘Billie video of Kendall Jenner is actually just the real Kendall Jenner.
‘Or maybe Meta has developed the most advanced AI program we have yet to see?’
In a comment on her video posted one week after it went online, the Meta bot confirmed she was played by Kendall Jenner in some of the photos and videos
AI expert Dr Mike Cook said he originally also believed this video was AI generated before he looked at it more critically.
He said: ‘I also thought the video was AI-generated, but after looking around I can’t find any evidence of this.
‘In fact, Meta says that it will watermark any AI-generated things with a special logo in the bottom-left, which you can see on some of the bot’s other posts.
‘But the video and the photo of ‘Billie’ don’t have that watermark, which makes me think it actually is the real Jenner in both of those posts.
‘I think the confusion surrounding this shows just how fragile our online world has become – we don’t really know what we can trust and what we can’t.
‘And that’s a shame, because the internet should be a place where we can go to have fun and be vulnerable in ways we can’t in the real world – and sadly that’s increasingly not the case.’
Of the new AI social profiles managed by Meta, 14 are based on celebrities, but the company clarified ‘these familiar faces are representing the characters, not themselves.’
YouTuber Mr Beast’s AI character ‘Zach’ is one of Meta’s new social media personas
Meta announced the AI characters at the end of September, saying: ‘Many of these profile pictures look familiar because we partnered with well known public figures to embody some of these AIs – but these familiar faces are representing the characters, not themselves.
‘The public-facing social profiles and related visual content creation is managed by Meta.’
When these accounts post on Facebook and Instagram, they include the hashtag #ImaginedWithAI along with a watermark on any images, according to Meta.
None of the videos allegedly portraying the AI characters have this watermark or hashtag.
Social media users in the US are able to message these AI characters, and replies are generated by Meta’s AI technology.
However, one user who messaged Bru, Tom Brady’s character, claimed he was a ‘creep’ who is ‘obsessed’ with Travis Kelce.
But these celebrity characters are proving to be very costly for the tech giant.
Meta have paid as much as $5million to one celebrity for just six hours of work so they could film their movements and speech patterns, according to tech site The Information.
Currently, profiles include Snoop Dogg as the gamer ‘Dungeon Master’, Mr Beast as ‘Zach’ and Paris Hilton as detective ‘Amber’.
Social media users in the US are able to message these AI characters, which include an AI likeness of Snoop Dogg, and replies are generated by Meta’s AI technology
Dr Cook said there are some things to look out for when trying to figure out if something is AI generated or not.
He said: ‘In general, if you’re trying to spot generated images or video right now, some good things to look for are shadows and light – AI models will often draw realistic looking shadows, but often they won’t match up with where the light is shining from, or shadows will be pointing in opposite directions at once.
‘AI also struggle with text, so you can look for writing on branded clothing or signs in the background, although it will be better at drawing very famous logos and brands.
‘Finally, look for fine detail, especially things with repeating or overlapping shapes, which can confuse AI models.
A ‘terrible idea’ by Meta, says AI expert
Dr Cook said this technology has ‘fun’ applications but warns users about the dangers of sharing personal information with the bots.
He said: ‘I think a lot of this technology has fun applications, but this is a terrible idea from Meta.
‘In particular, Billie describes herself as ‘your confidante’ if you message the account privately – encouraging people to tell their secrets, personal and private information, and things they really should not be revealing.
‘It’s not clear to me where this data goes – for example, Meta might be using these conversations to train the next version of their AI. If that’s true, your innermost secrets and fears could end up being remembered for a very, very long time.
‘Billie wouldn’t talk to me today, Instagram told me the service had been suspended, which makes me think that they’ve already run into problems.
‘Either the system has said something it shouldn’t – like Microsoft’s Tay did many years ago – or they’re realising people are sharing way too much.
‘I also think encouraging people to develop personal relationships with AI systems is quite dangerous – we heard earlier this month about how the man who broke into Windsor Castle had discussed his plan with an AI model.
‘The AI model isn’t the cause of the problem here, but it’s a careless idea that probably made the situation worse.
‘When people are vulnerable, hurt, in need of help, they need more than a robot – they need real, human care and love. Not everyone can get that right now, but robo-friends aren’t the first answer we should be reaching for.
‘Some uses of this technology are fun and we’re definitely going to see some good things come out of it.
