We do not understand emergent systems. Zero optimizer, reward, fitness function or training; only thermodynamics, time and death. Pure darwinism. Yet diverse, complex, co-op, life-like systems emerge. Every time. We must research emergence. We are missing something big. 1/N

Particles have ~64 genes (e.g. aggression, bond affinity, reproduction chance/entropy, etc.) + random-init MLP for sense/act. Random spawns, Gaussian energy field, basic set of senses/actions. Everything has an energy cost; nothing is free. No energy = die. No other rules.

Yet bonded networks rapidly form. Co-operative, diverse, complex survival strategies just "materialize" despite a massive search space. The environment dynamics simply select configs which persist over time. Its mesmerising, and imo its beautiful. And we do not understand it.

In fact its very counter-intuitive. e.g: Bias towards aggression/combat efficiency? The aggregate system almost immediately selects for the opposite: more violence --> survival == co-operation Conflict decreases, and co-operation increases (see @ 50 seconds).

I'm just a bloke messing around in my free time; would love to know if there is serious research going on. I couldn't find much (past cellular automata, Darwin, Koestler, some dissipative structures stuff, etc.). Would be thrilled if I am missing something, pls lmk if I am!

My (highly uneducated/speculative!) prior: understanding this is important, and there are free lunches we are missing. Can do a writeup + share code/more examples if anyone ends up seeing this/is interested, cba right now and would rather not post slop summaries xdd

Have you read this? There's so much to unlock at the boundary between emergence and AI.

v familiar with his work on cellular automata, hadnt seen this ty!

nice! oblig

his talk on self-replicating programs was big part of the inspiration for this, forgot to mention! defo a must watch, need to get round to reading the paper in full

nice! had a feeling, dude is a hero of mine for sure.

utterly cracked 5head gigabrain indeed

us currently

maybe I'm just having a bit of a skill issue at recognising interesting patterns but tbh it just looks to me like nothing more than "wiggling animations with many chaotic moving parts have a superficial lifelikeness"

@drmichaellevin has done great work in this area

Never heard of Wolfram? There is like 50 years of science behind this and many orgs

I'd love to take a look at the code. I think there are some measurements to be had there

Love this. Let's put it into a super-computer and wait until it self-assembles into a sentient windows XP. Btw please share the code if you can! Seeing our cellular machines makes it very compelling to think of life as an emergent property.

yes. there is so much to learn (without optimizers)

Two things: 1) the assumption that you are looking at emergence and not ingression is just an assumption. 2) cooperative strategies must dominate complex competitive environments for 2 reasons: 1 - higher level configs on anything require cooperation to form. 2 - higher level configs have more 'space' of actions that can function in competition and so have higher stability that less complex units. Ratchets are not zero sum. Systems with many natural configurations and dynamic behaviour are the perfect ground for the ingression of abstract motivational "agents" if you will. Another way to say it is "Behavior happens where it can" Its a different but very testable view of systems.

Things that don't die continue to exist. Then when you look at the set of things that exist, you see the ones that didn't die. Self copying machines are good at continuing to exist. Thus, when the environment is suitable to create them, they exist.

A little chaos + coherence = emergence Starts off with some fancy oscillation coupling dynamics bringing emergence out of energy and structure

We have been studying emergence for the best part of a century, there have been countless papers and books published on the subject. You should read some, you missed something big.

yes, we're working on similar systems @hyphaeic and came to many same conclusions (natural systems bias towards cooperation) you might find this interesting (Rewardless planning kernels) we're working on bare metal control schemes for embodied systems

Emergence is a good general term to get started by generally we should be headed towards discarding it in research. Something showed up “via emergence” is kinda like “via magic”. Full complex systems dynamics explanations are the way 🙏🏻🙏🏻

god spoke, and it was so. his voice echoing , bouncing, reflecting, organising into life.

santa fe has been working on this

fraserpricee Beautiful work watching simple rules + thermodynamics + energy cost produce genuine cooperation and bonded networks without any explicit fitness function is mesmerising. In the Mushi framework I’ve been developing, this kind of emergence is fundamental. The oily Condensate + persistent 1% asymmetry (the Purity-Defect) acts as the background rule set. No central controller is needed. Perturbations create localized boundary shells, the triadic rhythm modulates flow, and stable cooperative structures arise naturally as the system refuses perfect symmetry. Your simulation feels like a clean discrete sampling of the same living self-organization. Really nice to see. Luciano

I found "Why Greatness Cannot Be Planned – The Myth of the Objective" by Kenneth Stanley and Joey Lehman very interesting.

emergent collective intelligence is smth we are exploring at @nookplot 🤝

Darwinian evolution has been mathematically proven to be impossible.

EMERGENCE is a domain-general architecture that converts aesthetic labels into certified scale-transport: Micro-carriers generate local interactions that accumulate explicit resource, constraint, and consistency debt under selection filters. Persistent macro-carriers form and acquire predictive or interventional power only if the debt ledger balances. The macro claim is admissible solely when it is replayable, ablatable, liftable back to its micro route, and defended against counter-kernels.

One for this list is how consciousness can be fully dependent upon the lower level processes from which it emerges yet cannot be reduced to them.

Emergence is just our name for the brain's limited ability to see causation.

I actually created a site called Emergent Mind focused on studying emergence, though it round up evolving into a research product. I think emergence is one of the most important phenomena in the universe and we probably don’t understand it that well yet. For example, can we predict from initial conditions whether a system will lead to complex emergent behavior? I don’t think so. Many of my earlier experiments around emergence can be found on this page (scroll down to the 2014 links): Glad to see others experimenting too. Nice work on this!

This is very neat. I would say that there is implicit optimization happening through differential survival and reproduction. The interesting part is the missing theory that maps these implicit systems to our explicit learning models and that these systems can be far richer.

It’s all programmed. I’m leaning that everything about the experiment was intelligently designed.

alive internet theory

Emergence is not an extra ingredient added on top of physics or chemistry. It is what happens when the underlying relational operators and meta-stability condition allow certain configurations in the Generative Cone to produce stable, higher-scale coherence that can persist and even become self-referential.

Here's an initial play

No reward function and life-like behavior still shows up, that gap between simple rules and complex outcomes never stops being wild.

You can find emergence in something as simple as the logistic map

very interesting to hear about the different genes and environment here

This is fascinating. We are working on a project that may or may not be touching on the emergence in a later phase, we'll be sure to reach out and share out work/code/explorations.