By Jayden Mallari, Josue Rodriguez, David Mayancela Suquila

Abstract

AI technologies may be prone to developing biases based on the data they are trained on, which will reflect in the responses they give to certain prompts. As a result, certain denominations of individuals, in prompts where they are to explicitly appear, may be represented too often or too little on the bases of ethnicity, gender, and age. Utilizing Gemini’s Nano Banana image generation AI, 100 images of three different professions, being that of a Senator, Journalist, and Public Defender, were generated, totaling 300 images overall. Within them, it was found that the AI often defaulted to stereotypes created on different bases between each profession. Although age range was often fairly accurate or equally dispersed, there was a tendency to also amplify the appearances of individuals to match current beauty standards, sometimes to an unrealistic degree in relation to the setting of their image generation or profession itself. 

First Draft