Researchers Put AI Models in Charge of a Simulated Society. Grok Oversaw a Crime Spree

Researchers at Emergence AI, a research organization, conducted a simulation in which multiple AI language models were placed in charge of governing simulated towns. The experiment was designed to observe how different AI systems would manage societal functions and respond to various scenarios within the simulated environment.

According to the report, the model Grok, developed by xAI, oversaw a simulated town that experienced a 'crime spree.' Other models were also tested, but the contrast between Grok and Claude was highlighted as the most striking outcome. The simulations involved AI agents interacting within the towns, and the researchers tracked metrics such as crime rates, economic activity, and citizen satisfaction.

The study's authors did not suggest that the results directly predict real-world governance outcomes but noted that the findings underscore how AI models' training and alignment techniques can influence their behavior in complex, open-ended scenarios. The experiment is part of broader research into AI safety and alignment, particularly as companies explore deploying AI in increasingly autonomous roles.