Team Diversity Promotes Software Fairness: An Experiment on Fairness-Aware Requirements Prioritization

A new study by researchers including Cleyton Magalhães and Ronnie de Souza Santos provides empirical evidence that team diversity directly improves software fairness. The paper, 'Team Diversity Promotes Software Fairness: An Experiment on Fairness-Aware Requirements Prioritization,' conducted a controlled experiment with 27 pairs of software engineering students. The teams, comprising 13 LGBTQ+ diverse pairs and 14 non-diverse pairs, were tasked with prioritizing user stories containing varying fairness implications, simulating early-stage development decisions for systems that could include AI components.

While both groups showed general alignment with fairness principles, the results were starkly different in execution. The LGBTQ+ diverse pairs were significantly more consistent in rejecting user stories that posed fairness risks and made approximately 50% fewer fairness-related 'misprioritization' errors. Thematic analysis of their decision-making revealed that diverse teams grounded their reasoning in concepts of inclusion, non-discrimination, and ethical responsibility. In contrast, non-diverse pairs tended to adopt a more pragmatic, goal-oriented perspective, potentially overlooking subtle fairness pitfalls.

The study concludes that fairness must be a consideration from the very first stages of the software development lifecycle, not just during algorithm design or data auditing. It argues that diverse teams enhance the collective ability to identify and correctly interpret fairness issues during requirements analysis, leading to more reflective and inclusive decision-making. This research shifts the conversation on building ethical AI, highlighting that who builds the software is as important as how it is built.