AI Literacy and Procedural Justice in Algorithmic Systems: A Conditional Account on Data Evaluation Practices and Human Oversight

Abstract

Procedural justice is a central lens to understand how individuals evaluate algorithmic procedures. It involves different dimensions, reflecting both relational concerns about how people perceive they are treated, and instrumental concerns about whether individuals perceive control within algorithmic processes. Focusing on data evaluation practices (rule-based vs. AI-based) and human oversight (human-in-the-loop vs. human-out-of-the-loop) as two distinct algorithmic procedures, we advance a conditional account of procedural justice. We posit that the relationship between algorithmic procedures and perceived procedural justice depends on individuals’ AI literacy and on the dimension of procedural justice considered. We test these arguments using data from a large-scale vignette experiment in a behavioral regulation context (N = 3,203), and find that distinct aspects of AI literacy operate asymmetrically across algorithmic procedures and dimensions of procedural justice. AI awareness conditions how data evaluation practices translate into procedural justice perceptions, but plays a substantially weaker role for evaluations of human oversight. Human-out-of-the-loop oversight increases perceived control, but simultaneously reduces relational perceptions. These findings demonstrate that aggregate evaluations of procedural justice can conceal important heterogeneity, as algorithmic procedures may produce opposing effects across perceived procedural justice dimensions or across individuals with different levels of AI literacy. Our study contributes to the debate on the ethics of algorithmic systems by showing that perceived procedural justice depends on how individuals understand algorithmic procedures, and that it is shaped by the multidimensional nature of procedural justice.