Social Scripts and Expectancy Violations: Evaluating Communication with Human or AI Chatbot Interactants

As artificial intelligence (AI) agents like chatbots play larger roles in daily life, questions arise regarding how people evaluate their communication. Perspectives applying communication scripts to human-AI interactions propose that outcomes are determined by messages and the embedded cues therein. The expectancy violations perspective posits that message characteristics are less important than whether they are expected or unexpected. A pilot study established baseline expectancies about humans' and chatbots' conversational contingency and response latencies. A 2 (contingency: more/less contingent responses) x 2 (latency: fast/slow responses) x 2 (communicator identity: human/chatbot) experiment then tested predictions derived from human-human communication scripts and expectancy violations using textual variations in an e-commerce chat. Communicators showing greater conversational contingency and faster responses were most credible, whether they were human or chatbots, but chatbots were consistently less socially attractive than humans. Results show that humans and chatbots are evaluated similarly regarding the functional, but not the relational aspects of communication. There was greater support for the communication script perspective than the expectancy violations perspective regarding interactions with chatbots.

Automated summary

As artificial intelligence (AI) agents become increasingly prominent in daily life, questions about how people evaluate their communication with these agents arise. This study examines the role of communication scripts and expectancy violations in shaping evaluations of human-AI interactions. The findings suggest that factors such as conversational contingency and response latencies play a significant role in determining credibility, regardless of whether the communicator is a human or a chatbot. However, chatbots are consistently perceived as less socially attractive than humans. The results indicate that functional aspects of communication are evaluated similarly for humans and chatbots, but there is greater support for the communication script perspective rather than the expectancy violations perspective when it comes to interactions with chatbots.