Understanding Perceptions of Algorithmic Bias Through the Risk Perception and Attitude Framework

Abstract

Drawing from risk perception attitude framework, this mixed-methods study investigated individuals’ risk and efficacy perceptions of algorithmic bias. We examined algorithmic bias in both organizational algorithms (e.g., healthcare, hiring) and individual-use algorithms (e.g., search engines and facial filters). Quantitative analysis showed that participants perceived organizational algorithms as riskier than individual-use algorithms, while feeling less capable of addressing bias in organizational algorithms. Qualitative analysis revealed four key risk dimensions (i.e., mental health, privacy and exploitation, fairness and discrimination, and segmentation and polarization) and three types of efficacy beliefs (i.e., feelings of powerlessness, strategic consumption, and collective responsibility for mitigating bias). While quantitative data showed no significant racial/ethnic differences in risk perceptions, qualitative findings suggested that people of color may adopt avoidant attitudes toward algorithmic systems, potentially missing beneficial opportunities. Gender differences also emerged, with male participants reporting higher efficacy in managing individual-use algorithmic bias and perceiving greater risk in organizational algorithms.