DIVERSIFYING REPRESENTATION IN THE GLOBAL SOUTH

Creative Artificial Intelligence (AI) are trained on datasets that often represent WEIRD (Western Educated Industrialised Rich and Democratic) societies and contexts. Data used to train creative AI tools dominantly draw from Anglo-Saxon, middle class, heterosexual, and white demographics. These datasets lack diversity, especially in the representation of the Global South where 90 percent of the world’s youth currently live – and have fast become the next billion user markets. Biased data perpetuates and amplifies existing societal biases, which could lead to inappropriate, and unusable designs using generative media.

A major public data source for training creative AI is the Creative Commons (CC). The Creative Commons, while serving as a valuable platform for open content sharing and building of generative media tools, currently lacks diversity and representation. The current landscape of CC-licensed content often reflects the broader disparities present in the creative industry, with certain voices and perspectives dominating while others remain underrepresented.

This ‘Debiasing Creative Data’ project contributes to broadening perceptions of equity around the world through the lens of content, starting with India. This project strives to capture critical process measures prior to image/content acquisition from diverse underrepresented communities. DCD identify gaps in frameworks/approaches and serves as the first step towards bringing an international perspective to the content acquisition discussions.

Our Lab has pioneered a  7-step approach to debiasing data that is bottom up and led by ‘fellows’. This is an innovative and unique approach that bridges the humanities and technology for more targeted and relevant results. Contact us to learn more or to contract us to deploy such measures at your organization.