Investigating Practices and Opportunities for Cross-functional Collaboration around AI Fairness in Industry Practice

Fairness work requires significant "bridging labor" across teams: Practitioners from different roles (engineers, product managers, ethicists, etc.) struggle with different understandings of what AI fairness means and how to evaluate it. The study found that employees often take on invisible, unpaid work to translate fairness concepts across these disciplinary divides, which represents a major hidden cost in organizations.

Companies are creatively repurposing existing processes for fairness: When organizations lack dedicated resources and incentives for fairness work, practitioners strategically attach fairness considerations to existing requirements like privacy assessments or use established quantitative evaluation metrics. However, the researchers note practitioners themselves worry these workarounds may be "fundamentally compromised" and insufficient for genuine fairness.

Organizational structure and incentives are critical barriers: The study of 23 practitioners across 17 companies reveals that lack of resources and formal incentives for fairness work is a major obstacle—not just technical limitations. This suggests that responsible AI requires structural and cultural changes within organizations, not just better tools or guidelines.

The invisible labor problem needs recognition: Much of the work to coordinate fairness efforts across functions goes unrecognized and unrewarded. Making this labor visible and formally valued could help organizations better support employees working on fairness issues and improve overall effectiveness.