Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fairness under Competition

Ronen Gradwohl, Eilam Shapira, Moshe Tennenholtz

TL;DR

This work addresses whether fairness guarantees at the level of individual classifiers extend to fairness in ecosystems with competing firms. By formalizing Equal Opportunity under Competition ($EOC$) and a welfare-based $v$-$EOC$, the authors show that even when each lender uses an EO classifier, the aggregate outcome across lenders can be unfair due to correlations and overlapping borrower pools. They derive worst-case bounds and generalize to multiple classifiers and to welfare-based notions, demonstrating that fairness adjustments at the individual level can paradoxically worsen ecosystem fairness. The theoretical results are complemented by Lending Club data experiments, which reveal that post-processing fairness constraints often increase ecosystem unfairness, though the effect diminishes with larger data. The study highlights the need to consider competition dynamics when designing fair ML systems and motivates regulatory and methodological exploration for ecosystem-level fairness.

Abstract

Algorithmic fairness has emerged as a central issue in ML, and it has become standard practice to adjust ML algorithms so that they will satisfy fairness requirements such as Equal Opportunity. In this paper we consider the effects of adopting such fair classifiers on the overall level of ecosystem fairness. Specifically, we introduce the study of fairness with competing firms, and demonstrate the failure of fair classifiers in yielding fair ecosystems. Our results quantify the loss of fairness in systems, under a variety of conditions, based on classifiers' correlation and the level of their data overlap. We show that even if competing classifiers are individually fair, the ecosystem's outcome may be unfair; and that adjusting biased algorithms to improve their individual fairness may lead to an overall decline in ecosystem fairness. In addition to these theoretical results, we also provide supporting experimental evidence. Together, our model and results provide a novel and essential call for action.

Fairness under Competition

TL;DR

This work addresses whether fairness guarantees at the level of individual classifiers extend to fairness in ecosystems with competing firms. By formalizing Equal Opportunity under Competition () and a welfare-based -, the authors show that even when each lender uses an EO classifier, the aggregate outcome across lenders can be unfair due to correlations and overlapping borrower pools. They derive worst-case bounds and generalize to multiple classifiers and to welfare-based notions, demonstrating that fairness adjustments at the individual level can paradoxically worsen ecosystem fairness. The theoretical results are complemented by Lending Club data experiments, which reveal that post-processing fairness constraints often increase ecosystem unfairness, though the effect diminishes with larger data. The study highlights the need to consider competition dynamics when designing fair ML systems and motivates regulatory and methodological exploration for ecosystem-level fairness.

Abstract

Algorithmic fairness has emerged as a central issue in ML, and it has become standard practice to adjust ML algorithms so that they will satisfy fairness requirements such as Equal Opportunity. In this paper we consider the effects of adopting such fair classifiers on the overall level of ecosystem fairness. Specifically, we introduce the study of fairness with competing firms, and demonstrate the failure of fair classifiers in yielding fair ecosystems. Our results quantify the loss of fairness in systems, under a variety of conditions, based on classifiers' correlation and the level of their data overlap. We show that even if competing classifiers are individually fair, the ecosystem's outcome may be unfair; and that adjusting biased algorithms to improve their individual fairness may lead to an overall decline in ecosystem fairness. In addition to these theoretical results, we also provide supporting experimental evidence. Together, our model and results provide a novel and essential call for action.

Paper Structure

This paper contains 28 sections, 14 theorems, 32 equations, 12 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Related literature
  4. Model
  5. Fairness with a single lender
  6. Fairness with multiple lenders
  7. Equal Opportunity under Competition
  8. Correlations between classifiers
  9. Example \ref{['ex:third-party']} continued
  10. Generalization: $v$-EOC
  11. Example \ref{['ex:third-party']} continued
  12. Example \ref{['ex:third-party']} continued
  13. Different sets of borrowers
  14. Harmful fairness adjustment
  15. Experiments
  16. ...and 13 more sections

Key Result

Proposition 1

For two EO classifiers $c_1$ and $c_2$ with false-negative rates $\beta_1$ and $\beta_2$, the level of EOC is $\sigma_1\cdot\sigma_2\cdot\left|{\rho^0-\rho^1}\right|$. In the worst case, the level of EOC is $\min\{\beta_1, \beta_2\}-\max\{0, \beta_1+\beta_2-1\}$.

Theorems & Definitions (20)

  • Example 1
  • Example 2
  • Proposition 1
  • Corollary 1
  • Proposition 2
  • Corollary 2
  • Proposition 3
  • Definition 1
  • Proposition 4
  • Corollary 3
  • ...and 10 more