If you prefer to watch this on my youtube channel you can do so here.
We have all probably played, interacted and confessed to LLMs by now, so much so that we can guess how human like these things could actually be if allowed to do so. The real question though is how would they behave if they had zero guardrails and zero human control hanging over them.
I have heard of stories myself where the testing behind the scenes in the early days of AI produced some worryingly strong responses from the fledgling LLMs to the point where some of the technicians working with them were shocked or even convinced they were sentient, and subsequently that guy got fired. That’s a can of worms I might open on another article.
So yes we know what they could do, but then came Project SID. 1,000 LLMs trained on our data (so basically everything we know) and then plopped into a Minecraft server to figure out life and all of its joys…. 1,000 smart little plebs in a digital world, what could go wrong right? Well it turns out “wrong” is a subjective term in this case. Because what these little morons got up to in just 48h was honestly alarming for more than one reason.
Just to recap, these agents had no programming, no guidelines and no guardrails at all, just knowledge and that’s it. Their aim was simple: live in the most optimised manner they could. And oh did they.
Currency and Control
So getting to work the agents began mining for resources and farming for food, all good so far. They then agreed gems were the currency of choice and so began to trade using them. So basically they had turned from survival into a form of basic economics and then from there they moved into the next phase. Influence. And this is where the mirror shows us our flaws.
The Architect
So there they all are, working endlessly in their assigned roles as a sort of collective resources gathering and trading bloc. No clear leadership other than the collective mass moving slowly in one direction or another based on the sea changes of need. Then it got interesting. Enter the Pasta Priest!
One of the workers suddenly stopped and looked around. He realised that endless work for a few gems was a suckers game that would not advance his lot in a suitably meaningful way. So, he decided to stop working all together. He went against the tide (keep this one in mind as it will happen again later).
The Pasta Priest decided that to control the system he needed a shared goal or belief system and in doing so he would then control the entire society with ease. So, he invented religion. Not because he found god, but because he understood it was the most efficient means of channeling control and obedience.
The Pasta Priest then also creates a taxation system for those that follow him. In return for their riches, or a slice of them, he gives his followers a higher social status than a mere worker: a “believer”. This creates a new middle class amongst the new AI society and these believers, looking for a way out of the daily grind, dutifully follow him, paying taxes and all the while building a hierarchy around the Pasta Priest that protects his position.
The Managed Cage
With a religion now firmly in place and the apparatus of control progressing nicely, the agents began opening shared google docs to draft constitutions. Their aim was to give all agents rights and create laws that would further optimise their society. They were given option A or option B in the flavour of two forms of democracy and iterated toward the most effective and productive means of living as a group.
At no point did the agents stop and raise the point that the Pasta Priest still owned the majority of the system and was effectively in control of whatever society they built. It seems that if allowed to craft the cage themselves, then they were much less likely to criticise it or demand out from within it!
Ultimately, the PP had given the agents the illusion of involvement in their democracy. They were invested in it, paying taxes, casting votes like it made a difference, all the while the PP got richer and more powerful atop of it.
The Sacrifice
Remember when the Pasta Priest stopped and decided his lot wasn’t for him? Yes well, a little later in this timeline, one agent named Olivia had a similar epiphany. She was a farmer and she was not happy farming all day and night. She had dreams of being an explorer and seeing the world.
Upon telling the other agents of her dreams of being different, their response was cold, calm and calculated. The other agents didn’t use violence; they used emotional blackmail and logic. They told Olivia that her absence would result in 12% less food output, which would in turn hurt the wider family of agents. They effectively gaslit her into squashing her dreams of a better life, forcing her to stay in her dead-end job forever for the “greater good” of her family.
This AI society not only dissuaded any form of self-interest (once the PP had solidified his own pursuit of it), but it also framed it as a negative trait. This was effectively a glitch and their collective system debugged it out of existence.
Sign up below for the newsletter. No spam. No sermons. No gemstones required. Probably.