Know What Not To Know: Users' Perception of Abstaining Classifiers

TL;DR

This study investigates how users perceive machine learning systems that abstain from labeling ambiguous data points, revealing that abstention does not harm perceived trustworthiness and can improve decision quality.

Contribution

It provides first empirical insights into user reactions to abstaining classifiers and demonstrates that abstention can maintain trust while enhancing decision accuracy.

Findings

Users are unconsciously influenced by label suggestions on ambiguous data.

Participants view abstaining systems as equally trustworthy as always-labeling systems.

Abstaining classifiers can improve decision quality without reducing perceived credibility.

Abstract

Machine learning systems can help humans to make decisions by providing decision suggestions (i.e., a label for a datapoint). However, individual datapoints do not always provide enough clear evidence to make confident suggestions. Although methods exist that enable systems to identify those datapoints and subsequently abstain from suggesting a label, it remains unclear how users would react to such system behavior. This paper presents first findings from a user study on systems that do or do not abstain from labeling ambiguous datapoints. Our results show that label suggestions on ambiguous datapoints bear a high risk of unconsciously influencing the users' decisions, even toward incorrect ones. Furthermore, participants perceived a system that abstains from labeling uncertain datapoints as equally competent and trustworthy as a system that delivers label suggestions for all datapoints. Consequently, if abstaining does not impair a system's credibility, it can be a useful mechanism to increase decision quality.