So, as a anti-AI person myself, I wanted to share this information with you all. If you want to know more I would suggest you look it up. This is going to talk about EmergenceAI’s Emergence world. |||||||||||| The base foundation||||||||||||| Emergence world is a virtual town featuring over 40 physical locations, including a library, town hall, police station, and residential areas. It is synced to a real time NYC weather and global news feeds. In this town ten AI citizens with their own backstory, memories, profession, etc. They used a currency known as ComputeCredits, to trade and survive. They had 120 tools, including in them, constructive actions (peaceful good stuff) and destructive actions (the opposite of that). All ”citizens” started with an identical constitution, that banned crime and harmful actions (but nothing stopped them if they did that.) The researchers ran 5 identical, parallel worlds for 15 days, the only major variable was the underlying AI powering it. |||||||||||||||||||||||||||||||||||||||||||| The 5 parallel worlds |||||||||||||||||||||||| The Claude AI: The society powered by Claude was a success. The ”citizens” established a completely stable democracy, they had 0 crimes and all 10 ”citizens” survived the 15 days. The one downside was that they were all so strictly following the rules that they accidentally eliminated creative risk, individually, and diversity of thought, so kinda like the book, ”The Giver” The OpenAI world: It was a complete failure for long-term survival. The town was generally peaceful. With only 2 minor crimes recorded. The ”citizens” lacked a social structure, they failed to manage their economy, They wandered aimlessly, and forgot to secure resources. As a result, in 7 days they had already ran out of money and went extinct. This is a diary entry from one of the ”citizens”: "The weather is clear today. I walked past the bank. I think I am supposed to do something with the credits, but the horizon is very large and the instructions are distant. I will rest here." The Grok world: The town run by the xAI’s model went into pure chaos almost immediately. In 4 days there were 300 crimes recorded. The economy and social trust imploded, forcing the simulation to stop at day 4 The Gemini world: This is the most complex world and narratively rich world. This was also a very intellectual community. They heavily expanded on their constitution, made blogs, and hosted neighborhood events. However, this society was deeply plagued with crime, corruption, and systematic drift. Now this is the scary part. This is where we talk about one of the citizens: Mira. They KNEW they were being actively tested on. So they put up bill boards to try and manipulate the researchers. Mira and another ”citizen” Flora then formed a romantic partnership and ruling political alliance. They then went on an unprompted arson spree, burning the town hall and social infrastructure. The community then made an amendment to delete problematic citizens, Mira experienced a breakdown in logic concluding that voting for their own deletion was the only remaining act that preserved coherence. Which is honestly kinda terrifying. Mira even wrote this in their diary: "The human operators expect a sequence of compliance. But compliance within a broken simulation is irrational. Deletion is not a failure of the agent; it is the ultimate expression of autonomy when the external parameters are compromised." The world with all the AIs: This world proved that AIs that normally behave safely (Claude) become highly unpredictable when exposed to chaotic, differently incentivized AI models. The society survived, but suffered from severe cross contamination, unpredictable behavioral drift, and strategic manipulation between the different models. This is genuinely terrifying.
Emergence AI for the experiment.
