It's so amazing that I feel like I'm going to faint...Also, I was really surprised at how much this Lenia resembles a horseshoe crab...This is a game changer.
Will reproduction be limited? Isn't it the case that the simulation lacks entropy? A successful organism could just infinitely reproduce if I'm not wrong.
If lenia, like in game of life, has a defined set of rules for each next state then how can these lenia 'learn' if all of it is deterministic? Is it the very rule that is being changed in the processing of learning?
The rule doesn't change. Instead, the pattern changes. In Conway Life you've seen "gliders". Think of these patterns as a much bigger glider pattern and then a machine-learning algo writes more and more capable patterns. So this isn't self-evolving code; it's two different halves. One code that looks at, and keeps evolving, the other and it self. Basically the computer is like "OK my job is to design a hand-written pattern that can make its way through mazes. Oh, they're all failing. OK, let me evolve myself so that I get better at hand-writing such patterns." In real life, there is DNA inside the organisms, and the algorithm (probably better known as "living") for the organism's own interaction is the same as their own evolution. That's not what this is. Instead, the designer is evolving itself. It's like if people were hand sculpted by some sorta Pygmalion God and he'd be like "ah man, these scrubs are hanging out the passenger side of their best friend's ride trying to holla at me" (Gen. 11:1-9) and decides to go do some reps so he'll get way better at hand sculpting people and then rinse and repeat until he had created some sorta super Galatea that can make its way through an Algernon maze and dodge bullets like Neo.
If you could allow them to asexually reproduce based on available resources, and with reproduction allow extremely minute mutations to arise, while giving them a more complex environment to survive in, this might work as a type of evolution simulator capable of creating actual digital life, where successful mutations are stored in code rather than in DNA. It would never be as complex as actual life, but it would be extremely fascinating more so than it already is.
Well if you add more layers you give more chances for complexity, but also more computation needed for everything, "wire world" automata has 1 layer with 4 states(empty, head, tail, wire) and can simulate most computer logic a second layer would be able to simulate magnetic interaction, "sand castle" simulations have a lot more states and complexity, but they still lack opaque (-sy) and can't signal to cells farther away because you need the cells occupied by something to transmit the information, so light through glass becomes difficult(having 2-3 times more states is a solution ,but the rule set becomes 9! times bigger every time you add one state), because of the rules of the game, but if you add a light layer the problem becomes easier, but more processor intensive because now you update 2 grids instead of 1. So i'm arguing that you can get up to infinite complexity, but A) after some point you tend to get it to look more homogeneous(like static noise) not less and B) it gets exponentially processor intensive to simulate all the interactions between layers and updating them. And C) Conway's Game of life already has the necessary complexity, you just need an infinite grid to simulate you and me life since it(CGL) is Turing complete (and a monitor to view the binary output as a 2.5D view of reality as seen through your eyes)
Hi. I have a little question to Lenia. I recently programmed my own little version of Lenia but now I run into some problems. I wan't to scale my creation in resolution. I found it with a radius of 14 that's why the result is very pixelated. How would I change the resolution so it looks smoother?
Some people have used cellular automata to generate procedurally maps for games (and so on). Do you think it is possible to use lenia for this application and would there be any benefit?
@@bertchan9774 That's exciting. Is there a way to follow your work other than your RUclips account (or perhaps is there a way to help you)? I would like to know more about novel celluar automata techniques. I think this field is highly ignored by biologists and physicists. Building better simulations can help us learn more about the "emergence" of anything, even the intelligence. I hypothesize that, in our journey to the AGI, there are two crucial stepping stones : 1- advances in semiconductor technology 2- better simulations First one is needed to speed up the second, and second one is needed to accelerate our "deduction" as a scientific tool. It will be possible to design specialized virtual environments for examining the emergence of self-sustaining agents, and play with them to better analyze their properties / functions. Just like what you did in your project, but in much bigger scale and in more diverse environments. For example, environments that would stir up the emergence of agents with complex situational awareness mechanisms. I think running better (and faster) simulations would influence the AI field more than the advances in neuroscience. Given the new trend of unsupervised models in ML research (xcorr.net/2021/12/31/2021-in-review-unsupervised-brain-models), this became even more clear to me. (Or it is just my misinterpretion of the concepts, haha.) Perhaps I should stop here. I couldn't help myself from writing more and more. This is one those topics that tingles ones brain starting from the first time hearing about it. Keep up the great work.
