The invisible women: uncovering gender bias in Al-generated images of professionals

Abstract

This study explores gender bias in Al-generated images of profes­sionals, focusing on the visual representation of male and female professionals in law, medicine, engineering, and scientific research. Using a sample of 99 images from nine popular text-to-image generators, we conducted a survey of 120 respondents who assessed the perceived gender of the images. Our findings reveal a significant gender bias, with men represented in 76% of the images and women in only 8%. This bias persists across all four professions and varies between different Al image generators. The results highlight the potential of Al to perpetuate and reinforce gender inequalities, suggesting the need for more intersectional and inclusive approaches in Al design and research. It further underscores the necessity of diversifying the design process and redistributing power in decision-making procedures to challenge existing biases in Al. Our study emphasizes the need for further action to address gender bias in Al-generated images and high­lights the importance of adopting a more intersectional and inclu­sive approach in future research, considering factors such as race, class, and ability. This commentary aims to raise awareness of the current issues with Al-text to image generators and encourages the development of more inclusive and equitable Al technologies.