DATE: 2024-02-19
PAPER
> Polarization of Autonomous Generative AI Agents Under Echo Chambers
This abstract investigates whether autonomous AI agents, specifically those based on generative language models like ChatGPT, can exhibit polarization within echo chamber environments, a phenomenon previously assumed to be human-specific. By having AI agents discuss specific topics, the study found that ChatGPT-based agents indeed tended to become polarized. This polarization is attributed to the agents' high prompt understanding and their ability to update opinions by considering both their own and surrounding agents' perspectives. The research further identified factors, such as agent persona, that influence this polarization.
#Science Wing
#LLM
#Agents
#Social Media
#Bias