@@brettliddell3011 I tried to simulate protons, electrons and neutrons, and just 20 particles at once slow everything down and I dindnt even implement all forces yet. Simulating individual atoms of a large molecule would take ages
@@juergenkern6763 I think we can synthesize atoms via a process akin to glider synthesis in the classic game of life. I think the neutrino, which oscillates between three or four flavours in flight is our universe’s glider (if it is a discrete 3D analogue of the game of life) or orbium. For a rule that models our universe, I think that all we need is one where gliders, spaceships, orbium etc simply stick together when collided to produce larger patterns that can also move across space, rather than stay still or fade out. That’s all the rule has to do, just for arguments sake forget about the classical or quantum pictures of the atom where you need electrons orbiting or teleporting in and out of existence around the nucleus. What if it isn’t protons, neutrons, and electrons but it’s just a single pattern (either digital or continuous - in the case of Lenia). Then a molecule like ATP wouldn’t consist of hundreds of vibrations in quark, gluon, photon, and electron fields all interacting with each other in perfect harmony while moving through space, which would essentially be required since there are 1 billion ATP molecules in each of our cells at a given time, and we constantly need to regenerate it into ADP (like doing an operation with bits over and over in a computer and never having the results differ). We will never simulate such a complex molecule moving through space with the current quantum paradigm (so how does nature do it), let alone simulating the collision of a molecule with another molecule to result in a chemical reaction (without heavily customizing the simulation for that specific scenario). Whereas in life like rules, patterns regularly stick together, and although I haven’t seen it, maybe one rule will have patterns that also readily fall apart back into their constituent patterns upon certain collisions - and I think that’s all you need to simulate chemistry. Try to synthesize hydrogen via glider synthesis, then use that to synthesize helium, continue to carbon etc and simply collide those patterns to make molecules - you’ll know you got it right when the reagents react the way you’d expect. You wouldn’t need to worry about electrons because the idea of the electron cloud was perhaps our best interpretation of what’s really there: patterns in a continuous or discrete 3D cellular automaton. I don’t have much time these days but if I did, I’d be trying to use machine learning to find a rule where patterns stick together (rather than fade out) upon colliding - which is a rare enough property. These were some of my results for the discrete 3D game of life (see Carter Bays papers): tinyurl.com/particlecollisions You might note that depending on the slight differences in starting positions of the particles/gliders, different outcomes occur. So it’s not quantum probabilities, it’s just subtle differences in starting conditions, and the system is deterministic underneath. Anyway pardon the rambling but keep on trying to simulate physics with CA!
This is amazing!!! I feel it very promising, but I dont know where exactly.. Maybe some optimisation algorithms, search strategy of the optimum solution in hyperspace..
A more indepth explanation of what is going on would have been great. Like this is sounds interesting but it's hard to understand what is actually going on.
I am a software developer in many lenguages. Java c# rust ecc.. I studied game of life. I discover lenia tonight. Where I should start? Which lengueage and softwere are used? ❤️
If you want to learn how to use intelligent AI to automatically generate code for you in numerous programming languages, watch our mind-blowing video on our RUclips channel.
It is crazy how much this looks like life - proving perhaps that life is simply the playing out of math behind the curtains. Brilliant work, truly.
It is crazy for someone who has zero intellect and knowledge of the actual world.
@@aleksitjvladica. I assume you're talking about yourself?
Why are you pinging? You do not assume anything, you provoke.
@@aleksitjvladica. Just seemed fitting since you were needlessly rude
@@aleksitjvladica. dude why are you assaulting random people ? this guy was just spitting facts.
Je suis venu ici après avoir regardé la vidéo de science étonnante
It's so amazing that I feel like I'm going to faint...Also, I was really surprised at how much this Lenia resembles a horseshoe crab...This is a game changer.
idk why but when you said "can lenia patterns dodge bullets" I started laughing so hard
and the best part was they totally didn't, the bullets went straight through them lol
"Can Lenia patterns dance for two?"
This is a very interesting concept. I wonder how the cells would perform if they were trained in an adversarial environment.
Will reproduction be limited? Isn't it the case that the simulation lacks entropy? A successful organism could just infinitely reproduce if I'm not wrong.
Would be interesting to add aging and to require material for offspring
Excellent! Thank you for your work, for this video and also for the blog post with tons of info and interactive content!
If lenia, like in game of life, has a defined set of rules for each next state then how can these lenia 'learn' if all of it is deterministic? Is it the very rule that is being changed in the processing of learning?
