Practitioner Insights on Fairness Requirements in the AI Development Life Cycle: An Interview Study

Fairness awareness exists but practices are inconsistent: While the 26 practitioners interviewed across 23 countries recognized the importance of fairness in AI/ML systems, the study found that "practices are inconsistent, and fairness is often deprioritized," indicating a significant gap between understanding fairness concerns and actually implementing them in development.

Fairness gets deprioritized against other business pressures: The research identified that fairness considerations are frequently overshadowed by competing priorities such as "meeting critical delivery deadlines" and "addressing all software functionalities," suggesting that time and resource constraints push fairness concerns to the background during AI development.

Significant knowledge gaps persist among practitioners: The study revealed "noticeable knowledge gaps" among AI developers regarding fairness requirements, indicating that many practitioners lack the necessary expertise or training to effectively identify and address fairness issues throughout the software development lifecycle.

Fairness assessment needs to start earlier in development: The research emphasizes the importance of "assessing [fairness] arising early in the SDLC," suggesting that fairness considerations should be integrated from the initial stages of AI project planning rather than being addressed as an afterthought during testing or deployment.

Stakeholder agreement on fairness definitions is lacking: The findings highlight "the need for agreement with relevant stakeholders on well-defined" fairness requirements, indicating that organizations often lack clear, shared definitions of what fairness means for their specific AI applications and contexts.