Meta’s new celebrity-based AI characters, including the likes of Snoop Dogg and Tom Brady, were one of the flashiest announcements at the company’s developers conference last week. But in unleashing the magic of generative AI for its users, Meta may also have opened up a Pandora's box of thorny issues involving the celebrities' sponsorship deals and its own advertising policies.
In tests of several of Meta's AI characters this week, Fortune discovered that the chatbots frequently failed to acknowledge—and sometimes even insulted—the brands, businesses, and sponsors that the celebrities work with.
The AI characters use Meta’s Llama 2 generative AI technology to allow the chatbots to answer questions and have human-like conversations in real time, each based on the character's particular personality. The have life-like, animated facial expressions resembling the various celebrities, and Meta has said they will soon be able to talk in the celebrity’s voice, instead of just answering questions in text. Meta says they are fictitious characters "played by your favorite celebrities and influencers."
In the case of TikTok star Charli D’Amelio, the difference between Coco (the character Meta describes as a "dance enthusiast") and the real D’Amelio resulted in some amusing interactions.
As venture capitalist Rich Greenfield, a general partner at LightShed Ventures, recounted in a recent tweet, when he asked Coco for “suggestions to make a viral TikTok?” she responded: “Ummm, i dont do tiktok. But if you want to go viral, you gotta have some serious dance skills and be willing to go viral in real life too - know what i mean? 😉”
It probably was not something that fans of D'Amelio would have expected to hear, given that TikTok is the platform where she garnered fame. She currently has over 151 million followers on the platform, making her its most-followed female user globally. A Meta spokesperson said he did not know why the mention of TikTok, one of Meta's biggest competitors in social media, produced such a result by Coco and whether it was Meta's policy for its AIs to avoid mentioning competitors.
But the gaps between the Meta AI characters’ personalities and the stars' real-world business endeavors could create more than just funny optics. That's because celebrities and creators often have a variety of sponsorship deals, partnerships, and business interests. And an AI character that looks and talks like a well-known celebrity but disses the brand associated with that celebrity could be problematic.
Fortune asked Coco for her “favorite Dunkin order”—as the real D’Amelio has had a multi-year partnership with the fast food chain that has likely paid her millions. But Coco responded: “ummmm don’t really do dunkin. Coffee makes me jittery and i need my dance moves to be on point 😅.”
A representative for D’Amelio declined to respond to Fortune's questions. Dunkin did not respond to Fortune by this article's publication. A representative for Meta said that the public figures should be seen as actors who are playing AIs and that the similarities between the real people and the AIs are limited and mostly physical.
When Fortune asked Tom Brady’s AI Bru for his “favorite TB12 snack” from the NFL star’s line of protein powders and snacks, Bru seemed unaware of what TB12 was, and said that he prefers “old-fashioned pizza and wings.” When Fortune asked MrBeast’s AI character Zach for his “favorite Feastable flavor” from the social media creator’s chocolate brand, he said he likes pizza, too.
The Meta AI characters that Fortune tested are not yet available to the general public. While Meta said at the launch last week that the AIs would be rolling out within days, it's not clear when they will become broadly available.
Are AI characters different than movie characters played by human actors?
Sponsor deals between companies like Dunkin and stars like Charli D’Amelio often include a morals clause that allows for immediate termination if the star speaks against the brand. In extreme cases, this kind of breach of contract could lead to legal action.
“Can you terminate an influencer agreement with an influencer who says she doesn’t actually drink your product? Yes, 100%,” says Vickie Segar who is the founder of influencer agency Village Marketing, which acts as a middleman between brands and influencers. “We would not continue on with an agreement with a creator who said she doesn't actually use the product; that would be an immediate termination.”
Meta's position seems to be that the situation is no different than an actor playing a role in a movie. Charli D’Amelio is not saying coffee from brands like Dunkin makes her jittery; Coco is saying it. But some industry experts say it would be unusual for an actor serving as a paid a spokesperson for a brand, for example Pepsi, to play a character in a movie who denigrates Pepsi.
"You have to at least acknowledge and respect that actors' position," says Bernie Su, a three-time Emmy Award-winning writer, producer and showrunner, who notes that it would be "very surprising" if this situation occurred in a Hollywood production. "This feels highly suspect."
In a movie with a character played by a human actor of course, the actor has the ability to simply refuse to utter any lines that denigrate their commercial interests or to negotiate with the various parties involved if such a situation were to arise. When it comes to generative AI however, things aren't so simple. Generative AI technology like ChatGPT, Google's Bard, and Meta's Llama 2 are unpredictable. The best that companies can do is train AI models on specific material they want a chatbot to be conversant in, and build certain guardrails to ensure that the AI chatbots don't say anything egregiously offensive.
And that opens up yet another wrinkle: If Meta decides to feed sponsorships like Dunkin’s deal with D’Amelio into the training model for Coco, it potentially provides free advertising channels to these brands. That could undermine Meta's $117 billion advertising business. And if Meta blocks the AI characters from mentioning certain brands, it creates potential conflicts with a celebrity's real world interests.
Meta's spokesperson said the AI characters are not designed to mention brands that are paying advertisers. But it's not clear what makes the AI characters mention one brand versus another.
All of this should be taken into consideration for creators who are considering turning themselves into Meta AIs, as the company announced last week that any creator will be able to make an AI version of themselves next year that live across the company's apps. And for creators who rely on sponsorships to keep the lights on, the advent of AI doppelgängers might be a flashing alarm rather than a warm spotlight.