The rule doesn't change. Instead, the pattern changes. In Conway Life you've seen "gliders". Think of these patterns as a much bigger glider pattern and then a machine-learning algo writes more and more capable patterns. So this isn't self-evolving code; it's two different halves. One code that looks at, and keeps evolving, the other and it self.
Basically the computer is like "OK my job is to design a hand-written pattern that can make its way through mazes. Oh, they're all failing. OK, let me evolve myself so that I get better at hand-writing such patterns."
In real life, there is DNA inside the organisms, and the algorithm (probably better known as "living") for the organism's own interaction is the same as their own evolution. That's not what this is. Instead, the designer is evolving itself.
It's like if people were hand sculpted by some sorta Pygmalion God and he'd be like "ah man, these scrubs are hanging out the passenger side of their best friend's ride trying to holla at me" (Gen. 11:1-9) and decides to go do some reps so he'll get way better at hand sculpting people and then rinse and repeat until he had created some sorta super Galatea that can make its way through an Algernon maze and dodge bullets like Neo.
@@sandcjw9901 oh I see! Now it makes sense, thanks a ton for going through all the trouble of explaining :) really helped
If you could allow them to asexually reproduce based on available resources, and with reproduction allow extremely minute mutations to arise, while giving them a more complex environment to survive in, this might work as a type of evolution simulator capable of creating actual digital life, where successful mutations are stored in code rather than in DNA. It would never be as complex as actual life, but it would be extremely fascinating more so than it already is.
The guy with the vision gets the job. The code for Lenia is freely available.
Well if you add more layers you give more chances for complexity, but also more computation needed for everything, "wire world" automata has 1 layer with 4 states(empty, head, tail, wire) and can simulate most computer logic a second layer would be able to simulate magnetic interaction, "sand castle" simulations have a lot more states and complexity, but they still lack opaque (-sy) and can't signal to cells farther away because you need the cells occupied by something to transmit the information, so light through glass becomes difficult(having 2-3 times more states is a solution ,but the rule set becomes 9! times bigger every time you add one state), because of the rules of the game, but if you add a light layer the problem becomes easier, but more processor intensive because now you update 2 grids instead of 1.
So i'm arguing that you can get up to infinite complexity, but A) after some point you tend to get it to look more homogeneous(like static noise) not less and B) it gets exponentially processor intensive to simulate all the interactions between layers and updating them. And C) Conway's Game of life already has the necessary complexity, you just need an infinite grid to simulate you and me life since it(CGL) is Turing complete (and a monitor to view the binary output as a 2.5D view of reality as seen through your eyes)
Hi. I have a little question to Lenia. I recently programmed my own little version of Lenia but now I run into some problems. I wan't to scale my creation in resolution. I found it with a radius of 14 that's why the result is very pixelated. How would I change the resolution so it looks smoother?
Some people have used cellular automata to generate procedurally maps for games (and so on). Do you think it is possible to use lenia for this application and would there be any benefit?
This is absolutely incredible.
i see lenia more like the fields of quantum mechanics than biologic celulas or bacteria.
This has nothing to do with quantum mechanics?
@@loiman4179 indeed not, but for my personal point of view, 'feels like' heheheheh
Well it has.@@loiman4179
That was beautiful! Thank you! 🥹
When they reproduce, can the offspring "mutate"?
Not for now, but we want to design the system to allow this, so the system can evolve in a natural way.
@@bertchan9774 That's exciting. Is there a way to follow your work other than your RUclips account (or perhaps is there a way to help you)? I would like to know more about novel celluar automata techniques. I think this field is highly ignored by biologists and physicists. Building better simulations can help us learn more about the "emergence" of anything, even the intelligence.
I hypothesize that, in our journey to the AGI, there are two crucial stepping stones :
1- advances in semiconductor technology
2- better simulations
First one is needed to speed up the second, and second one is needed to accelerate our "deduction" as a scientific tool. It will be possible to design specialized virtual environments for examining the emergence of self-sustaining agents, and play with them to better analyze their properties / functions. Just like what you did in your project, but in much bigger scale and in more diverse environments. For example, environments that would stir up the emergence of agents with complex situational awareness mechanisms. I think running better (and faster) simulations would influence the AI field more than the advances in neuroscience. Given the new trend of unsupervised models in ML research (xcorr.net/2021/12/31/2021-in-review-unsupervised-brain-models), this became even more clear to me. (Or it is just my misinterpretion of the concepts, haha.)
