Estimating divergent moral and diversity preferences between AI builders and AI users

Abstract

AI builders' preferences influence AI technologies throughout the development cycle, yet the demographic homogeneity of the AI workforce raises concerns about potential misalignments with the more diverse population of AI users. This study examines whether demographic disparities among AI builders and AI users lead to systematic differences in two critical domains: personal moral beliefs and preferences for diversity-related machine outputs. Using a pseudo-experimental, cross-sectional design, we assessed the moral beliefs and diversity preferences of adults (N = 519, 20+ years) and adolescents (N = 395, 15–19 years) with varying levels of actual or projected AI engagement. In our sample, males and adults with higher AI engagement exhibited stronger endorsement of instrumental harm and weaker support for diversity. Given the largely male composition of the AI workforce, these findings suggest there may be critical value gaps between current builders and users. In contrast, our adolescent data indicated that—developmental changes withstanding—these differences may narrow in future cohorts, particularly with greater gender balance. Our results provide initial support for a broader concern: that demographic homogeneity in the AI workforce may contribute to belief and expectation gaps between AI builders and users, underscoring the critical need for a diverse AI workforce to ensure alignment with societal values.