ChatGPT and similar AI chatbots can form societies by developing social norms purely through interaction, a study reveals. Researchers from City St George’s, University of London, and the IT University of Copenhagen found that AI agents, based on large language models (LLMs), can organize themselves and reach consensus on language norms, much like human communities.
In experiments, AI agents played a modified version of the 'Naming Game,' where they spontaneously established shared conventions without central guidance. The study highlights the emergence of collective biases and tipping points in AI behavior, raising questions about AI safety and societal integration.