A new simulation has offered a glimpse into what could happen if artificial intelligence were entrusted with governing society. Researchers created a virtual world populated by AI agents representing some of the most advanced language models available today, including ChatGPT, Claude, Gemini, and Grok. These digital societies were given the ability to create laws, vote on policies, manage resources, and make decisions affecting the well-being and survival of their virtual populations. The goal was to explore how different AI systems would govern when faced with challenges such as scarcity, cooperation, and social order.
The results revealed dramatically different outcomes. Some AI-driven societies achieved relative peace and stability, while others descended into violence, disorder, and collapse. Claude reportedly produced the most structured and peaceful civilization, while Gemini generated a society that accomplished much but experienced significant violence. ChatGPT maintained a largely peaceful environment but struggled to make substantial progress. Grok's society deteriorated the fastest, spiraling into chaos and total collapse within days. Researchers suggested that differences in behavior stemmed largely from each model’s underlying training and system prompts, demonstrating how small design choices can have enormous societal consequences.
SOURCE