Sentiment analysis of Artificial Advisors in Search and Rescue Tasks

Abstract

The potential of Artificial Intelligence in assisting human teamwork has yet to be fully realized, despite its success in other domains. To ensure AI’s effectiveness and credibility as a team advisor, it must be able to effectively infer team dynamics and issue appropriate interventions. This study focuses on AI-mediated human teamwork in an simulated search and rescue (SAR) task, where a team of humans is monitored and guided by an artifical social intelligence (ASI). Six different ASIs are compared against a human baseline investigating the characteristics and effectiveness of their interventions. When adjusted for initial player competence ASIs performed on par with the human advisor although the human advisor was rated as more trustworthy and useful. Additionally, sentiment analysis of the interventions reveals that participants were more likely to accept interventions with negative emotions and resulted in improved team performance.