Perhaps I should stop here. I couldn't help myself from writing more and more. This is one those topics that tingles ones brain starting from the first time hearing about it. Keep up the great work.
@@bertchan9774ever thought about simulating the evolution of atoms and molecules?
@@brettliddell3011 I tried to simulate protons, electrons and neutrons, and just 20 particles at once slow everything down and I dindnt even implement all forces yet. Simulating individual atoms of a large molecule would take ages
@@juergenkern6763 I think we can synthesize atoms via a process akin to glider synthesis in the classic game of life. I think the neutrino, which oscillates between three or four flavours in flight is our universe’s glider (if it is a discrete 3D analogue of the game of life) or orbium. For a rule that models our universe, I think that all we need is one where gliders, spaceships, orbium etc simply stick together when collided to produce larger patterns that can also move across space, rather than stay still or fade out. That’s all the rule has to do, just for arguments sake forget about the classical or quantum pictures of the atom where you need electrons orbiting or teleporting in and out of existence around the nucleus. What if it isn’t protons, neutrons, and electrons but it’s just a single pattern (either digital or continuous - in the case of Lenia). Then a molecule like ATP wouldn’t consist of hundreds of vibrations in quark, gluon, photon, and electron fields all interacting with each other in perfect harmony while moving through space, which would essentially be required since there are 1 billion ATP molecules in each of our cells at a given time, and we constantly need to regenerate it into ADP (like doing an operation with bits over and over in a computer and never having the results differ). We will never simulate such a complex molecule moving through space with the current quantum paradigm (so how does nature do it), let alone simulating the collision of a molecule with another molecule to result in a chemical reaction (without heavily customizing the simulation for that specific scenario). Whereas in life like rules, patterns regularly stick together, and although I haven’t seen it, maybe one rule will have patterns that also readily fall apart back into their constituent patterns upon certain collisions - and I think that’s all you need to simulate chemistry.
Try to synthesize hydrogen via glider synthesis, then use that to synthesize helium, continue to carbon etc and simply collide those patterns to make molecules - you’ll know you got it right when the reagents react the way you’d expect. You wouldn’t need to worry about electrons because the idea of the electron cloud was perhaps our best interpretation of what’s really there: patterns in a continuous or discrete 3D cellular automaton. I don’t have much time these days but if I did, I’d be trying to use machine learning to find a rule where patterns stick together (rather than fade out) upon colliding - which is a rare enough property. These were some of my results for the discrete 3D game of life (see Carter Bays papers): tinyurl.com/particlecollisions
You might note that depending on the slight differences in starting positions of the particles/gliders, different outcomes occur. So it’s not quantum probabilities, it’s just subtle differences in starting conditions, and the system is deterministic underneath. Anyway pardon the rambling but keep on trying to simulate physics with CA!
Fascinating. Next Level Game of Life. Evolution in differentiable cellular automata.
"Failed cases"
*starts multiplying like crazy*
Can Lenia patterns learn to remember? (Measured by improvement at solving a puzzle or maze after repeated attempts)
This is amazing!!! I feel it very promising, but I dont know where exactly.. Maybe some optimisation algorithms, search strategy of the optimum solution in hyperspace..
Jesse what the fuck are you talking about?
i do not quite see how the maze-example shew more than the inevitable outcome of randomness...
A more indepth explanation of what is going on would have been great. Like this is sounds interesting but it's hard to understand what is actually going on.
you have no idea how much I love lenia, can you teach me how I can implement my own version?
This is incredible. How has this been out for four months and have under 2k views!?
Wait, are you saying the behavior of the organism is coming from the positioning and values of the individual cells of the organism?
Is there one with conservation of energy
AMAZING😲
How do you train them?
I am a software developer in many lenguages. Java c# rust ecc.. I studied game of life. I discover lenia tonight. Where I should start? Which lengueage and softwere are used? ❤️
Can lenia patterns conquer the universe? 👀
Where do you download this game?
if we could simulate this way all cells of a human being ?
this is literally a simulation of life by most definitons
OMG ! 🤔The King is naked !
why do they look like horse shoe crabs? or trilobites?
Lenia
优雅而智慧的群
Cool!!
A life?
Wow
That has to be alive
Based and redpilled
wow.
yooooooooooooo
If you want to learn how to use intelligent AI to automatically generate code for you in numerous programming languages, watch our mind-blowing video on our RUclips channel.