|
Eduardo Israel

Are Decision Makers Really Averse to Algorithms? Challenging the Findings on Current Research on Algorithm Aversion and Suggestions For Future Research

Tuesday, 31. May 2022 - 12:00 to 13:00

Talk title: Are Decision Makers Really Averse to Algorithms? Challenging the Findings on Current Research on Algorithm Aversion and Suggestions For Future Research

Speaker affiliation: Dr. Ekaterina Jussupow is a Post-Doctoral Researcher in Information Systems at the University of Mannheim, Germany. Her research interests include the human-AI-collaboration, specifically how decision makers evaluate and make joint decisions with AI systems. She received her Ph.D. in Information Systems from the University of Mannheim in 2021 for her dissertation on the implications of Artificial Intelligence on medical expertise and decision making. Her research has been published in ISR, JMIR Formative Research, and BISE.

Talk abstract: Even though algorithms increasingly influence humans in daily life and workplaces, scientific research has shown that decision makers have an aversion toward algorithms. Specifically, decision makers often have more favorable evaluations of humans and prefer advice provided by humans instead of algorithms. As a result, decision makers fail to fully benefit from the computational abilities of algorithms. This working paper aims to critically examine the experimental studies on algorithm aversion to discuss whether decision makers are really averse to algorithms. Specifically, we investigate four problem areas that make the interpretation and synthesis of algorithm aversion research difficult as these problems hinder the synthesis of current experimental findings. First, we consider what scholars mean by the term algorithm aversion and which theoretical foundations are used to hypothesize on influencing factors and the boundaries of algorithm aversion versus related constructs. Second, we investigate how the experimental design confounds the findings on algorithm aversion. Third, we discuss the lack of consistent measurement of algorithm aversion. Lastly, we demonstrate in the context of advice-taking decisions which further moderators need to be considered. Based on those four problem areas, we derive suggestions for future research that help to avoid confounding effects.