[breuleux]

> Another interesting thought is if we ever become capable of simulating such a universe, it seems much more likely than not we're also a simulation, since there's a possibility for each simulation to produce it's own simulation or collection of simulations.

There's a lot of "ifs", though, for the argument to work:

The simulations must support a higher density of life than the universe does. If you need a planet the size of Jupiter and a hundred years to simulate one second of life on Earth, then each nested layer of simulation is going to make the number of simulated humans smaller. It seems highly dubious to me that simulations could be dense enough: if you can build a simulator capable of evolving a million life forms, couldn't you slash all the overhead of environmental simulation and use these same resources to build ten million life forms that interface with the real world directly (imagine nano-life, if that helps)?

There must be sound economic reasons to build these simulations. Otherwise, too few of them will exist, and they will not be run for long enough. I can think of a few reasons, like research or entertainment, but I do not think they would suffice. Specialized, focused simulations are useful. Simulating whole universes is overkill. It's a waste of resources.

[simdude]

There is no reason to assume that the same physical rules and limitations of our own universe apply to the universe that is simulating us. Our universe could be a drastic simplification of the "real" one, indeed this simplification is the kind of potential evidence the article mentions us looking for.

I remember reading once that the limitation on speed of light could be a performance optimisation. Isolating each "light cone" of local spacetime allows the simulation to be distributed and run on separate compute clusters, because it limits things to local interactions rather than universal ones.

Things in superpositions only becoming "set" after observation... that is just lazy evaluation. It's fun to think about.

[nickcano]

>I remember reading once that the limitation on speed of light could be a performance optimisation.

>Things in superpositions only becoming "set" after observation... that is just lazy evaluation.

I've always thought about it this way also, I'd love to get a source on where you read it so I can see a different take than my own.

Some other things to consider:

1. Planck length is the most granular unit or pixel the computer can measure.

2. The time it takes to move one Planck length at the speed of light is the time of one iteration of the simulation's "main loop".

3. The reason time dilates as you approach the speed of light is because the faster a particle moves, the more the main loop must access that particle, and the more the particle's state may be in a "being processed" lock where it can't be mutated by anything else.

Just some thing I've always thought about when trying to see if I could use code patterns and processes to quantify the behavior of the universe. Of course, I wouldn't say I believe these things as fact or anything, just awesome to think about.

[breuleux]

> The reason time dilates as you approach the speed of light is because the faster a particle moves, the more the main loop must access that particle, and the more the particle's state may be in a "being processed" lock where it can't be mutated by anything else.

There's a simpler reason, I think. Any mechanism or "clock" must have moving parts of sorts (atoms, exchange particles like photons, etc.) which means that when functioning or "ticking", some parts move faster than some other parts. But if the mechanism is moving at the maximum speed, it can only be the case that some parts move at c, and others move slower, but these slower parts can never catch up, so the clock disintegrates. It can only remain whole if all parts move at maximum speed, which means it is static: it is frozen in time.

I imagine that given a speed distribution for all components of the mechanism, there would be a smooth time dilation effect as the speed of the whole mechanism increases. I'd have to calculate. In any case, I think time dilation is basically necessary in a system with a maximum speed. I'd expect to see something similar in cellular automata, regarding complex objects that can move at various speeds.

[breuleux]

I feel these limitations are pretty central to the way our universe works, and that without them the universe would be very deeply different. You can only "simplify" the world so much before it becomes something else entirely, or loses its interesting properties. For instance, if you use fluid dynamics to simplify the simulation of the ocean, you'll save a lot of computation and it'll look the same, but life won't evolve in it.

> Isolating each "light cone" of local spacetime allows the simulation to be distributed and run on separate compute clusters, because it limits things to local interactions rather than universal ones.

I don't know if that would make a difference. If you split the universe into zones and assign a cluster to each, they still have to interact with each other all the time because of the boundary. Besides, in a universe where interactions can be "universal", what's the performance gain of local algorithms?

> Things in superpositions only becoming "set" after observation... that is just lazy evaluation.

Really, though, what collapses a quantum state is interaction in general. It happens everywhere, all the time, so I don't know if there's actually anything to gain with this "lazy" evaluation.

[NZGumboot]

>Things in superpositions only becoming "set" after observation... that is just lazy evaluation. It's fun to think about.

Lazy evaluation is generally used to defer some expensive computation until later. In this case, though, the pre-measurement state (a superposition of states) is MORE complicated (computationally speaking) than the post-measurement state, which is nominally a single definite state. It's not a good analogy I'm afraid.

[nickcano]

How, though? Assume a particle has some Flatten() method that is called only when it interacts with another particle. Until this is called, certain data points (let's say, anything not related to it's movement through space) could be either undefined or out of date.

When this function is called, it causes some algorithm to set up these data points. You can extend this to treat entanglement as copy-on-write style branching, where entangled particles A and B have some subset of their properties stored in the same physical memory, but calling this methods causes writing to those properties, and, thus, when the function is called for particle A, a copy of the properties is made for both A and B, and the evaluation is done for both.

[NZGumboot]

You're assuming that movement properties are special, and that non-movement properties don't affect the environment until they are collapsed. This is not the case -- quantum mechanics treats all quantum properties in the same way. An indeterminate spin can be detected in the same way an indeterminate position can be detected (i.e. indirectly, via diffraction and interference effects). It is true that if particle A and B are entangled so that spin A + spin B = 0, then that may help you optimize your "universe computer", but it doesn't gain you much since any gains will be offset by needing to store the fact that A and B are entangled. Not just that, but you have to take into account the fact that any number of particles can be entangled, with any arbitrary linear combination of properties.

