MullenLowe hopes to correct AI bias with image library of professional women
AI tools including midjourney, DALL·E 2 and Stable Diffusion have been used to create an image database that helps address gender imbalance.
MullenLowe boldly claim to have cracked AI gender bias
MullenLowe MENA claims to have developed a surefire technique to banish gender bias in artificial intelligence (AI). Its dedicated website, FixingthebAIs, lets participants play their part in challenging established gender stereotypes, ensuring that they do not carry over into the AI era.
As part of this research, the agency asked the tools to create images of various professions, including mechanical engineers, F1 drivers, mathematicians, CEOs, boxers, football players and many more. However, the results consistently showed only male representation in these roles. This highlights the gender bias that exists in AI and the need for corrective measures.
In response, the agency has built an image bank of women performing professional roles, populating a new gender-neutral dataset that can train automated systems to recognize and combat existing bias.
Advertisement
Speaking about the campaign, MullenLowe MENA’s CEO, Mounir Harfouche, commented: “Rather than just highlighting the issue, we wanted to do something about it by creating an image bank of women in different professions, we hope to address these issues head-on and create a more equitable world for all. We’re asking colleagues, partners, clients and the wider industry to join us in helping retrain AI to break the AI gender bias.”
Prerna Mehra, head of art and creative director at MullenLowe MENA added: “As a team of all female creatives working on this campaign, we have brought unique insights and experiences to the table, leading to innovative and meaningful work. By making our images royalty and rights-free, we are not only encouraging their widespread use but also contributing to the growth of AI.”
The work dovetails with a UNESCO report that raises concerns over the role of growing automation on women’s participation in the labor market. Seeking to turn AI from a problem to a solution, MullenLowe put existing systems to the test, by asking for images of specific professions such as footballers, mechanical engineers and mathematicians – all consistently shown as male only.