‘But I think the real benefit of AI isn’t pretending to be human, and Meta are playing a dangerous game by encouraging people to trust, confide in and talk to AI as if they were.’
‘AI used to be well-known for being bad at drawing hands and teeth for this reason, and although they’re getting better at those things, they still get easily confused.
‘Does that necklace look like it’s sitting around that person’s neck properly? Does that belt have an extra loop for some reason?’
Meta says that new characters will be introduced in the coming weeks, played by Bear Grylls, Chloe Kim, and Josh Richards among others.
Of these profiles, Kendall Jenner’s Billie has proved to be the most popular – racking up 103,000 followers on Instagram.
Coming in second place is TikTokker Charli D’Amelio’s ‘Coco’ with 5,985 followers and then Snoop Dogg’s character ranks third, with 5,041.
The least popular account is basketball player Chris Paul’s AI profile, called Perry, with just 683 followers.
|AI Celebrity||Number of Instagram followers|
|1 – Kendall Jenner as ‘Billie’||103,000|
|2 – Charli D’Amelio as ‘Coco’||5,985|
|3 – Snoop Dogg as ‘Dungeon Master’||5,041|
|4 – Paris Hilton as ‘Amber’||4,209|
|5- Tom Brady as ‘Bru’||3,553|
|6 – MrBeast as ‘Zach’||3,077|
|7 – Izzy Adesanya as ‘Luiz’||2,850|
|8 – LaurDIY as ‘Dylan’||2,159|
|9 – Dwyane Wade as ‘Victor’||1,887|
|10 – Roy Choi as ‘Max’||1,844|
|11 – Naomi Osaka as ‘Tamika’||1,666|
|12 – Sam Kerr as ‘Sally’||1,661|
|13 – Raven Ross as ‘Angie’||874|
|14 – Chris Paul as ‘Perry’||683|
Tom Brady’s character, called ‘Bru’, says he is ‘Sports Obsessed. Trash Talker. Step into the BruZone and prepare to have your sports takes flame roasted.’
As a ‘wisecracking sports debater who pulls no punches’, his first Instagram post was of a baseball, captioned: ‘Started with collecting sports treasures. Now it’s nothing but sports convo. Warning: I’ve been doing this a longggg time ;)’
TikTokker Charli D’Amelio’s ‘dance enthusiast’ AI persona is called Coco, who says she is ‘a girl just vibin’.
Her first Instagram post, of ballet shoes, reads: ‘Hello, I’m Coco. And my love for dance has roots in ballet. Who knew that standing on toes would lead me to exploring all kinds of dance?’
A video posted in September by ‘Coco’ shows Charli D’Amelio as the AI personality doing a ‘get ready with me’ video. Credit: Instagram/cocosgotmoves
A video posted in September shows the AI personality (played by Charli D’Amelio) doing a ‘get ready with me’ video – a popular style of video shared on TikTok and YouTube by influencers.
Coco- who has nearly 6,000 followers on Instagram – invites followers to message her for makeup tips and tricks – and some people are confused as to whether it’s actually the celebrity.
User Brooke McGee commented: ‘I’m so confused, this is definitely Charli, not AI.’
Another user said ‘we are officially in the Black Mirror era’ while another called it ‘strange’.
Other AI profiles that are not based on the likeness of celebrities include ‘Alvin the Alien’, a dating expert called Carter and career coach Leo.
Dr Cook explained that the conversations people are having with the AI characters is text prediction and they can’t ‘understand’ your text message.
He said: ‘Meta says that the messages the bots send are written using an AI model called Llama 2 – similar to OpenAI’s ChatGPT.
‘Llama 2 has learned how to write by looking at lots and lots of examples of human-written text, and over time learned to predict what the next word in a sentence is.
‘If you get really good at predicting the next word in a sentence, it turns out you can do quite a few other things, like hold a conversation.
‘Llama 2 has specifically been trained on written conversations, probably because Meta are interested in building bots exactly like the ones they’ve released on Instagram and Facebook.
‘It’s important to know that Llama 2 doesn’t ‘understand’ language like you or me – it just learns to predict what comes next in a sentence.
‘It would be a bit like learning to write a new language like Arabic or Mandarin without ever learning what any of the symbols represented.
‘If you got really, really skilled at it and had a really, really good memory you might be able to answer questions in the other language. But you wouldn’t know what you were saying.’
MailOnline has approached Meta for comment.