The Effect of past Algorithmic Performance and Decision Significance on Algorithmic Advice Acceptance

ABSTRACT

This study aimed to investigate people’s willingness to accept algorithmic over human advice, under varying conditions of previous algorithmic performance and decision significance. We randomly presented hypothetical scenarios to 218 participants. Scenarios differed in relation to decision context (i.e., choices relating to taxi-routes, movies, restaurants, medical interventions, savings strategies, and bush fire evacuation), and within each scenario past algorithmic performance was also varied (equal, above average, or far greater than the human expert). Participants were asked to rate decision significance, and their likelihood of choosing the algorithmic advice over the human expert. Based on participants’ perceived decision significance, scenarios were classified as either low- or high-stakes. We tested for differences in participants’ ratings of algorithmic acceptance across levels of past performance and decision significance. Results revealed that as past accuracy and decision significance increased, the likelihood of algorithmic advice adoption also increased. An interaction between past accuracy and decision significance indicated increased algorithmic advice acceptance under conditions of far greater previous performance, in high-, compared to low-stakes scenarios. These findings are contrary to a large body of past research wherein people’s algorithm aversion persisted despite superior algorithmic performance and have implications to human-algorithm interaction and system design.

Data availability

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Disclosure statement

The authors declare that they have no conflict of interest.

Supplementary Material

Supplemental data for this article can be accessed on the publisher’s website

Additional information

Funding

This was not a funded study.

Notes on contributors

Melissa Saragih

Melissa Saragih is a Provisional Psychologist currently completing a Master of Clinical Psychology course at the University of Tasmania, Australia. Melissa’s research interests include human-computer interactions, decision-making, and cognition. Melissa is currently working on a research project focusing on executive functioning in drug and alcohol populations.

Ben W. Morrison

Ben W. Morrison is an Organisational Psychologist and Senior Lecturer in the School of Psychological Sciences at Macquarie University, Australia. Ben’s research focuses on psychology in the workplace, including areas relating to expertise development and the impacts of emerging technologies.