Black History Month: Generative AI and diversity

Generative AI has seamlessly integrated into our daily lives, becoming a valuable tool for work, education, and beyond. However, when utilizing such tools and the resulting outputs, it is becoming increasingly important to take diversity and representation into consideration. As humans develop machines and automated services, it's vital to recognize that these creations, like us, may harbor racial biases inherited from their creators.

A recent exploration of AI generators by founder Karin Ginena of RAI Audit, a consultancy focusing on AI governance and research, found some very important observations about AI generated images of the workplace. When prompted for images using terms such as "professional headshot or "professional dress code", AI programs consistently delivered pictures of white individuals as an output. This reveals a significant underrepresentation of people of color and highlights the bias of AI image generation towards white stereotypes; stereotypes that can be harmful to advancing diverse perspectives of what is considered "professional". Furthering the lack of diversity in generative AI, gender bias was also apparent when searching terms related to "professionalism": generated images that would mostly contain younger white males.

Training data plays a critical role in generative AI diversity. Tools that create new content based on existing patterns and have biased or incomplete training data can lead to skewed and incomplete AI models. A study conducted by Joy Buolamwini at MIT, revealed a lack of diversity in facial recognition systems' accuracy, emphasizing the need for representative training data to avoid negative consequences. "For instance, ...researchers at a major U.S. technology company claimed an accuracy rate of more than 97 percent for a face-recognition system they’d designed. But the data set used to assess its performance was more than 77 percent male and more than 83 percent white." A lack of diversity in data sets can be harmful in creating accurate representations within generative AI technologies.

While there are many concerns about AI's impact on everyday life, experts emphasize that improving bias should be a top priority. AI is currently in a critical development stage where addressing issues in diversity bias is of upmost importance; rectifying these issues in the future will bring immense challenges. With these challenges of racial bias and inaccuracy in mind, it is imperative to foster diversity not only in training data but also within the teams developing generative AI. A diverse team can be able to better identify potential biases and ethical concerns, contributing to more innovative and creative solutions. Research affirms that diverse teams excel in problem-solving and decision-making, bringing a broader range of perspectives to the table (Johansson, 2023). 

To create ethical and inclusive technologies that truly reflect the needs and experiences of all individuals and communities, it is important to prioritize diversity and inclusion in generative AI development. In our efforts to embrace diversity both in our personal lives and professional endeavors, prioritizing education remains paramount. As we move forward, recognizing the impacts of diversity gaps is crucial for fostering an environment that embraces and celebrates the myriad of perspectives from diverse backgrounds.

Interested in learning more? Check out the following articles on generative AI and diversity:

How can AI support diversity, equity and inclusion?

Humans are biased. Generative AI is even worse

Diversity in AI is a problem – Why Fixing it will help everyone