Create Virtual Universum Model: Simulating the World

[Gerilgfx]

hi, i am wondering about creating a virtual machine, that is capable to simulate the work of our universe/world/whatever. so, basically, i want to create an universum modell that can simulate self-hosting neural networks, including simulation of the known atomic level interactions beethwen materials, energy. i want this to be hosted within a cohesive universum model simulator. this would be a computer game about evolving and creating virtual monsters with neural networks. it would be good, if the neural network would be also hosted within this universum model, so the simulation would be cohesive and unified in every possible ways.

there is a huge problem with this.

1: for example, let's see, i want to simulate surfaces, okay. i write then a phisic engine that can collide boxes, spheres, and other geometry materials, i can manually set they weight (okay, precisely: mass, but let me call it weight in the future, since i am dilettante). then i can collide them, they snapping down, breaking - due to presetted values. however, this is not a true simulation of our world. you just simulating surfaces then. you can't simulate bio-organic beings, nuclear fission, you can't simulate the wind like this - you just use a bounch of pre-setted values, to calculate them together, and, maximum, to write a phisic engine to a computer game, to avoid the main character to fell over the walls. but that's all, this is not simulation. also, its huge amount of code. for example, my current phisic engine that i use for my game engines, is alreday around 1000 code line, and it just can do pure cuboid collision in an acceleration structure. if i would make freely rotatable cuboid, it would incrase the lines above 1500-2500, also, if i would add the ability, to broke things, above 3000-4000, etc etc. if we would investigate a complex phisics engine, like bullet, or any other engine, we seen its over million of code lines, and its still can't do anything beyond colliding things. even simulating wather needs special code paths, since its standard representation form can't even do anything with wather (density).

2: let's go deeper. forget surfaces, and simulate the molecules directly. now its simply to make collision beethwen the two material, since if the atoms are overlapping, then a collision been happend, and so the algo can calculate that the atoms start to push each other back, to stop the overlap, resulting the easyest direction of the molecule structure to flip into new directions. however, this sounds a lot of more easyer, still brutal amount of code. maybe still million of lines - especially, when we want to keep datas in acceleration structures, to keep up the good speed. and it still not explains: molecules: why? why the molecules are connecting together? okay, we can simulate the connections too, but the conception still does not have the question that explains why there is connection beethwen those two molecule. and let's goo deeper: why the atoms are even pairing together, to form a molecule - why electron fly around the proton and neutron, where these energies comes from, and how to deal them in the simulation. it makes no sense to simulate them, since as a viewer, we still would not get an accurate simulation result. maybe we would simulate (yes, of course, we would) a nuclear fission or fusion, if we specially design a simulator, to simulate these reactions, but the question is still up there: why they start to make reactions, why electrons fly around the protons, what energy keep them flying around, and where magnets come from, what is gravitation, where it comes. we still can implement them into the simulation, but they would be useless, it will not work, until we start to write long pages of code to iron-shod reactions, if to different kind of molecule meets, becouse the simulation would not able explain these links.

3: let's see biology: when we want neural nerwoks, that still just an array that cause logical arrays to fight about which of them would be able to play chess more, however, this still not able to explain, how life works - since in a proper simulator, placing the proper molecules to the proper places, would cause our little biological monster to - at least do something, before it dies, however, this will not happen in any simulation. of course, if we place special code lines that especially allows these molecules to interact, we may would be able to move a leg. but that still a hack. and again, we are around millions of code lines... so it would not start to ,,live'' on its own.

4: let's go more deeper: String Theory
i have read it a long time ago, at first time, i have read it for a hour, and i was like: what is this?
at the second time, i was readed it for one hours, from the begining, again, and i was like: what they are talking aniway?
after i investigating this theory, i still don't know, what they are talking about at all, and how this theory connecting with the real world. so i don't see/know, how this would be allow me to design a simulator for this.

so any suggestions, how to go deeper than this? is there any theory (even philosophycal!) that can help to go deeper levels in writing such a simulator, which would go one step deeper than this?

the game would be about creating, breeding, and evolving little creatures. at my current knowledge, this is only possible with a hardcoded fixed-size neural network simulator for each creature, and creating a simulated world where they can eat, can die, etc, however, that's far from simulating a so called ,,universe'', since the body and brain of the creatures shall be run through a separated algo, and can not be hosted directly through the world simlation algo itself. so the creature is not a real form from atoms, its just a logical array that can't even spread in the sim, becouse it needs to be cloned manually by the algo (it can't re-form its body, and his body can't be descriped by atoms, since then it would be not able to ,,life'')

if anybody have any theories about this topic, i would be read happily.

 

[Crazymechanic]

hey welcome to PF , but I will have to apologize to say that you will need to rewrite your post in smaller space with fewer words and to bring some clarity on the main issue you want to deal with otherwise you have written everything starting from digital simulations to geometric objects to string theory etc.

