Artificial intelligence (AI) can only be as good as the people who program it. As a result, she has faced allegations of inaccuracy and racial prejudice in the past. The bias of their programmers, who are often young, white men from Silicon Valley, seems pre-programmed. For example, she perceives black people less well in photographs because she was only trained with white faces, or she draws the logical conclusion from the lack of diversity in the workplace that people with other ethnic roots, or women, must be less suitable for this.
However, a female-founded Toronto-based start-up in the fashion and beauty industries is turning to AI to create a fairer future. Halt AI co-founders Amanda Cosco and Roya Sedigh, who have extensive experience in technology and fashion, had one thing in common: they regularly tested algorithms for bias. And so Halt AI became their joint project. FashionUnited spoke to Cosco about Halt AI and their Diversity Analytics Dashboard, which provides in-depth analytics on the diversity of over 500 fashion brands, media and retail companies.
“Most of the stories you hear about AI are cyborg doom and kind of scary, especially when it comes to facial recognition,” says Cosco. “But if we can put AI to good use to improve diversity, then that’s a countermove to that murky narrative.”
They found that most brands measure their diversity from the perspective of new hires, but neglect their marketing and content. So they decided to use their technology to give brands a snapshot of what their social media and marketing looks like from the outside. This aroused the interest of the brands. “Brands asked for contextual information, like how their efforts compare to other brands,” explains Cosco. “If they get a 7 out of 10 on ethnicity, is that really good?”
This motivated the creation of benchmarks for companies to evaluate their performance not only in comparison to other companies but also within the industry in general. Six dimensions of diversity – race, age, gender, body type, skin color and disability – are measured for each brand across its visual content, among other criteria.
Halt AI’s current client base includes Fortune 500 companies following pre-funding and a series of Canadian government grants and angel investments. The founders started with the fashion and beauty industry because both are industries that are particularly important for women. Cosco says, “The technology is ready to scale into other industries, it’s just a matter of building the brand’s reputation now.”
A handful of content review services already exist, such as the Geena Davis Institute, which tracks gender representation in media and entertainment. But Cosco says they’re expensive, don’t scale, and fraught with human error. “What makes Halt AI unique is that we have an accuracy rate of over 90 percent and that we can be fast, handling entire industries in a matter of weeks.”
Luxury fashion still has too little diversity
Halt AI’s data quickly confirmed that the luxury fashion industry, for example, is still very white and very young, with little representation in terms of disability, and that while there is diversity in terms of ethnicity, skin color still tends to remain light. “After BLM and the murder of George Floyd, companies made a lot of promises and the [POC-]Community has received a lot of attention, but the community of people with disabilities is next in the spotlight,” says Cosco. “Sports brands like Nike do very well in terms of diversity, while the usual suspects like Chanel don’t have much diversity.”
But halt AI isn’t about pointing fingers at anyone. The quantitative analyzes they perform provide companies with raw data that they can use to make marketing decisions. Cosco believes there is currently a lack of insight, leaving brands with a big blind spot.
Because of the NDAs, Cosco is limited in what it can say about specific brands, but there are some general patterns it has observed. “Brands are usually surprised by their lack of representation, especially in the space of people with disabilities. We try to create analytics for them so they can track improvements over time to understand better. But helping them understand where they stand within the industry, telling them they’re in the 10th or 90th percentile has proven useful.”
A pilot program for a sub-brand of a major cosmetics company proved so successful that it is now being rolled out internally for all brands so that all marketing teams can reference it in their decision making. While the technology is still a B2B tool for now, Cosco believes it will evolve into a B2B/B2C tool as consumers reap the benefits.
“We want to become the global standard for measuring diversity in every industry so that we can offer a seal of approval for diversity,” says Cosco. “There is no such thing at the moment.”
This article was previously published on FashionUnited.uk. Translation and editing: Barbara Russ