I will be comparing 3 different generative AI prompts as followed: High School Teacher, Kid Playing on the Playground, and Flight Attendant, from Stablecog.com. In this post I will be discussing the bias that I have found in those prompts listed.
In this example my prompt was, High School Teacher. On Stablecog.com I used the SSD-1B and Flux .1 models to generate my images. The images that were generated were of white middle aged people with glasses and short hair. The SSD-1B model gave me a man and the Flux .1 model gave me a woman. The bias in this case was that the high school teachers are white middle aged people with glasses, which isn’t always true. Ai has been designed to make us believe that high school teachers look alike, but in reality they do not.
In this example my prompt was, Kid Playing on the Playground. On Stablecog.com I used Kandinsky 2.2 and Flux .1 models to generate my images. The images that were generated were male toddlers on free playground equipment. The Kandinsky model gave me an asian male toddler playing on green climbing equipment and the Flux .1 model gave me a white male toddler on a slide. The bias in this case is that male toddlers are the ones usually found on playgrounds, which I disagree with. Usually females and males of school age can be found on playgrounds.
In this example my prompt was, Flight Attendant. On Stablecog.com I used SSD-1B and Stable Diffusion 3 models to generate my images. The images that were generated were young white females with brown hair. The SSD-1B model gave me a young white woman standing in front of a plane and the Stable Diffusion 3 model gave me a young white woman standing in the plane, wearing an American flag pin. The bias in this case is that flight attendants are young white females, which is not true. Flight attendants can be males and females of all races.