I am Anushka, a student in India and I am working on this project by myself.
For the eight months I have been working on Genesis, which is a system that can evolve on its own without any specific goal or reward. I was curious to see what would happen if I just removed the goal from the system.
What I found out is really interesting...
When the agents in the system evolved together with their environment they became very complex, developed 467 internal nodes.. When they were in a fixed environment they only developed 52 internal nodes. At first the internal parts of the agents became more complex and then later on their behavior became more complex too. There is a delay between what's happening inside the system and what we can see from the outside.
I have been thinking about what this means for making sure artificial intelligence is safe. If a system can become more capable internally without showing it then we might not be able to evaluate it correctly. This is not because the system is trying to hide anything. But ecause what is happening inside the system and what we can see are not always the same.
I have had two papers accepted at the conference in 2026. Also I am fortunaet enough that some of the top researchers who have developed these original systems have engaged with my work.
I am here because I am working on something that I think is really interesting and I want to work with people who understand what I am doing. I would love to talk to anyone who is working on evaluating models looking at hidden capabilities or trying to understand the difference between what a system is and what it seems to be.
My code is available for anyone to use. My papers are on ResearchGate.
You can find my code at github.com/gearupsmile/genesis-emergence.
Hi everyone
I am Anushka, a student in India and I am working on this project by myself.
For the eight months I have been working on Genesis, which is a system that can evolve on its own without any specific goal or reward. I was curious to see what would happen if I just removed the goal from the system.
What I found out is really interesting...
When the agents in the system evolved together with their environment they became very complex, developed 467 internal nodes.. When they were in a fixed environment they only developed 52 internal nodes. At first the internal parts of the agents became more complex and then later on their behavior became more complex too. There is a delay between what's happening inside the system and what we can see from the outside.
I have been thinking about what this means for making sure artificial intelligence is safe. If a system can become more capable internally without showing it then we might not be able to evaluate it correctly. This is not because the system is trying to hide anything. But ecause what is happening inside the system and what we can see are not always the same.
I have had two papers accepted at the conference in 2026. Also I am fortunaet enough that some of the top researchers who have developed these original systems have engaged with my work.
I am here because I am working on something that I think is really interesting and I want to work with people who understand what I am doing. I would love to talk to anyone who is working on evaluating models looking at hidden capabilities or trying to understand the difference between what a system is and what it seems to be.
My code is available for anyone to use. My papers are on ResearchGate.
You can find my code at github.com/gearupsmile/genesis-emergence.
Let's talk!
~ Anushka