Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Laypeople's Attitudes Towards Fair, Affirmative, and Discriminatory Decision-Making Algorithms

Gabriel Lima, Nina Grgić-Hlača, Markus Langer, Yixin Zou

TL;DR

This paper investigates laypeople's attitudes toward affirmative algorithms that explicitly prioritize historically marginalized groups in high-stakes decisions. Using two vignette-based experiments in hiring and bail decisions (N=1193), it contrasts affirmative, discriminatory, and fair algorithmic approaches. Results show broad public support for fairness and skepticism toward discrimination, while opinions on affirmative algorithms vary by political ideology and racial identity, with liberals and racial minorities more favorable than conservatives and dominant-group members. The lack of effect from contextual injustice framing suggests that changing beliefs about marginalization may be challenging; the authors discuss framing and design strategies to bridge divides toward affirmative algorithmic futures.

Abstract

Affirmative algorithms have emerged as a potential answer to algorithmic discrimination, seeking to redress past harms and rectify the source of historical injustices. We present the results of two experiments ($N$$=$$1193$) capturing laypeople's perceptions of affirmative algorithms -- those which explicitly prioritize the historically marginalized -- in hiring and criminal justice. We contrast these opinions about affirmative algorithms with folk attitudes towards algorithms that prioritize the privileged (i.e., discriminatory) and systems that make decisions independently of demographic groups (i.e., fair). We find that people -- regardless of their political leaning and identity -- view fair algorithms favorably and denounce discriminatory systems. In contrast, we identify disagreements concerning affirmative algorithms: liberals and racial minorities rate affirmative systems as positively as their fair counterparts, whereas conservatives and those from the dominant racial group evaluate affirmative algorithms as negatively as discriminatory systems. We identify a source of these divisions: people have varying beliefs about who (if anyone) is marginalized, shaping their views of affirmative algorithms. We discuss the possibility of bridging these disagreements to bring people together towards affirmative algorithms.

Laypeople's Attitudes Towards Fair, Affirmative, and Discriminatory Decision-Making Algorithms

TL;DR

This paper investigates laypeople's attitudes toward affirmative algorithms that explicitly prioritize historically marginalized groups in high-stakes decisions. Using two vignette-based experiments in hiring and bail decisions (N=1193), it contrasts affirmative, discriminatory, and fair algorithmic approaches. Results show broad public support for fairness and skepticism toward discrimination, while opinions on affirmative algorithms vary by political ideology and racial identity, with liberals and racial minorities more favorable than conservatives and dominant-group members. The lack of effect from contextual injustice framing suggests that changing beliefs about marginalization may be challenging; the authors discuss framing and design strategies to bridge divides toward affirmative algorithmic futures.

Abstract

Affirmative algorithms have emerged as a potential answer to algorithmic discrimination, seeking to redress past harms and rectify the source of historical injustices. We present the results of two experiments () capturing laypeople's perceptions of affirmative algorithms -- those which explicitly prioritize the historically marginalized -- in hiring and criminal justice. We contrast these opinions about affirmative algorithms with folk attitudes towards algorithms that prioritize the privileged (i.e., discriminatory) and systems that make decisions independently of demographic groups (i.e., fair). We find that people -- regardless of their political leaning and identity -- view fair algorithms favorably and denounce discriminatory systems. In contrast, we identify disagreements concerning affirmative algorithms: liberals and racial minorities rate affirmative systems as positively as their fair counterparts, whereas conservatives and those from the dominant racial group evaluate affirmative algorithms as negatively as discriminatory systems. We identify a source of these divisions: people have varying beliefs about who (if anyone) is marginalized, shaping their views of affirmative algorithms. We discuss the possibility of bridging these disagreements to bring people together towards affirmative algorithms.
Paper Structure (49 sections, 5 figures, 23 tables)

This paper contains 49 sections, 5 figures, 23 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Critiques of Algorithmic Fairness
  4. People's Perceptions of Decision-Making Algorithms
  5. Affirmative Algorithms as Attempts to Rectify Past Harms
  6. Potential Disagreements in Perceptions of Fair and Affirmative Algorithms
  7. Methods
  8. Study Design
  9. Measures
  10. Perceived Fairness of the Decision-Making Domain:
  11. Perceptions of the Algorithm:
  12. Beliefs About Marginalization:
  13. Demographics:
  14. Analysis Plan
  15. Participants
  16. ...and 34 more sections

Figures (5)

  • Figure 1: High-level overview of our methodology. Our vignette and manipulations are presented in Appendix \ref{['supp:methods']}. We focus on participants' perceptions of the algorithm and beliefs about marginalization due to space constraints (see Appendix \ref{['supp:analysis']} for supplementary analysis).
  • Figure 2: Participants' mean judgments of fairness, trust, objectivity, support concerning different types of algorithms deployed to assist hiring and bail decisions. The x-axis refers to the algorithm type experimental manipulations: control (Ctrl), affirmative (Aff), discriminatory (Disc), and fair (Fair). Standard errors are included in the figure but are not visible due to their small values.
  • Figure 3: Participants' judgments of fairness, trust, objectivity, support concerning different types of algorithms depending on participant's political leaning. The x-axis represent political leaning on a 5-point scale, in which -2 refers to conservative (Con), 0 to moderate (Mod), and 2 to liberal (Lib).
  • Figure 4: Participants' mean judgments of fairness, trust, objectivity, support concerning different types of algorithms depending on participant's racial group. The x-axis refers to the algorithm type experimental manipulations: control (Ctrl), affirmative (Aff), discriminatory (Disc), and fair (Fair). Standard errors are included in the figure but are not visible due to their small values.
  • Figure 5: Participants' mean judgments of fairness, trust, objectivity, support concerning different types of algorithms depending on participant's self-reported race. The x-axis refers to the algorithm type experimental manipulations: control (Ctrl), affirmative (Aff), discriminatory (Disc), and fair (Fair).