# Conversational Swarm Intelligence, a Pilot Study

**Authors:** Louis Rosenberg, Gregg Willcox, Hans Schumann, Miles Bader, Ganesh Mani, Kokoro Sagae, Devang Acharya, Yuxin Zheng, Andrew Kim, Jialing Deng  
**arXiv:** [2309.03220](https://arxiv.org/abs/2309.03220)  
**Date:** 2023-08-31  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2309.03220v1/

## Key points

- The pilot study introduces Conversational Swarm Intelligence (CSI) as a method for facilitating real-time networked conversations among large human groups.
- CSI utilizes conversational agents powered by Large Language Models (LLMs) to enable local dialog within small deliberative groups and global propagation of conversational content across a larger population.
- The aim of CSI is to combine the advantages of small-group deliberative reasoning with the benefits of large-scale collective intelligence.
- Participants using the CSI structure produced 30% more contributions than those in the centralized chat room, with 7.2% less variance in contribution quantity among users.
- These findings suggest that CSI has the potential to enhance collaboration and information sharing within large human groups by leveraging swarm-like dynamics and advanced language models.
- The study is pending for presentation at the Collective Intelligence 2023 conference organized by ACM.

## Summary

In the pilot study titled "Conversational Swarm Intelligence, a Pilot Study," conducted by Louis Rosenberg, Gregg Willcox, Hans Schumann, Miles Bader, Ganesh Mani, Kokoro Sagae, Devang Acharya, Yuxin Zheng, Andrew Kim and Jialing Deng, the researchers introduce Conversational Swarm Intelligence (CSI) as a novel method for facilitating real-time networked conversations among large human groups. Inspired by the dynamics of biological swarms, CSI utilizes conversational agents powered by Large Language Models (LLMs) to enable both local dialog within small deliberative groups and global propagation of conversational content across a larger population. The aim of CSI is to combine the advantages of small-group deliberative reasoning with the benefits of large-scale collective intelligence. To evaluate its effectiveness, participants engaged in text chat-based deliberations in conversational swarms were compared to those using a standard centralized chat room. The results showed that participants using the CSI structure produced 30% more contributions than those in the centralized chat room (p<0.05). Additionally there was 7.2% less variance in contribution quantity among users in the CSI structure. These findings indicate that users contributed more content and participated more evenly when utilizing the CSI structure for their conversations. This suggests that CSI has the potential to enhance collaboration and information sharing within large human groups by leveraging swarm-like dynamics and advanced language models. The study is pending for presentation at the Collective Intelligence 2023 conference organized by ACM. Overall this pilot study highlights how Conversational Swarm Intelligence can effectively facilitate real-time networked conversations among large human groups while promoting increased participation and balanced contribution levels.