Risk-Informed and AI-Based Bias Detection on Gender, Race, and Income Using Gen-Z Survey Data

Abstract

In the digital world, growing bias in diversity, equity, and inclusion (DEI) is increasingly influencing people's thoughts and behaviors. Though younger generations have a large presence on social media, there is minimal DEI research focusing on Generation Z (Gen-Z). This paper examines the DEI landscape through an online survey targeting the Gen-Z population. We analyzed 675 survey responses to train an AI-based model to assess both the existence and risk of gender, race, and income bias, attempting to foster a healthier online community. We built a Natural Language Processing-based risk assessment model using Word2Vec with a dimensionality reduction algorithm. We then trained the model with Gen-Z's vocabulary using the results from our survey. Our findings show that words traditionally associated with a particular gender are now being used to describe both genders (e.g., strong), and some gender-specific words are fading away (e.g., doll). This suggests that society is moving away from certain gender stereotypes. In the responses to the survey's race-related questions, all racial groups mentioned racial justice, reflecting Gen-Z's awareness of the racial issues that exist in today's society. Also, our K-means clustering algorithm identified social justice issues as a common theme across the gender, race, and income areas studied in our survey, highlighting Gen-Z's understanding of injustice in these areas of society.