A Black-Box Technique to Mitigate Gender Bias for Generative AI

Through the generation of visuals that are both creative and realistic, generative artificial intelligence makes a significant contribution to the education of today. On the other hand, prejudices that are inherent in the training data can cause these models to perpetuate gender stereotypes, particularly when it comes to the creation of images associated with occupations that do not include explicit gender advice. These kinds of prejudices not only impede diversity but also run the risk of strengthening gender stereotypes that already exist. This work offers an automated black-box strategy that tries to reduce gender bias by employing face detection and gender classification on images generated by generative artificial intelligence. The strategy is effective because it addresses the problem at the input and output levels by altering the prompts used with the AI models. This is because the method is flexible and may be used with a broad variety of generative artificial intelligence models.