Enhancing or impeding? Exploring the dual impact of anthropomorphism in large language models on user aggression

Abstract

This study explores the impact of anthropomorphism in large language models (LLMs) on user aggression through the lens of dehumanization theory. Specifically, it analyzes how chatbots’ human nature and human uniqueness traits influence user aggression by triggering perceived identity threats. Drawing on an online survey of 1,000 LLM chatbot users in China and employing structural equation modeling, the research reveals that chatbots perceived as competent and rational tend to reduce user aggression by alleviating identity threats. In contrast, chatbots exhibiting empathetic and moral traits are more likely to heighten identity threats, thereby increasing aggression. The study further demonstrates that perceived economic value mitigates the negative impact of identity threats, while perceived emotional value exacerbates it. These findings highlight the critical need for AI designs that not only enhance user interaction but also carefully manage the potential for eliciting adverse behaviors.