Long before human robots stroll the streets with us, they’ll be posting and socializing right alongside us on the internet, piercing the meniscus between reality and unreality. They are already doing it. Yumi, a new brand ambassador for Japanese cult skincare brand SK-II, is the latest member of a growing gang of digital humans and faux-influencers. What is most remarkable about her is not that she is a fake model doing brand work. She’s not an influencer slinging #spon or hosting an account takeover. Yumi is a one-brand gal and will belong to SK-II, forever.
She is the product of a collaboration with Soul Machines, a New Zealand startup that uses artificial intelligence to create digital humans that look and move like we do. Built on Google’s natural language processing platform, Yumi’s main purpose will be to advise SK-II fans on how to take better care of their skin, says Soul Machines CBO Greg Cross.
“She’s targeted to talk to twentysomething-year-old Japanese women about how they look after their skin,” he says. Rather than hire an army of people to answer beauty questions around the clock, SK-II is unleashing a single artificially intelligent woman to simultaneously answer beauty questions from customers all over the world at any time on the company’s website.
To date, Soul Machines has created digital humans for AutoDesk, Mercedes Benz, ABC bank in Bahrain, and Bank of Scotland among others. The bots have served as customer service representatives and, somewhat ironically, as human resources training assistants, onboarding new employees. Yumi will be performing some elements of customer service, but she’s also the new face of SK-II.
Brand ambassadors or mascots like Progressive Insurance’s Flo or Toyota’s Jan are nothing new. Verizon once had Paul Macarelli—the “Can You Hear Me Now?” guy—but alas, Sprint offered a better deal, so he now represents one of Verizon’s competitors. The “savage” acquisition of Macarelli by Sprint represents just one of the problems of working with humans. Contracts are finite, disagreements can erupt over pay, and humans are prone to having opinions, emotional outbursts on Twitter, and encounters with the law. Digital humans, on the other hand, won’t pose the same problems to brands that real humans do, and they have the potential to cost less than their flesh-and-blood ambassadors.
“When we go talk to big companies, they don’t really blink an eye,” says Cross. “Using a help desk to do customer service they could be paying an average cost of let’s say $15 per interaction—I can do that for it for a dollar already.” At the moment, clients pay annual subscription fees between $150,000 and millions for very high volumes, according to a representative for the company.
Yumi is based on a real person. Soul Machines can use detailed face scans to build its creations, though it can also generate human faces without a reference, conjuring new people seemingly out of nothing. SK-II declined to provide details about the woman Yumi is based on or how the company is compensating her for the perpetual use of her likeness.
On appearance alone Yumi seems to fit neatly into fashion’s current fascination with women of color, particularly those who don’t easily fit stereotypical racial profiles. Digital influencer Lil Miquela is part of this trend, as is Shudu, who was created by a 28-year-old white photographer named Cameron-James Wilson. While the trend is leading to more visibility of non-white faces, often these models still have features prized by European beauty standards. (SK-II plans to debut Yumi in Japan before bringing her to the U.S. and other markets.)
The Procter & Gamble-owned skincare brand could have asked Soul Machines to build a digital human entirely from scratch, but there was the possibility that she might not feel authentic.
“We wanted to make Yumi as lifelike as possible,” says SK-II CEO Sandeep Seth.
While Yumi may start out as a scan of a real human, she will develop her own personality and movement over time, he says. For now, however, not much differentiates Yumi from a creation like Miquela, a synthetic Instagram influencer created by L.A. ad agency Brud in 2016. Looking closely, Yumi looks more detailed, more 3D, less airbrushed than Miquela, but their movements are not so discernibly different, at least not to my eye.
Cross argues otherwise, insisting that the technology behind his digital humans is substantially different from the current roster of CGI influencers. Soul Machines has focused on creating artificial intelligent movement, so that its digital humans move and react on their own rather than scroll through a series of pre-programmed gestures. Cross sees these subtle reactions as the root of human personality rather than just the words we string together.
“The implementations of artificial intelligence are still relatively simple,” says Cross. But there’s a big shift coming, he says. Current commercially available technology is able to understand human language and offer scripted responses—think Siri or Alexa. However, the coming wave of technology will be able to not only understand natural language, but generate unique responses to it, he says. A recent example of this technology is Google Duplex, which makes calls on behalf of users to book restaurant reservations and other appointments and sounds remarkably human. It was spellbinding when it debuted last year, though reporting from The New York Times in May indicated its very real-sounding artificial humans are still very much in development.
“In the next 12 months, we’re going to move from [natural language processing] engines to [natural language generating] engines, where conversational content is getting created on the fly from data—and for these scripted chatbots that’s the first point at which they break.”
Cross referring not only to text bots, but to other computer-generated digital humans. If a gesture feels stilted, or if lips fail to adequately form the proper shape when speaking, the believability of these bots is broken. For brands, that sense of credibility and human connection is crucial.
Last year, SK-II launched a chatbot to help answer questions that customers might have about how to use a product they bought or how to integrate it into their existing regimen. “What’s missing from these chatbots, which are much more text based, is that human connection,” says Seth. He’s hoping that Yumi will be able to foster more personable interactions.
Creating a lifelike presence that truly feels human remains an incredibly difficult task. Last November, China state news agency Xinhua debuted an artificially intelligent news anchor who was based on a real newscaster named Zhang Zhao. The idea was that an AI newscaster could just as easily read written text as a human newscaster, and do it anywhere and at any time. But while the computer-generated anchor looked very real, his movement was ever so slightly stiff and his voice was completely monotonous with no inflection, no character. Could he make a human audience love him?
I ask Seth what personality characteristics he wants Yumi to have. “One is super intelligence,” he says, meaning that Yumi needs to be able to offer beauty advice that isn’t necessarily specific to the company’s products, but she also needs to reflect the brand’s knowledge and principles. “If I look at the core value of SK-II, it’s really all about authenticity.”
We are running out of time on our call, so I don’t have a chance to tell him that beauty advice and brand knowledge are not quite the underpinnings of a realistic human personality. But perhaps brands don’t really want their robots and fake humans to have real personality. After all, character is what contains our flaws and quirks, the stuff that makes us laugh or panic under duress. Passing the Turing Test may prove less important than the other benefits of a faux workforce. Not only can Yumi be available to anyone at any time, she doesn’t need to be replaced, only upgraded as the technology improves.
“Yumi is forever,” says Seth.
Updated with specific pricing information from Soul Machines