After AI failed to generate an image of a Black surfer, David Mesfin set out to change how technology sees people of color in the outdoors.
(Photo: Kory Lamberts)
Published August 19, 2025 03:42AM
David Mesfin was coming up empty. The prompt seemed simple enough: “Black man with a surfboard.” Over and over, he typed it into the search bar, and over and over, the same glaring blind spot came up on his computer screen. Was this the best AI could do? Something felt off.
“When I searched…the program would end up giving me a white man with dark skin holding a board,” recalls Mesfin. “Black people, Black surfers, just weren’t part of the data set.”
The filmmaker and agency creative director had just spent the last year filming and producing “Wade in the Water,” a documentary about the primarily overlooked 1,000-year history of Black surfing. As a Black creative producing a film about Black representation in water sports, the moment hit with acute irony. Even after producing a film that proved Black people were participating in water sports, leading-edge technology had no evidence of their existence. And the online imagery gap didn’t stop at surfers. The more Mesfin looked, the more he found the lack of Black—and all BIPOC—faces extended to all outdoors imagery, then to photos in workplaces, schools, and homes.
However, instead of being discouraged by the lack of visual references in AI and the rapidly evolving world of information and technology, Mesfin saw an opportunity to rewire the machine. Enlisting the help of Pocstock, a BIPOC imagery database, and dozens of other creative agencies, Mesfin and his team at Innocean have begun Refacing the Future. This project not only re-trains popular AI models for industry leaders like OpenAI, Google, and Microsoft to support imagery of people of color, but also creates the first AI guidebook on how to continue diversifying those datasets for years to come.
“We need a broader group of people that can actually look at this technology and what is being developed,” explains Mesfin. “If you’re going to create a tool for the world, it should represent the world accurately, right?”
The outdoors is no stranger to the pitfalls of misrepresentation, just ask the generations of BIPOC skiers, surfers, hikers, and climbers that watched white athletes dominate the videos, ad spreads, and magazine pages of their youth. Mesfin grew up consuming that same media, watching athletes who didn’t look like him in between paddle outs around St. Augustine, Florida. Eventually, though, he did find other surfers of color, realizing that much of what he saw on the silver screen simply reflected the perspectives of its (white) creators.
In many ways, he feels the same about AI.
“Speaking to under-representation, I think it’s simply because of the individuals that are behind creating the content or technology,” explains Mesfin.
Throughout the short history of generative AI, those individuals have typically been both white and male.
Additionally, these language learning models draw from already published material to populate their program, material that continues to perpetuate biases and stereotypes across generations.
“Media has always painted a negative image of BIPOC communities…we’ve never been represented in our true sense,” says Mesfin. “[AI] is picking up on those things, those stereotypes—it’s reflecting what’s already out there.”
When the algorithm fails to recognize antiquated depictions of people of color, it often misidentifies them altogether. A federal study in 2019 concluded that Asian and Black people were up to 100 times more likely to be incorrectly identified or depicted by AI models in comparison to white men. While advancements have been made in the technology over the last few years, it’s still easy to find holes in the ship. Innocean points to the Asian dentists that make up 22 percent of the entire U.S. dentist population. When AI is asked to generate an image of a dentist, an Asian person only appears 2.3 percent of the time.
Refacing the Future is working to eliminate these digital information shortcomings by strategically flooding the system with information. Mesfin says an alliance with Pocstock, an image database featuring over 1.8 million images of people of color, has provided the initiative a platform to stand on, but considers the way Pocstock organizes these images to be the key to a more equitable online future. Each image in Pocstock’s database is manually tagged for skin tone, gender, race, age, and a range of other cultural data, which makes it easier for AI models to identify and incorporate these details into auto-generating image processes. With Refacing the Future, the alliance aims to collect more of these photos and intentionally attach these kinds of physical identifiers to each one, making the information more readily available for popular image-generating technologies and helping to train the models to see the world in all its color.
From there, Innocean has identified 22 other creative agencies to produce BIPOC-specific media, not only to host imagery and video on the Pocstock platform but also to tag and make that work available to OpenAI, Canva, Google, and other diversity data-starved AI models. The result? According to Refacing the Future’s website, over 16 BIPOC photographers are working to produce 96,000 pieces of original media within a year. Hispanic skateboarders, Native American watermen and waterwomen, Asian snowboarders—a list of historical blind spots shrinking by the day.
As the database continues to grow, Refacing the Future has also prioritized establishing what that growth will look like. For one, the guidebook encourages creative agencies to curate photoshoots with diverse talent (including those behind the lens). Additionally, the guidebook provides creative resources and cultural curators that can help media producers steer a more inclusive course.
Currently, AI is a powerful tool without a clear code of conduct, something Mesfin and others see as an opportunity to establish some good practice ground rules. That’s why Refacing the Future published the first BIPOC AI guidebook, a comprehensive tool for agencies and creatives to explore who to shoot, how to shoot, how to prepare files, and how to tag, all to eliminate bias.
“I want to hand agencies this guidebook and say, ‘Hey, here’s how we did it,’” explains Mesfin. “Here’s how you can find photographers, how you should approach your photoshoots. That way it can be more a reflection of the community and authentically capture the demographic.”
Still in its early stages, Refacing the Future has identified several areas for improvement, including the representation of women in STEM fields and the changing roles of women in the household. However, when Mesfin was tapped to produce his creative shoot for the newly established platform, he returned to what he knew best: the ocean.
On a sunny day last February, Mesfin and a team of photographers and surfers of color descended on Huntington Pier in Southern California. The ocean spray cast a morning chill, but the air was warm as the crew walked toward the iconic wooden pylons. It all felt like a full circle, as Mesfin was now charged with filling a gap he had identified during his fruitless internet search just a few years prior.
One of the photographers, Kory Lamberts, had met Mesfin at Great Day in the Stoke, the largest gathering of Black surfers in the world, and felt inspired by the creative director’s vision. When the call came to join up for this first Refacing the Future project, Lamberts knew he needed to be a part of it.
“Here was this chance to reshape and recreate the future from the lens of people that haven’t been able to tell it yet,” he explains. “It’s time we figure out a way to move forward with these technologies to utilize them for our communities.”
With four athletes in the water and photographers both on land and ducking waves, the team stacked shots and videos that would later make their way to Pocstock and further into today’s popular AI modeling programs, letting a new digital world know exactly who they are.
