Strategic Costs of Perceived Bias in Fair Selection

A team of researchers including L. Elisa Celis and Nisheeth K. Vishnoi has published a groundbreaking paper, 'Strategic Costs of Perceived Bias in Fair Selection,' accepted at NeurIPS 2025. Using a game-theoretic model, they analyze meritocratic systems like college admissions and corporate hiring. The core finding is a 'perception-driven bias': when candidates from different socioeconomic groups perceive different post-selection values (e.g., future salary or career success), they rationally invest different levels of effort. This effort directly translates to observable merit, meaning disparities emerge even when selection is perfectly fair and based solely on that merit.

The paper specifically highlights the role of modern 'techno-social environments,' including AI-powered tools that provide personalized career or salary guidance. These tools can inadvertently shape and solidify group-based perceptions of value. The researchers characterize the unique Nash equilibrium in their model and provide explicit formulas showing how valuation disparities and institutional selectivity jointly determine outcomes like effort, representation, and social welfare. They also propose a cost-sensitive optimization framework, offering institutions a way to quantify how modifying selectivity or working to change perceived values can reduce disparities without sacrificing their core goals. This work bridges rational-choice and structural explanations of inequality, showing how individual incentives are shaped by a broader feedback cycle linking technology, social context, and perceived rewards.