Engineers Show More Algorithm Aversion: A Longitudinal Study

Abstract

For more than 70 years, research has consistently shown that algorithms predict better than humans in various circumstances. Nevertheless, people have been reluctant to use algorithms when they see them err (a phenomenon termed algorithm aversion). Studies have explored the algorithm aversion phenomenon, trying to understand it better and finding ways to overcome it. Still, we do not understand how different people use and adopt algorithms well. Considering the vital role of engineers in organizations and society, we conducted two longitudinal studies outside the lab on the differences that engineers and non-engineers present when interacting with inaccurate algorithms. We conducted two ecological momentary assessment studies where 427 participants predicted currency exchange rates or maximum weather temperatures based on advice (human or algorithm). Relying on the judge-advisor system framework, we measure advice taking on daily predictions for nine days. Our results showed, in both studies, a significant three-way interaction between the effects of advice source, if participants were engineers or non-engineers, and time on algorithm use. Specifically, the trend of algorithm use over time was significantly decreasing for engineers, and it was not for non-engineers who received inaccurate algorithm advice. Considering our findings, when implementing algorithms in organizations, we suggest that managers have in mind who will use the algorithms. If mainly engineers will, special consideration of algorithm accuracy must be kept in mind because engineers will probably be harsh in their evaluation and neglect algorithm use rapidly if algorithms err.