Towards Design Justice and Gender Perspectives - An Inclusive Audit of Facial Recognition Systems
This study presents a qualitative audit of commercially available facial recognition tools featuring Automated Gender Recognition (AGR), specifically Amazon Rekognition, Face++, and Google's Vision AI, to investigate their inherent intersectional biases and broader ethical and social implications. While Artificial Intelligence offers efficiency, evaluating AI bias using isolated demographic categories overlooks intersectional discrimination, allowing deeper systemic inequalities to persist, disproportionately harming marginalized groups such as Black, transgender, and non-binary individuals. To counteract these biases, we evaluate the tools across four critical dimensions: legal and corporate responsibility, ethical considerations, social implications, and Justice, Equity, Diversity, and Inclusion (JEDI). Our findings reveal a mixed landscape. Google demonstrated a proactive shift toward algorithmic justice by removing gender labels from its Vision AI, acknowledging the problematic nature of inferring gender from appearance. Conversely, Amazon Rekognition and Face++ exhibit a concerning lack of transparency regarding their training datasets and misalignment with inclusive best practices, particularly concerning non-binary gender views and the necessity of gender identification. Ultimately, the findings reinforce that achieving algorithmic justice requires moving beyond mere inclusion to ensure user autonomy, privacy, and systemic change. Relying solely on technical accuracy is insufficient; early integration of frameworks like Design Justice and IEEE Ethically Aligned Design is essential. Furthermore, policy must evolve to prevent AI from being constrained by binary legal definitions. Establishing ethical, human-centric technology demands strong interdisciplinary collaboration to protect vulnerable populations and ensure algorithms genuinely respect the diversity of the communities they serve.