Effect of Confidence Indicators on Trust in AI-Generated Profiles

Abstract

Artificial Intelligence (AI) is increasingly augmenting and generating online content, but research suggests that users distrust content which they believe to be AI-generated. In this paper, we study whether introducing a confidence indicator, a text rating of an algorithm's confidence in its source data alongside rationale for why the data is more or less trustworthy, affects this distrust in Airbnb host profiles believed to be computer-generated. Our results indicate that a low-confidence indicator decreases participant trust in the rental host, but high-confidence indicators have no significant impact on trust. These findings suggest that user trust of AI-generated content can be negatively, but not positively, affected by a confidence indicator.