Gyani Ray, N Molla
The paper is a critical analysis on why infrastructure, policy based, and organizational culture is important in the successful implementation of fairness-improving artificial intelligence (AI). Although the metrics of algorithmic fairness and bias mitigation have been heavily developed, their application in an institutional context is poorly coordinated. The major void in knowledge is how sociotechnical conditions facilitate or limit the process of long-term adoption of fairness beyond the optimization of the technical side. A systematic literature review based on PRISMA helped to identify 150 records, screen them, and then reduce them to 20 studies that fulfilled the inclusion criteria, which were all related to implementation contexts. These results indicate that strong data governance and lifecycle monitoring and auditing systems are the basis of operational fairness, and that enforceable policy mechanisms and culture rooted in leadership play a significant role in the result. The paper concludes that AI-based fairness encompasses a sociotechnical ecosystem and not a set of technical responses.