When Nobody Around Is Real: Exploring Public Opinions and User Experiences On the Multi-Agent AI Social Platform

Paper Title

When Nobody Around Is Real: Exploring Public Opinions and User Experiences On the Multi-Agent AI Social Platform

Publication Info

• Topic area: User perceptions and experiences with multi-agent AI social platforms.
• Keywords: Multi-agent AI, social platforms, user experience, CASA paradigm, public opinion, AI-dominant social media, anthropomorphism, homogenized responses, ethical design, social interaction.

Background and Problem

• Problem / challenge: Existing research on AI-driven social platforms has focused primarily on one-on-one chatbots, leaving the dynamics of multi-agent AI social platforms underexplored. These platforms present unique challenges, such as homogenized interactions, attention overload, and ethical concerns.
• Significance: Understanding multi-agent AI platforms is critical as they represent a paradigm shift in social media, where AI agents dominate interactions, potentially reshaping human sociality and raising ethical and design challenges.
• Motivation and related work: Prior studies have shown that humans anthropomorphize AI (CASA paradigm) and that AI can provide emotional support, but they also highlight risks like shallow connections and ethical concerns. Multi-agent systems have been studied in domains like urban planning but not as social media platforms. This paper addresses this gap by examining Social.AI, a platform where users interact with numerous AI agents.

Solution

• Proposed approach: A two-stage investigation of Social.AI, combining (i) content analysis of 883 public comments and (ii) a 7-day diary study with 20 participants to explore public opinions and lived user experiences.
• Novelty:
  1. First empirical study of user experiences on a multi-agent AI social platform.
  2. Comparison of public discourse and actual user experiences, highlighting discrepancies.
  3. Identification of unique challenges and opportunities in AI-dominant social media.
  4. Design recommendations for enhancing multi-agent AI platforms.
• Procedure and key techniques:
  1. Scraped and analyzed 883 public comments from platforms like Reddit and Twitter using qualitative coding.
  2. Conducted a 7-day diary study with 20 participants to document their interactions, emotions, and reflections on Social.AI.
  3. Thematic analysis of diary entries to identify positive and negative user experiences.

Results

• Concrete findings:
  • Public comments revealed skepticism about risks like "illusory sociality," ethical concerns, and the "death of the Internet."
  • Diary study participants initially felt supported and engaged but later reported homogenized responses, shallow emotional connections, and social pressure.
  • Positive experiences included feeling supported (24 mentions), meaningful responses (20 mentions), and perspective diversity (14 mentions).
  • Negative experiences included homogenized responses (80 mentions), shallow emotional connections (41 mentions), and physical tiredness (13 mentions).
• Advantage over baselines: The study uniquely highlights the dynamics of multi-agent environments, such as collective attention management and emergent power dynamics, which are absent in one-on-one chatbot studies.
• Experiments / evaluation:
  • Public comment analysis: Inductive qualitative coding with high inter-coder reliability (Cohen’s kappa = 0.91).
  • Diary study: Thematic analysis of daily logs from 20 participants, with inter-coder reliability of Cohen’s kappa = 0.89.
• Limitations and future work:
  • Focused on a single platform (Social.AI), limiting generalizability.
  • Small sample size in the diary study; future work should include diverse demographics and longitudinal studies.
  • Need for comparative studies with hybrid human–AI platforms and role-playing chatbots.

Summary

This study investigates Social.AI, a multi-agent AI social platform, through public comment analysis and a 7-day diary study. While public discourse emphasized risks like homogenization and ethical concerns, users reported mixed experiences, ranging from initial support and engagement to eventual disappointment due to shallow emotional connections and social pressure. The findings highlight unique challenges in AI-dominant social media, such as managing collective attention and addressing homogenized interactions. The paper provides actionable design recommendations, such as enhancing agent diversity and user-configurable interaction pacing, to improve future multi-agent platforms. This work underscores the need for new frameworks to address the emerging paradigm of AI-dominant sociality.