June 2024 | MARC CHEONG, EHSAN ABEDIN, MARINUS FERREIRA, RITSAA RT REIMANN, SHALOM CHALSON, PAMELA ROBINSON, JOANNE BYRNE, LEAH RUPPANNER, MARK ALFANO, COLIN KLEIN
This article systematically investigates the gender and racial biases present in DALL-E Mini, a popular image generation model. The authors find that DALL-E Mini tends to represent occupations as either predominantly male or female, with a significant overrepresentation of White individuals and a lack of diversity in other racial groups. These biases are reflected in the images generated by the model, which often depict occupations as being populated by either all men or all women, and predominantly by White people. The study uses a rigorous methodology involving coding images based on gender and race, comparing the results to real-world labor statistics from the US Bureau of Labor Statistics. The findings highlight the need for critical scrutiny and potential regulation of generative AI technologies to prevent the perpetuation and amplification of existing social biases. The authors also discuss the implications of these biases for the field of generative AI, emphasizing the urgent need for ethical evaluations and regulatory oversight to address the potential harms caused by such technologies.This article systematically investigates the gender and racial biases present in DALL-E Mini, a popular image generation model. The authors find that DALL-E Mini tends to represent occupations as either predominantly male or female, with a significant overrepresentation of White individuals and a lack of diversity in other racial groups. These biases are reflected in the images generated by the model, which often depict occupations as being populated by either all men or all women, and predominantly by White people. The study uses a rigorous methodology involving coding images based on gender and race, comparing the results to real-world labor statistics from the US Bureau of Labor Statistics. The findings highlight the need for critical scrutiny and potential regulation of generative AI technologies to prevent the perpetuation and amplification of existing social biases. The authors also discuss the implications of these biases for the field of generative AI, emphasizing the urgent need for ethical evaluations and regulatory oversight to address the potential harms caused by such technologies.