The Impact of Appearance and Task Outcome on User Psychology in Interactions with LLM-Based Agents

Abstract

In this study, we investigated the impact of an agent's appearance and its task outcomes on user psychology during interactions with agents based on Large Language Models (LLMs). We conducted a between-subjects experiment in which participants used an agent to perform a creative task, specifically writing a long-form text for a prize contest. We manipulated the agent's appearance with four variations and the task outcome with three levels (success, partial success, and failure). Satisfaction with the results of the application and impressions of the agent's task execution were investigated in 20 questionnaire items. The results indicated that while the agent's appearance did not significantly influence user perceptions, the task outcome had a significant effect on the user's evaluation of both the agent and the text it generated. Notably, these evaluations dropped markedly when the task resulted in failure, suggesting a disproportionately high perceived risk associated with agent failure. Furthermore, the task outcome affected the perceived locus of responsibility. These findings highlight that when designing LLM-based agents for tasks with uncertain outcomes, it is crucial to consider not only the agent's functionality but also the psychological aspects of how users perceive and interpret these outcomes.

Ähnliche Arbeiten