Reproduce “A Gray Literature Study on Fairness Requirements in AI-enabled Software Engineering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Thanh Nguyen, Chaima Boufaied, Ronnie de Souza Santos
Researchers at the University of Calgary conducted a gray literature study examining how fairness requirements are addressed in AI-enabled software engineering. The analysis of industry reports and articles indicated that fairness is often managed reactively, with data bias identified as the leading cause of violations that frequently result in harmful biased predictions for individuals and society.
Key insights
- Gray literature predominantly defines fairness as non-discrimination, context-dependent, and relative to sensitive attributes, with less emphasis on formal or technical definitions.
- Fairness management is largely reactive, focusing on model training, bias mitigation, and post-deployment monitoring, rather than proactive integration in early SDLC stages like requirements elicitation.
- Data bias (imbalanced, unrepresentative, or historically biased datasets) is identified as the most frequent cause of fairness violations in AI systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.