[Paper Review] Dimensions of Diversity in Human Perceptions of Algorithmic Fairness

Nina Grgić-Hlača, Gabriel Lima+2 more|arXiv (Cornell University)|May 2, 2020

Ethics and Social Impacts of AI6 citations

TL;DR

This study investigates how sociodemographic factors and personal experiences shape human perceptions of algorithmic fairness in bail decision-making. Using a survey of 329 participants, it finds that political ideology and direct experience—particularly attending a bail hearing—significantly influence fairness judgments, especially regarding the use of juvenile criminal history, highlighting the need for diverse representation in algorithmic oversight boards.

ABSTRACT

A growing number of oversight boards and regulatory bodies seek to monitor and govern algorithms that make decisions about people's lives. Prior work has explored how people believe algorithmic decisions should be made, but there is little understanding of how individual factors like sociodemographics or direct experience with a decision-making scenario may affect their ethical views. We take a step toward filling this gap by exploring how people's perceptions of one aspect of procedural algorithmic fairness (the fairness of using particular features in an algorithmic decision) relate to their (i) demographics (age, education, gender, race, political views) and (ii) personal experiences with the algorithmic decision-making scenario. We find that political views and personal experience with the algorithmic decision context significantly influence perceptions about the fairness of using different features for bail decision-making. Drawing on our results, we discuss the implications for stakeholder engagement and algorithmic oversight including the need to consider multiple dimensions of diversity in composing oversight and regulatory bodies.

Motivation & Objective

• To investigate how sociodemographic factors and personal experiences affect human perceptions of fairness in algorithmic decision-making.
• To examine the influence of political views and lived experiences on judgments about using specific features (e.g., criminal history) in bail decisions.
• To inform the composition of diverse oversight and regulatory bodies by identifying which dimensions of diversity significantly impact fairness perceptions.
• To move beyond demographic diversity by exploring the role of context-specific personal experiences in shaping moral judgments about algorithmic fairness.

Proposed method

• Conducted a human-subject survey with 329 participants from the U.S. to assess fairness perceptions of algorithmic features in bail decisions.
• Collected data on demographics (age, education, gender, race, political views) and personal experiences (e.g., attending a bail hearing).
• Used a structured questionnaire to rate the fairness of 12 algorithmic features for bail decisions on a Likert scale.
• Applied statistical analysis to examine the relationship between demographic variables, personal experiences, and fairness judgments.
• Controlled for confounding variables and tested for significance using regression models.
• Explored potential mechanisms such as egocentric bias and information acquisition from personal experiences.

Experimental results

Research questions

• RQ1 How do political views influence people’s perceptions of fairness in using specific features for algorithmic bail decisions?
• RQ2 To what extent do personal experiences—such as attending a bail hearing—affect fairness judgments about algorithmic features?
• RQ3 Are there specific demographic factors (e.g., race, education, gender) that significantly predict differences in fairness perceptions?
• RQ4 How do personal experiences interact with political ideology in shaping fairness judgments?
• RQ5 What role does contextual knowledge (e.g., awareness of legal protections for juvenile records) play in fairness perceptions?

Key findings

• Left-leaning individuals consistently rated algorithmic bail decisions as less fair than right-leaning individuals, regardless of the specific features used.
• Participants who had attended a bail hearing were significantly less likely to consider using juvenile criminal history as fair, indicating a strong influence of direct experience.
• The experience of attending a bail hearing was negatively correlated with perceived fairness of using juvenile criminal history, even after controlling for political views.
• Political ideology emerged as a stronger predictor of fairness judgments than most demographic factors, suggesting it is a critical dimension of diversity in oversight bodies.
• Personal experiences such as attending a bail hearing may lead to more informed or context-specific fairness judgments, though they may also introduce bias depending on the context.
• The study highlights that legal protections for certain data (e.g., juvenile records) may be less known to the general public, and personal exposure can significantly alter fairness perceptions.