- Generative AI has enabled advertisers like Levi’s to create BIPOC-presenting models for cheap.
- This practice has been sharply criticized.
- Some advertisers have created internal guidelines to make sure AI-generated models don’t undercut BIPOC talent.
Advertisers using generative AI to depict BIPOC human models are raising alarm bells among critics who say this use case is Digital Blackface.
When Levi’s said in April it would use generative AI to increase “the number and diversity” of its models, it was accused of taking away opportunities from diverse models and production talent.
Non-white models spoke about how they book fewer jobs in a white-dominated industry.
“People of color don’t get as many bookings or don’t get requested nearly as much as white models,” Efosa Uwubamwen, a Black model, told Insider in a previous story.
There are cost-savings from using generative AI, not just for not hiring models but also to replace sets and backgrounds. However, companies using generative AI to quickly signal diversity, as Levi’s did when it said it wanted to show diverse models “in a sustainable way,” can end up excluding the very people they intended to include.
A Levi’s spokesperson pointed to a March statement addressing these criticisms, and described its upcoming work with generative AI as “a small, controlled pilot” for its website. “We are not scaling back our plans for live photo shoots, the use of live models, or our commitment to working with diverse models,” the spokesperson told Insider.
Renee Miller, chief diversity, equity, and inclusion officer at the ad agency Code and Theory, understands that attraction to save time and money, while seeming to promote diversity.
“It’s now gotten to a place where, particularly for smaller clients, it is tempting,” she said. “‘Oh man, we can have all the models we wanted, but for a smaller price!'”
Apparel company Revolve debuted billboards that used AI-generated models wearing some of its upcoming fashions and was met with little fuss.
“We won’t have to spend on a photoshoot potentially, tens of thousands of dollars building a set,” Michael Mente, Revolve’s co-CEO, previously explained to Insider.
“You want inclusion, but these shortcuts to inclusion often actually create the opposite,” said Code and Theory’s Miller. The key, Miller said, is to ask the right questions early on around the consequences of using a generative AI model.
“Who are you paying, who are you hiring, what ends up happening in the end?” Miller said. “It’s not like AI models are only created by white folks who then are using this digital blackface, digital yellowface, et cetera. There are people of color who have created generative AI models.”
She added: “We don’t want to end up in a position where we have a Shudu.” Shudu Gram was a heavily-criticized digital creation from 2018 who presented as a Black woman but was designed by a white man, and touted brands like SOULSKY and Fenty on Instagram.
In response to these concerns, advertisers are starting to put guardrails in place.
Agencies like Code and Theory, VMLY&R, and Media.Monks, for instance, have positioned internal groups and Slack channels to focus on these issues and provide resources for creative talent who might have questions over the ethical use of AI.
Code and Theory has an AI ethics board, Miller said. And VMLY&R is integrating a framework to make sure the agency uses generative AI in a way that’s ethical and legal, said its chief innovation officer Brian Yamada.
“We’re building that into our processes to have humans in-the-loop and ask the right questions along the design and development cycle,” Yamada said.
Despite the need for caution, some advertisers want to train generative AI engines to be more representative of diverse experiences.
For instance, Alex Coles, a motion designer at Decoded Advertising, helped run a project to see if generative AI could synthesize the nuances of a “Black voice.”
“We found that yes, it can capture the Black voice, however it captures the biases that are seen of the Black voice,” Coles said. “It will come out with an output speaking like Megan Thee Stallion.”
Decoded Advertising staffers see an opportunity to make sure data about diverse peoples informs generative AI.
“Without the data being put in from Black experiences, AI tools have been limited,” said Decoded Advertising junior strategist Cheyanne Moore. “In order for it to be accurate, it must have diversity included.”