Fairness perceptions of algorithmic decision-making: A systematic review of the empirical literature

Abstract

Algorithmic decision-making increasingly shapes people's daily lives. Given that such autonomous systems can cause severe harm to individuals and social groups, fairness concerns have arisen. A human-centric approach demanded by scholars and policymakers requires considering people's fairness perceptions when designing and implementing algorithmic decision-making. We provide a comprehensive, systematic literature review synthesizing the existing empirical insights on perceptions of algorithmic fairness from 58 empirical studies spanning multiple domains and scientific disciplines. Through thorough coding, we systemize the current empirical literature along four dimensions: (1) algorithmic predictors, (2) human predictors, (3) comparative effects (human decision-making vs. algorithmic decision-making), and (4) consequences of algorithmic decision-making. While we identify much heterogeneity around the theoretical concepts and empirical measurements of algorithmic fairness, the insights come almost exclusively from Western-democratic contexts. By advocating for more interdisciplinary research adopting a society-in-the-loop framework, we hope our work will contribute to fairer and more responsible algorithmic decision-making.

Humans versus machines: Who is perceived to decide fairer? Experimental evidence on attitudes toward automated decision-making

“…A literature review by Starke et al. 11 identified several papers that investigated humans’ perceptions of algorithmic fairness. We focus on four key dimensions that have been investigated with respect to perceptions of algorithmic fairness: (1) the context in which an ADM system is applied and the type of impact the system makes, (2) the degree of human involvement in decision-making, (3) the features used by an algorithm, and (4) the characteristics of the individual that may influence perceptions of algorithmic fairness.…”

“…The first dimension is concerned with the contexts in which ADM systems are used and the impact of a decision for an individual’s life. 11 , 12 Previous research highlighted that empirical results on perceptions in specific ADM contexts may not translate into other contexts, cautioning researchers against over-generalizations. 10 Although each context comes with myriads of idiosyncrasies, it appears likely that the stakes of the decision-making context are one crucial differentiating factor.…”

“…Relating to the second dimension, some research exists on direct comparisons between human and (purely) ADM for specific contexts 11 and concludes that there is great variation in relative perceived fairness across contexts and that characteristics of the task impact fairness perceptions. In a series of survey experiments, Nagtegaal 17 found that public sector employees perceived human decision-makers as procedurally fairer for tasks with high complexity, and that adding an algorithm to a human in the decision-making process may increase justice perceptions.…”

“… 20 With an Amazon MTurk sample, Starke et al. 11 found that ADM decisions overseen by a “privacy professional” increased perceived legitimacy of the decision compared with purely algorithmic or human decisions. Overall, a literature review by Langer and Landers 21 suggests that hybrid decision-making is preferred over fully ADM, at least in specific contexts.…”

“…However, the review study by Starke et al. 11 finds no clear public preference for whether solely human decision-making or a hybrid process involving humans and algorithms was preferred and conclude that no general statement on the preference for either human or ADM could be made. The literature may therefore profit from a systematic comparison of degrees of automation in several major ADM applications contexts with a large and probability-based sample.…”