Analysing and Organising Human Communications for AI Fairness-Related Decisions: Use Cases from the Public Sector

Communication gaps between developers and non-technical stakeholders create fairness risks: The paper finds that policy-makers, civil servants, and domain experts are significantly less involved than developers throughout an AI system's life-cycle in public sector applications. This forces developers to take on advisory roles they may not be qualified for, while domain experts who understand the policy context are sidelined from critical fairness decisions.

Non-technical decision-makers lack skills to evaluate AI limitations: End-users and policy-makers often cannot independently interpret a system's technical limitations and must rely entirely on developers for guidance. This dependency creates a power imbalance that can lead to misuse of algorithms, particularly in high-stakes decisions like social benefit allocation or fraud prediction.

Organizational structure matters for responsible AI deployment: The research reveals that fairness-related decisions in public sector AI depend not just on the algorithm itself, but on how communication flows between different roles and expertise areas. Missing stakeholders or unclear role divisions directly undermine the ability to make fair decisions.

Public sector AI requires intentional stakeholder coordination: Rather than treating AI fairness as purely a technical problem, the paper suggests that organizations need to deliberately structure communication processes that bring together developers, domain experts, policy-makers, and civil servants at appropriate phases of the system's life-cycle.