Fairness and Diversity in Information Access Systems

Diversity and fairness are distinct but interconnected concepts: The paper clarifies that while diversity and fairness are often used interchangeably in AI discussions, they represent different values. However, they cannot be treated as completely separate—fostering diversity in information systems can effectively lead to fairer outcomes.

Diversity alone does not guarantee fairness: Simply increasing diversity in an AI system does not automatically result in fair treatment. Organizations must intentionally design systems with fairness as an explicit goal, rather than assuming that diversity measures will automatically produce equitable outcomes.

Information access systems require special attention: Recommender systems and search engines can have "disparate impact" on different groups of users. The paper emphasizes that these widely-used systems need specific methods to mitigate unfair outcomes, suggesting that general fairness principles may not be sufficient without targeted interventions.

Fairness is a core requirement for trustworthy AI: According to the European Commission's High-Level Expert Group on AI, achieving trustworthy AI requires addressing diversity, non-discrimination, and fairness throughout an AI system's entire lifecycle—not just at deployment, indicating this is a foundational concern rather than an afterthought.