Algorithms in decision-making : exploration of algorithm aversion and appreciation

Abstract

As artificial intelligence and machine learning progress, algorithms are increasingly outperforming humans in various decision-making tasks. Such advancement has led to a growing reliance on algorithmic decisions. Yet, perhaps surprisingly, not everyone is keen to follow algorithmic advice, even when it outperforms human judgment. This reluctance, known as algorithm aversion, is a behavioural anomaly that poses a significant challenge to the optimal use of algorithms. This research initiates an exploration to understand this paradox through a systematic literature review and empirical studies. It aims to synthesise existing findings, uncover novel influencing factors, and propose frameworks for understanding and mitigating algorithm aversion and enhancing algorithm appreciation.

The dissertation comprises four publications. The first publication systematically reviews 80 empirical studies and categorises the factors influencing algorithm aversion under four themes: algorithm, individual, task, and high-level factors. Additionally, this publication highlights gaps and identifies future research avenues. The second publication, drawing on innovation resistance theory and the technology readiness index, proposes and tests a structural model relating perceived barriers to algorithm aversion. It also examines how technology readiness might alter the influence of these barriers on aversion to algorithmic decision-making. The third publication investigates how individuals’ familiarity with algorithms and tasks affects their attitudes towards algorithms and how these attitudes influence algorithm aversion. The fourth publication extends the investigation to algorithm appreciation, examining the roles of familiarity with algorithms, tasks, and algorithm performance in enhancing trust in algorithms and fostering algorithm appreciation.

The findings reveal that while algorithm and individual factors have been extensively studied, task and high-level factors have received less attention to date. Perceived barriers related to value, tradition, and image significantly affect algorithm aversion, while technology readiness reduces the impact of value barriers. Familiarity with algorithms positively impacts attitudes towards algorithms, which, in turn, negatively influences algorithm aversion. Furthermore, familiarity with algorithm performance moderates the relationship between familiarity with algorithms and trust, highlighting the importance of algorithm transparency in appreciating algorithms.

This dissertation contributes to the theoretical understanding of algorithm aversion and appreciation and offers practical implications for algorithm designers and managers. Furthermore, this dissertation provides valuable insights for overcoming algorithm aversion and increasing algorithm appreciation. It emphasises the necessity of addressing perceived barriers, enhancing technology readiness, familiarisation with algorithms and their performances, and building trust to increase algorithm acceptance. The comprehensive framework and novel measurement scale developed through this research lay a solid foundation for future inquiries.