IMO the biggest reason to think the universe isn't a computer is just how darn non-computable quantum mechanics is. The complexity grows at an absurd rate. Even for just the position of two particles you have a 6-dimensional wavefunction with (as far as we know) an infinity of possible values for every (infinitesimal) point in that 6-dimensional state space.

[nickcano]

>the biggest reason to think the universe isn't a computer is just how darn non-computable quantum mechanics is

If an intelligence is simulating a universe, they aren't exactly going to be doing it von-neumann style.

[NZGumboot]

That's a good point, but the fact that all Turing machines are equivalent (and the fact that the equivalence is purely mathematical, and does not rely on any laws physics) implies that computers in ANY universe would be similar to ours. At least in terms of what it can compute. Even if the laws of the parent universe made their computers dramatically faster than ours, you would still expect that they would choose our universe's laws of physics so that they were relatively easy to compute, so as to save CPU time. That doesn't seem to be the case -- it takes us a supercomputer to accurately simulate a volume a fraction of the size of a proton. Of course this doesn't prove anything -- it's possible that computers in the parent universe are so ridiculously powerful that even simulating a trillion galaxies is just a drop in the bucket -- but IMO it is suggestive.

[Klathmon]

This gets a bit philosophical, but they may not be simulating the whole universe, just you.

They could just be simulating the universe as you view it, everything outside of your view doesn't exist, you are just a complicated AI that could be fed data.

You have no way of proving that others have "consciousness", and simulating what a single person can experience is MUCH easier than simulating an entire universe, hell I believe we could have access to technology that could do that within my lifetime.

It's also fun to think about. What if my forgetfulness is just an optimization?

[breuleux]

That could happen, but I don't think it makes much sense for it to happen on a large scale, so I wouldn't put a lot of probability on it. Entertainment is the most likely use for that kind of simulation, but I like to imagine they would be less boring than my life is right now.

[nikkwong]

Well, what if you're just 'one of the zombies' and I'm actually the only conscious being just reading your comment? Joking, but in all truthfulness I don't think it's all that unlikely. What if the simulation you are experiencing is being put in place by yourself—that is, 'you' in the higher dimension? Or, it could also be some kind of experiment to see how 'you' given your brain, environmental inputs, etc. grow and respond to stimuli. That doesn't seem unlikely at all.

[Klathmon]

Well depending on scale, your entire "existence" could be a few "minutes" in the "real world" (that's a lot of quotes...)

Much like how we are making ML applications now, you could be a complicated AI who's purpose is to calculate something encoded in your experience. You wouldn't need to know your purpose in order to complete it.

I'm clearly grasping at straws here, its just fun to think about :)

[breuleux]

I find it far-fetched, but you're right, it is fun to think about :)

To be fair, far-fetched things happen all the time, the trick is that we're usually wrong about which ones do.

[taneq]

Look at it this way - if at any point in the future, we ever run even one simulation of the world at our current time, then our odds are 50/50 of being in that simulation. That's only extrapolating from our current capabilities, and not assuming either nested simulations or 'non-historical' simulations. (If we're in a simulation, but not one based on the real universe's past, then we can't even begin to extrapolate from our world to 'universe zero'.)

As for practicality, simulations don't need to support a higher density of life (although they very likely will use approximations and variable level-of-detail to effectively simulate space using less resources than the space itself would use) because they have other advantages; you can pause them, take snapshots, revert to earlier states, branch, etc.

Think of it as like virtualisation for the universe. Even though virtualisation takes up some system resources, there are still huge advantages to running virtual hosts.

As for simulating whole universes being a waste of resources - if we're in a non-historical simulation, we don't even have any conception of what "a waste of resources" means in universe zero. Maybe our entire universe, simulated on a regular grid smaller than the Planck length, is as trivial to store and manipulate on a 5-dimensional universe-zero computer as a 1-dimensional cellular automaton is on a modern desktop computer for us.

[breuleux]

> if at any point in the future, we ever run even one simulation of the world at our current time, then our odds are 50/50 of being in that simulation

That's the contention, though. Someone in the future might simulate some unknown one-year slice of the past, and leave it at that. That would lower the probability significantly below 50%.

> (although they very likely will use approximations and variable level-of-detail to effectively simulate space using less resources than the space itself would use)

The devil is in the details, though. You can save a lot of time approximating an ocean using fluid dynamics, but you're axing all the low level chemical reactions that would make live evolve in it. There are no ways to use variable levels of detail that won't make the simulation diverge from the original. And I'm not even mentioning the synchronization issues and the overhead you'll need to resolve them.

> they have other advantages; you can pause them, take snapshots, revert to earlier states, branch, etc.

Only if you use hardware that's analogous to modern computers, but that hardware is not good at simulation to begin with. Unless you use specialized simulation hardware you're probably going to have to endure billion-fold slowdowns, and I'm being generous.

Assuming you have good hardware for simulation, you can't assume the ability to pause, snapshot or branch, you have to build it in. Each of these features will add overhead and make the simulation slower and larger. Nothing is free.

> As for simulating whole universes being a waste of resources - if we're in a non-historical simulation, we don't even have any conception of what "a waste of resources" means in universe zero.

That's possible, although the fact that it is possible is basically all we can say about the idea. There is no way to evaluate the probabilities involved.

[taneq]

Ah, but I said "even one simulation of the world at our current time" which side-steps that issue.

Agreed about the fact that too many shortcuts in the simulation might make it unable to develop or sustain life, but then we don't know when the simulation would have started or how much it might have diverged from universe zero by now. Presumably it would be intended to diverge anyway because they already know the outcome if it doesn't.

Of course, while we can guess about this stuff, ultimately at this point our guesses are unfalsifiable so as fun as it is to speculate, we're really just shooting the breeze. :)