TPRC 2024: The 52nd Research Conference on Communications, Information and Inter

Link to paper

Abstract:
This study examines the posited relationship between diversity and the supply of debiased AI, using a cross-sectional survey sample of AI professionals working in Silicon Valley. The results of preliminary analyses show that increased diversity, when indicated by value diversity as opposed to demographic diversity, had a significant effect on decreased AI bias. Further, the analysis found that there was no significant difference in manufactured AI bias or the effort to debias AI attributable to the socio-demographic diversity alone, as self-reported by AI industry insiders. The study concludes with a call for much closer attention 1) to diversity in its conceptual and regulatory operationalization and 2) to conditional institutional variables in translating diversity into discernible effects. The author of this study emphasizes a preliminary nature of the findings, with suggestions for the potential areas of improvement in the future AI debate.