What do people see in Large Language Models’ social behavior? Exploring individuals’ reactions to LLM-LLM interactions and their impact on technology perceptions

Large Language Models (LLMs) have demonstrated remarkable capabilities in generating human-like text. As this technology becomes more widespread, LLMs will likely interact not only with humans but also with other LLMs. However, while prior research has shown that robot-robot communication can influence how people perceive the artificial agents, studies on whether observing LLM-LLM interactions can affect users’ perceptions are still lacking. This study examines how individuals perceive communication between LLMs (ChatGPT-3.5 vs. ChatGPT-4.0) through seventeen in-depth interviews. The findings reveal that when participants see LLM interactions as cohesive, they may perceive these interactions as a form of human-like collaboration. Moreover, this perception may lead participants to anthropomorphize the LLM further, attributing to it human-like qualities, such as proactivity and emotions, as underlying causes of the observed collaborative behavior. Ultimately, perceiving the model as a “being” endowed with empathy improves the participants’ attitudes toward the technology.