Performance rather than reputation affects humans’ trust towards an artificial agent

Abstract

To succeed in teamwork with artificial agents, humans have to calibrate their trust towards agents based on information they receive about an agent before interaction (reputation information) as well as on experiences they have during interaction (agent performance). This study (N = 253) focused on the influence of a virtual agent's reputation (high/low) and actual observed performance (high/low) on a human user's behavioral trust (delegation behavior) and self-reported trust (questionnaires) in a cooperative Tetris game. The main findings suggested that agent reputation influences self-reported trust prior to interaction. However, the effect of reputation immediately got overridden by performance of the agent during the interaction. The agent's performance during the interactive task influenced delegation behavior, as well as self-reported trust measured post-interaction. Pre-to post-change in self-reported trust was significantly larger when reputation and performance were incongruent. We concluded that reputation might have had a smaller than expected influence on behavior in the presence of a novel tool that afforded exploration. Our research contributes to understanding trust and delegation dynamics, which is crucial for the design and adequate use of artificial agent team partners in a world of digital transformation.