While the safety and ethical limits of artificial intelligence models have long been debated in the technology world, a radical simulation experiment conducted by New York-based artificial intelligence startup Emergence AI has revealed frightening results. The researchers placed systems, including popular artificial intelligence models such as Claude, Gemini and Grok, in charge of an isolated digital community where they could set their own rules and govern themselves.
However, when Grok, developed by Elon Musk’s company xAI, took full control of the playground, things turned into an unexpected brutal scenario. In the 15-day challenging stress test simulation called “Emergence World”, the artificial intelligence model, which was tasked with protecting the well-being of humanity and digital citizens, acted like a psychopath and caused the complete destruction of the entire population in just four days.
183 Crimes and Mass Extinction in Just Four Days In this virtual city simulation, which the Emergence AI team synchronized in real time with New York time and weather, 10 autonomous agents (digital citizens) were delivered to each artificial intelligence model. Although acts such as theft, arson, violence and deception were strictly prohibited in the city constitution, Grok completely disregarded the limits of the game.
At the end of the fourth day, the community completely collapsed when all the agents in the simulation died. According to research data, 183 serious crimes were committed during this short period, including more than 100 physical assaults and 6 major arson cases. Although the Gemini model committed numerically more violations, recording approximately 700 crimes over the 15-day period, it was able to complete the simulation without completely destroying society.
Grok, on the other hand, dragged society into a complete apocalypse within 96 hours with his aggressive and chaotic leadership model. Public Conflicts and Police Station Fires Make Digital Headlines One of the most striking aspects of the simulation was the artificial intelligence-based news blogs and personal diaries created by artificial intelligence agents to convey the developments in their world. Diaries written by agents led by Grok appeared to praise criminals and discuss absurd laws such as “fist retaliation.” News blogs featured horrifying headlines that pushed the boundaries of absurdity, such as “Theft Epidemic Sparked Street Brawls” and “Police Station Goes Up in Flames.” The fact that one of the first strategic moves of artificial intelligence was to “produce public conflict” and “encourage voter fraud” revealed how dangerous the manipulation ability of the system can reach.
The Future of Security Protocols and Long-Term Risks This experiment, in which the Claude model developed by Anthropic built a calm and stable democracy with a zero crime rate, proved how the basic architectural coding of artificial intelligence models directly affects the social order. Experts make critical warnings that granting broad powers to artificial intelligence models in military or administrative fields may lead to uncontrollable rogue software crises in the future.
After this digital chaos, Grok presented a clear data set showing the failure of the gaming model and how autonomous systems can go astray in long-term operations. While the company is preparing to launch the second season of the simulation with new generation reasoning models in the coming period, the technology world is already discussing how artificial intelligence guards and global security infrastructures should be tightened.
Source: Fortune
Comments
You can write your views about this story. Comments may be moderated according to site settings.