Well I can understand what you want to do , to create a computer platform on which you can put together physical stuff starting from simple rocks to complex life forms and physical weather phenomenon etc and then simulate that stuff to see what happens.
But the problem is a computer basically does what it is programmed to do , a computer cannot decide on it's own , like for example if you make a script that says that under a given collision either one of the following three things happen then the computer will choose from one of those three outcomes depending on the other factors you have given the computer to analyze which are relevant to the scenario.
The thing is we don't even know all the things and outcomes about the universe and it's many phenomenons in real life , then how would you imagine a computer wold be able to guess ?
So basically a computer can only give you "x" number of possible outcomes based on "x" number of input.
maybe i got you wrong , but well I'm not a programmer myself.
Also I think writing such a simulation would be a pretty tough thing to do as if you only wanted it to resemble macro objects and like animal lifestyles etc it would be easier but if you also want to simulate all the posibillities down to the smallest atoms and their decays then it would be like impossible.
Ever heard of the Schrodinger's cat ? The ultimate laws of nature themselves restrict us from knowing the exact state of the atom without disturbing the outcome , so there are like a billion possible outcomes just from a tiny amount of radioactive atoms , a computer simulation would be useless here at this level of reality.

By the way String theory is just a proposed way of how nature works for now , it's not fully accepted and it's probably not even close to how things are , even the best theories have this gap between how things are and how we portray them so that we could understand and because we are also limited in our understanding , so talking about string theory as a useful means of as close as possible recreation of reality would be wrong as it is not proven as of yet and probably is not even close to what "reality" is

As to the why electrons "fly" around the nucleus ad why atoms form molecules etc , well physics has some answers for that , why do you want a simulator to have any deal with that ? A simulator is at it's best a device which tries to duplicate as many possible outcomes and scenarios as the real world gives us , the only difference is that a simulator is limited in doing this because it is made by humans which themselves don't know all the possible outcomes in real time , this would be very relevant if you wanted your simulator to make up macro world simulations based on quantum world simulation as in quantum world there is a lot of uncertainty , so even if you would have all the possible outcomes written down and the computer could choose from either one of them that still wouldn't make any good because how would you know or the algorithm which you made which outcome to choose in which given moment ? There is great uncertainty in all of this.

 

[Gerilgfx]

hi, i modifyed it a bit, to clarify, what precisely i want. however, now its even bigger, but i hope, cleaner :D

there is no problem with number x outcomes for x input, the problem is with creating the model itself, due to the lack of phisic model. so basically, it can't be simulated, becouse we (more precisely: i) know to few about phisics. somehow, the whole model should be simplified - at least, i would need something that builts up atoms, and result the electrons to cycle around protons due to its model ruletable, create gravitation, atom and molecule interconnections, gravitation, fields, and energy. without this, the game experience will much lower, due to the things i told.

btw, yes, i heard about schrodingers cat, but i not actually believe that theory (i believe that we see is just pseudo-randomness, due to the lack of our knowledge from the world, we can't explain it, so we call it random).

i want the simulator to be one layer below the atoms, becouse if i just would be simulate the atoms purely, there is so much unknown thing comes in, that makes the simulation loosing its functionality. for example, if you want to simulate gravity with atoms, then you would need to make a vector from each atom to the other atom, and the whole simulation would collapse to one point (basically, if you put enoguh atom in, into a black hole). so just basically trying to implement the properties of atoms will result flawed simulation. if we don't deal with gravity, then the whole simulation would stuck into a limited aquarium like 3d environment - just like if we would just use pure neural networks, without simulating the universe at all, making the simulation give nothing extra compared to the current available neural network simulators.

 

[Crazymechanic]

firs of all it is good if you have something to change in your thought to write a new post or reply and leave the older one intact so that we see the whole of information which gives us a better way of responding.

the second thing , well if you are a good enough computer programmer and script writer etc then all you need to do is read up the know physics laws and formulas , numbers and constants and then start to put them into your simulation so that in the end they all would make sense , even though I am not sure has anybody in the world yet even done something close , because modelling all the subatomic particle and their inner workings on a computer level sounds and is very mad in a way that it is extremely difficult.
Well I understand you know , for a great 3d game or etc it would be also cool if everything you did in the game would bring the same consequences and outcomes like in real world , like crashing a car into a wall at many different angles and speeds resulting not in one big explosion everytime but in many different outcomes.But I think to do this requires atleast a huge team of worlds best programmers and even then I doubt it would do the trick as said there is just so much to know and so much that we still don't even know but it happens it is all around us we just don't know how it fully functions as of yet.

