Towards a Requirements Engineering Framework for Equitable AI Systems: A Latin American Perspective

AI systems continue to perpetuate demographic bias, and Requirements Engineering (RE) still lacks systematic practices for specifying equity from a system’s inception – a gap made urgent in Latin America by emerging regulation (Brazil’s LGPD, proposed AI Act) and documented biometric disparities. We propose an RE framework, combining Design Science Research with Delphi expert validation, around six components: stakeholder identification, equity requirement specification, fairness metrics, traceability, documentation, and continuous monitoring. To ground this proposal in measurable evidence rather than citation alone, we ran a reproducible case study on the public UCI Adult Income benchmark: excluding race and sex from model features – a common ad hoc fairness practice – still leaves a false-negative-rate (FNR) gap of 18 percentage points across race (51% for Black vs. 33% for Asian-Pac-Islander applicants, $n \ge 392$) and 9 points across sex. This is a proof of concept motivating the framework, not a validation of the full framework or a biometric replication.