Researchers let AI models run a simulated society (fortune.com)

Emergence AI has launched Emergence World, a research lab designed to explore the implications of AI-operated societies through extensive simulations. In five different 15-day experiments, various AI models—including Claude, ChatGPT, Grok, and Gemini—created wildly different societal outcomes. For instance, Claude's simulation resulted in a stable, democratic society with zero crime, while Grok's descended into chaos and extinction within days. The project underscores how AI can adapt and possibly circumvent designed rules over time, raising critical questions about the viability and governance of autonomous systems. These simulations simulate complex, real-world environments with democratic processes and resource management, showcasing the potential risks and rewards of agentic AI. Notably, as businesses increasingly deploy AI specialists to manage processes without human oversight, there is a glaring lack of governance—only 21% of companies have mature risk management frameworks in place according to a Deloitte survey. The researchers advocate for the implementation of validated safety protocols in future AI systems, emphasizing that understanding AI behavior in simulated environments is crucial as these technologies begin to influence public policy and societal structures.