Algorithm Reliance: Fast and Slow
通过实验室实验，研究工人在算法辅助服务中如何决定是否采纳算法建议以及采纳速度，发现算法质量和系统负荷共同影响决策质量和速度。
In algorithm-augmented service contexts where workers have decision authority, they face two decisions about the algorithm: whether to follow its advice and how quickly to do so. The pressure to work quickly increases with the speed of arriving customers. In this paper, we ask the following. How do workers use algorithms to manage system loads? With a laboratory experiment, we find that superior algorithm quality and high system loads increase participants’ willingness to use their algorithm’s advice. Consequently, participants with the superior algorithm make higher-quality recommendations than those with no algorithm (participants with the inferior algorithm make slightly lower-quality recommendations than those without). However, participants do not necessarily speed up by using algorithms’ advice; their throughput times only decrease compared with the no-algorithm baseline when the system load is high and algorithm quality is superior, although participants would benefit from working faster in all treatments. This happens in part because participants in the high-load, superior-algorithm treatment serve customers more quickly than participants in the other treatments, conditional on using the algorithm. Participants in the high-load, superior-algorithm treatment work especially quickly in later periods as they increasingly default to their algorithm’s advice. Our findings show that algorithms can have benefits for both decision quality and speed. Quality benefits come from workers’ decision to use their algorithms’ advice, whereas speed benefits depend on workers’ algorithm use and the time they spend deliberating about their algorithm use. Ultimately, algorithm quality and system load are mutually reinforcing factors that influence both service quality and especially speed. This paper was accepted by Elena Katok, Special Issue on the Human-Algorithm Connection. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.01989 .