The main message and insights of this research are as follows:
- It has been confirmed that AI-generated text (smart replies) exhibits a strong positive bias, being much more positive than human-generated text.
- The overall messages of senders using AI text became more positive compared to senders not using AI text, although there was no change in the sender’s own language (excluding AI text).
- The positive bias in AI text may lead to a decrease in the evaluation of the social attractiveness of the conversation partner, suggesting unexpected negative effects.
- There were no detrimental effects observed on task performance (accuracy or conversation length) due to the introduction of AI text.
- The concept of adjacency pairs was shown to be useful in the analysis of AI-mediated communication, proposing a new analytical framework.
- Identifying a new type of message called “human-AI composite messages,” which combine AI text with human text.
Main contributions:
- Experimentally verifying the impact of positive bias in AI-mediated communication.
- Proposing a new analytical method for AI-mediated communication using the concept of adjacency pairs.
- Pointing out the potential negative effects of AI text on interpersonal cognition.
- Introducing the concept of “human-AI composite messages.”
- This research offers important insights into understanding the effects of language patterns, interpersonal cognition, and task performance as AI becomes integrated into everyday communication.