SLR on Algorithmic Decision Making

Abstract With the continuing application of artificial intelligence (AI) technologies into decision-making, algorithmic decision-making is becoming more efficient, even often outperforming human counterpart. Despite this superior performance, people often consciously or unconsciously display reluctance to rely on algorithms, a phenomenon known as algorithm aversion. Viewed as a behavioral anomaly, algorithm aversion has recently attracted much scholarly attention. With a view to synthesize the findings of this literature, we systematically review 80 empirical studies identified through searching in seven academic databases and performing citation chaining. We map the emergent themes following grounded theory and categorize the influencing factors of algorithm aversion under four main themes: algorithm, individual, task, and high-level. Our analysis reveals that although algorithm and individual factors have been investigated extensively, very little effort has been given to explore the task and high-level factors. We contribute to algorithm aversion literature by proposing a comprehensive framework, highlighting open issues in existing studies, and outlining several research avenues that could be handled in future research. Implications for research and practitioners about the findings of the study are discussed.

摘要 随着人工智能（Artificial Intelligence，简称AI）技术持续应用于决策领域，算法决策的效率日益提升，甚至往往优于人类决策者。尽管表现优异，人们仍时常有意或无意地表现出对依赖算法的抵触情绪，这一现象被称为算法厌恶（algorithm aversion）。作为一种行为异常现象，算法厌恶近年来受到了学界的广泛关注。为整合该领域的研究成果，本文通过检索7个学术数据库并开展引文链检索，系统梳理了80项实证研究。本文基于扎根理论梳理出新兴研究主题，并将算法厌恶的影响因素划分为四大核心类别：算法层面、个体层面、任务层面与宏观层面。分析结果显示，尽管学界已对算法与个体层面的影响因素开展了大量研究，但针对任务层面与宏观层面影响因素的探索仍极为有限。本文通过构建综合分析框架、点明现有研究中的开放性问题，并梳理出若干未来可探索的研究方向，为算法厌恶领域的学术研究作出贡献。最后，本文探讨了本研究发现对学术研究与实务从业者的启示。