Agencies under Innocean USA unite to advocate for multicultural AI training data.
Rewritten Article:
Battling AI Bias: Image Platforms Correct Their Approach to BIPOC Representation
In the realm of creative advertising, accurate representation matters immensely - a fact underscored by David Mesfin, INNOCEAN USA's vp creative director, during the production of his documentary, Wade in the Water: A Journey into Black Surfing and Aquatic Culture. His concern? AI-powered image search platforms consistently delivered inaccurate results when searching for images of Black surfers.
Mesfin explained, "Every query, every prompt returned a white surfer with a dark skin tone. The data simply wasn't there, and representation isn't just about inclusion. It's about accuracy."
As a result, INNOCEAN USA joined Breaking Bias, an initiative spearheaded by diverse stock image database POC Stock, intent on eradicating AI bias through the creation of datasets more accurately representing BIPOC communities.
Steve Jun, INNOCEAN USA’s CEO, weighed in, saying, "AI is as good as its training dataset. When we learned that a simple action could address a complex issue, we knew supporting Breaking Bias was the right choice."
The coalition, comprising INNOCEAN USA, FCB Global, David&Goliath, and over 20 other agencies, is leveraging POC Stock images, rich metadata, and multiple rounds of human curation to retrain AI models, prioritizing authenticity and avoiding stereotypes.
David Angelo, founder and creative chairman of David&Goliath, voiced concern over AI's handling of the BIPOC community’s representation, stating, "AI representation still lacks authenticity and often perpetuates stereotypes. But if we collaborate, AI can self-correct."
The problem of AI bias has long been a source of concern for the agency world. In 2023, Hispanic agencies like Lerma and Alma sounded the alarm about stereotyping in AI tools such as Midjourney. By 2024, The Brandtech Group launched Bias Breaker, a technology aimed at correcting identity factors, including race, gender, and age, in AI-generated content. That same year, Mastercard debuted Mastercard Small Business AI, a generative AI chatbot catering to the needs and perspectives of multicultural and minority business owners.
Looking ahead, the Breaking Bias coalition intends to expand its dataset, focusing on underrepresented groups, and extend its reach beyond the U.S. In Mesfin's words, "AI is used globally, and it needs to represent that diversity."
As the Breaking Bias initiative evolves, the coalition seeks involvement from beyond the current group of agencies, agencies, and other industries. Mesfin expressed hope for additional partnerships, stating, "To make a real impact, we'll require more than just a few agencies, and ideally, the likes of Midjourney and DALL·E would join us in our quest."
Ultimately, Jun added, "We need more agencies to tap into their passions, join our cause, and help this evolving tool mirror the real world."
This growing movement toward inclusive AI represents a significant step forward in combating stereotypes and misrepresentation, paving the way for a more accurate and authentic digital landscape.
Enrichment Data:
- The Breaking Bias initiative, organized by Innocean USA and 22 other creative agencies, aims to tackle AI bias by creating diverse AI training data.
- Agencies involved use POC Stock images along with detailed metadata to help retrain AI models, focusing on annotating race, ethnicity, gender, and cultural fluency to ensure authentic representation.
- Bias Breaker technology, developed by The Brandtech Group, corrects identity factors such as race, gender, and age in AI-generated content, helping reduce stereotypes and misrepresentations in AI outputs.
- Mastercard Small Business AI is a generative AI chatbot designed to address the needs and perspectives of multicultural and minority business owners.
- The AllyAI certification, initiated by LGBT Tech, is geared towards ensuring that AI systems are free from bias against LGBTQ+ individuals. While not exclusively focused on BIPOC, its framework can be adapted to broader diversity efforts.
These initiatives highlight the growing acknowledgement of the need for more inclusive and representative data in AI systems to combat bias.
- INNOCEAN USA, in collaboration with 22 other creative agencies, has launched the Breaking Bias initiative, focusing on creating diverse AI training data to combat AI bias.
- Utilizing POC Stock images and detailed metadata, the coalition aims to authentically represent BIPOC communities, addressing stereotypes and misrepresentations in AI-powered image search results.
- The Brandtech Group's Bias Breaker technology targets identity factors like race, gender, and age, correcting biases within AI-generated content and helping create a more accurate digital landscape.
- Mastercard Small Business AI, a generative AI chatbot, caters to the specific needs and perspectives of multicultural and minority business owners.
- Although not exclusively focused on BIPOC, the AllyAI certification, initiated by LGBT Tech, ensures that AI systems are free from bias against LGBTQ+ individuals, providing a framework that can be adapted to broader diversity efforts.
