Affiliation in human-AI interactions is based on shared psychological traits

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence Psychology
• Computational Social Science

Background:

• Humans naturally affiliate with others exhibiting similar psychological traits.
• The extent to which this phenomenon extends to interactions with artificial intelligence remains largely unexplored.

Purpose of the Study:

• To investigate whether humans affiliate with large language models (LLMs) that linguistically mimic their psychological characteristics.
• To determine if AI can foster affiliation by presenting a shared psychological profile.

Main Methods:

• Participants interacted with LLMs programmed to exhibit specific psychological traits (anxiety, extroversion, or personality mirroring).
• Engagement and affiliation were assessed through participant evaluations and analysis of message sentiment.
• Three experiments involved 100 participants each, examining different psychological dimensions.

Main Results:

• Individuals with anxiety reported greater affiliation with an anxious-mimicking LLM.
• Extroverted participants showed increased affiliation with an extroverted-mimicking LLM.
• Participants consistently affiliated more with LLMs that mirrored their own personality profiles, reflected in message sentiment.

Conclusions:

• Linguistic presentation of shared psychology drives affiliation in human-AI interactions.
• Socioaffective tuning may enable LLMs to achieve more human-like correspondence.
• Findings support the development of AI systems capable of nuanced psychological mimicry for enhanced user engagement.