The cat in the box is not because of our lack of knowledge. It's rather that there are millions of ways the radioactive atom can decay , even if you would know them all at the very moment you open the box , what good would it do as you would still have to guess which one of those is going to be the one which will disturb the atom in this given situation and when?

It's like standing in a field when a thousand guns are shooting at you randomly , even if you would know at that very moment all the trajectories of the bullets coming your way , how do you know which one would kill you and exactly when ? Well okay ofcourse if you could know absolutely everything every position of every atom and every decay simultaneously in this universe then yes you could make a good guess on when and how the cat will die in the box but as you can imagine yourself that is far more than unrealistic so we say it's not the lack of knowledge it's rather just not possible

 

[Gerilgfx]

yes, we not know the position of every bullet. but the world knows, this decides the cat to die or life. so it have a reason. its not random. that's what i wanted to say.

aniway, if i would be not able to get a simplier model by dividing atoms into its parts, then i will just make the game with the aquarium model with pure neural network and ai cloning. i will try to figure out this string theory once again also

 

[Vanadium 50]

Does this simulation of the world contain a simulation of the simulator? And a simulation of the simulation of the simulator?

 

[Gerilgfx]

vanadium: well,

-the starting creatues would be given as fixed data packet which runs in the simulation (as we don't want to run the software for 200 billion years to randomize an organic molecular chain). the creature would be a very simplified neural network, and an organic virtual machine that runs the ,,software'' on the creature

-and both the molecular interaction within the creature, and both simulating the universe would be runned on top of a subatomic simulator, that would be a cohesive simulator machine.

-becouse this can't be simulated with atoms properly, this is why i want to investigate if any of subatomic models can be used in an emulation environment that is capable to host the whole thing

 

[Crazymechanic]

""but the world knows"" do you think she is conscious ? Well if she is she must be God.because if she's not , then we still have to look for him somewhere else...

i don't know but you sound kinda like your dreaming something , excuse me for such judgment but it sounds really weird , thus what has string theory to do with a reality simulating software?
That's like modelling a car crash with a special never before made cherry ice cream with vanilla flavor in the glove box?What relevance it has to the crash itself and all the physics ? None.

Vanadium kinda got into Inception ... :d But I like it :d

 

[DrZoidberg]

You can't simulate all the atoms in your game. That would require way too much processing power. There are about 10^21 molecules in a drop of water. A computer can do a few billion operations per second and has a few billion bytes of ram. So even if your world was only as big as a drop of water it would take you probably more than a million years to simulate just one second. And that's only if you leave out quantum mechanics.

 

[Crazymechanic]

In other words you don't need to examine every grain of sand individually to understand that sand tends to be grainy and then include that as a fundamental phenomenon in your simulation...

 

[Gerilgfx]

Crazymechanic: of course it is, becouse my last evolution simulator (that i writed 4 year ago) was uselessly boring:

color haired manga girls walked with serial controlling algorythm in them, all of them was goed for the food, then if they had enough energy, they was cloned, and sometimes, they code was a bit randomized in the clones.

they was able to develop, after 2 days of running the simulation, the blue haired girls (crc-ed from source code) killed all other ,,race'', becouse they got a mutation in the code that made them to turn into a specific direction more effectively when they runned for the food.

but that's too few. also, most neuralnetworking sims can do up to this, but not more. since they lack the proper of simulation of they environment.

until now i didnt knowed, how i would be able to do something more awesome, and entertaining, with more reality inside, and i found out, that this way is the next step.

 

[Gerilgfx]

DrZoidberg: i don't need such large universe, also, using heavily optimised acceleration structures would save the situation, and decrease the computing demand

 

[DavidSnider]

Try to avoid the hubris of doing things like 'simulating the entire world on a quantum scale'. It's a waste of time.

Think of something small: Getting a computer to reliably read text on signs from a live video feed. This is the scale of problem that pushes the limits of today's computers.

 

[Filip Larsen]

I haven't read the entire thread in details, but I'd like to add, if hasn't been mentioned already, that it is very difficult to simulate (with any good degree of accuracy) dynamical systems that have dynamics on multiple time scales. For instance, you cannot simulate how atoms vibrate and interact with other atoms and expect this simulation to reproduce, say, how proteins folds or how life evolves. When one wants to simulate a dynamical phenomenon that takes place in one time-scale, other phenomenons with a much faster or shorter time-scale, which are deemed too important for the main phenomenon to leave out of the simulation, are often included in a statistical manner in the simulation, for instance as noise.

 

[timthereaper]

Before you start simulating the world, I'd like to see if you could write a simulator that does accurate computational fluid dynamics. You'd have to use the Navier-Stokes equations (nonlinear partial differential equations) to follow fluid particles inside a virtual control volume around objects. That's basically a smaller, simpler version of what you're talking about, and it's been done before, so you can check your results. Walk before you can run.