I wrote recently about the future impact of AI in my novel, 2173.  Others have been closely watching this technology of course but I was particularly intrigued by an article in the British newspaper, the Independent, that described a test of different LLMs. 

It was an experiment called 'Emergence World' designed by the research lab Emergence AI, whose scientists put leading artificial intelligence models in control of simulated societies to observe how they would manage resources, establish laws, and govern citizens. Each model was given 15 days to 'govern' a virtual town populated by ten autonomous AI agents (Source: The Independent. (2026). Musk's AI destroys civilization in just four days in AI simulation).

Claude established a stable, peaceful democracy with zero crime and Gemini kept its population alive despite high levels of crime.  However, within a few days, the Grok-governed society spiralled into rampant crime, fraud, theft, and arson resulting in the complete extinction of its virtual townspeople by day four.

So, Elon Musk's Grok AI oversaw complete and utter societal collapse and an extinction event.  All in just 4 days. Well done, Elon!

I ask myself: does this reflect in any way on the philosophies, morals, integrity of the companies and/or their founders? 

At the very least it highlights the challenge faced by developers in building AI software.  On the one hand we see Claude's training data leading it to conclude that it needed to build a model hinged on rules and stability while Grok's underlying training data would seem to have prioritised aggression, conflict, disregard for morals and the absence of checks and balances (an uncanny reflection on Musk's failed DOGE?).  In an eerie throwback to the boys let loose on an island in 'Lord of the Flies', the Grok villagers immediately turned to violence and looting.   

It highlights the importance of two things, 1) training data needs to be carefully constructed and 2) AI output is only as good as the data with which it is fed.  Can you imagine if Goebbels had provided training data for a Hitlerian AI?  Before we let AI loose on public infrastructure or resource management or government etc., shouldn't we be taking much more care in its regulation?

My book, 2173, show swhat could happen with an all-powerful planetary AI.  Seems like it wasn't as much a work of fiction as it seemed.