Hacker News Reader

> Further, simulating atoms requires more atoms. In other words there are impossible problems

Our universe has 10^70 atoms (or whatever). This is an arbitrary number. Maybe in the host universe there are 10^700000000 atoms, and 10^70 atoms are what fits inside a small "tenis ball" from that universe.

BTW, we are already running pretty accurate physicals simulations, if only for extremely short periods of time and small spaces - https://en.wikipedia.org/wiki/Lattice_QCD

Simulating a lot of atoms doesn't require more atoms, with quantum mechanics. You only need to compute the wave function! You need to instanciate atoms only when they're observed (and you can garbage collect them once they vanish again into the wave probability).

I think there are deeper problems in the argument than merely "it would take a lot of resources". I'm concerned by arguments in the form "I can imagine that something is true, therefore it is true". From this perspective, I see the argument like this:

1) (Assume that) we can create a simulation with intelligent beings in it.

2) Therefore, it is not impossible that someone has created a simulation in which we are the simulated intelligent beings.

3) Therefore, it is highly likely that we live in a simulation.

I simply don't get the logical implication between 2 and 3 from "not impossible" to "highly likely". I have the same concerns with the ontological argument for the existence of God:

1) I can imagine a being that is perfect and all-powerful.

2) This being therefore exists, because if it didn't it wouldn't be perfect nor all-powerful.

In both cases I see an extreme case of begging the question.

I don't think you are representing the argument accurately. The logic is as follows:

1. Assuming any rate of improvement of simulations there will reach a point where a simulation is indistinguishable from reality

2. Advanced species have reached this point and are interested in running simulated universes.

3. There are likely to be many more simulated universes than real ones

4. Given the previous 3 it is more likely that we are in a simulated universe than a real one.

Read the original paper from Bostrom for yourself. https://www.simulation-argument.com/simulation.html

By the same argument the people in the outer universe are in a simulation of the outer-outer universe and the people in the outer-outer universe are simulations in the outer-outer-outer universe, etc to infinity.

In other words the theory requires infinite complexity.

The argument says that there is a very high probability, not that it's for sure.

It's curious that the argument presented by Broston is not "it's likely that we're living in a simulation"; it's an either/or:

  Either we're living in a simulation, 
  OR an advanced civilization wouldn't be interested in running one. 
  (OR we won't become an advanced post-human civilization ourselves; 
  though that wouldn't prevent other species from running one).

Apparently too many people dismiss the second possibility and forget reporting about it when discussing the topic.

Also, the paper both assumes that galactic-wide computing power is physically possible, and that computing power is all it takes to generate civilization-deep simulations. I can think of several limits of physics and complexity that could make either hypothesis less than plausible, or at the very least, make option two (the post-humans not being interested in running such simulations) as likely as the "we're in a simulation" one.

>Apparently too many people dismiss the second possibility and forget reporting about it when discussing the topic.

I doubt most people even HN types have read the actual argument.

> I can think of several limits of physics and complexity that could make either hypothesis less than plausible

No one is saying this is feasible in the short term and no offense but I highly doubt that you are privy to the physics available to humans hundreds or thousands of years in the future.

I doubt the limits of physics themselves will change in the future :-P

There might be more knowledge on how to work around those limits to achieve efficiency closer to the theoretical limits. But to assume that a future civilization will have access to nearly infinite computing power (to the point of not caring about throwing several world-computers to a simulation side-problem), well, that's magical thinking.

I'm not saying that it can't happen. I'm saying that it's not the most likely possibility given our current understanding of how the universe works (which while not perfect, is quite a lot).

It all boils down to the first step in your version of the argument: if simulations are improving fast now, they will keep improving forever and reach any arbitrary level of power given enough time. It's the fallacy of infinite growth, ignoring the likely possibility of the growth process meeting some inner limits that will make it end.

I think you are seriously overestimating our current understanding of the universe/physics. For instance we don't even know if our universe is finite or not, if it's infinite then your point is moot and you literally would have access to infinite compute.

I disagree, I have seen research indicating that it's reasonable to have a computing system with on the order of 10^69 bits. There doesn't seem to be anything magical about this thinking at all. https://arxiv.org/pdf/1705.03394.pdf

Who says that the simulation has to be done in realtime?

Without any outside reference the NPCs inside the simulation have no way to know that one second in the simulation has taken them for example a year in real-life.

It is not done in real time, it is done in relative time!

Elon Musk did respond to a similar question that you have asked here. The entire Universe need not be simulated at the same time. It could be a generative program. The reality could be limited to just what we are observing.

If you have enough resources to lazily simulate the entire observed universe, you probably have enough resources to consistently simulate the entire observable universe.

I haven't kept up with the latest out-of-universe hardware upgrades, so I could be talking out of my ass here.

With that said, why would you simulate something so ridiculously large and detailed if you only had the resources to run a small portion of it at a time? It's like buying an 8K 3D monitor and hoping your 3DFx Voodoo 2 will do the trick.

You might do this if you didn't have the computational power to simulate the thing you wanted in real time. Games are hard locked to require the computation to be performed within the frame time of the gameplay, but much more detailed and complex computations are done all the time to pregenerate an animated movie or video that can then be "played" in real time. The generation of the content often takes many times as long as the content actually is, but the purpose of doing it this way is to generate more detail per frame than you would be able to compute in the time between each frame when played at the ideal speed.

What is "real time", though? Time is relative within the universe itself, what hope is there that it even exists outside of it?

Using crude current metaphors: If I build a game, the concept of time within that game is, essentially, each tick of the game (a frame, for example, or a server-side tick). You can even have portions of the game world tick faster or slower relative to others.

But those ticks mean nothing outside of the game. Yes, in "real time" the ticks are constant... but I can pause the game at any point, it stops progressing. I can step it one tick forward, and for the game one tick has passed, but for me, any amount of time can pass.

In other words, there's no relation between time inside and outside the simulation, because it's a simulation.

Sorry if I was unclear, I didn't mean to imply a link between "real time" in the game/simulation and the real world, because you are completely correct in noting that there is no specific relation between game time (ticks) and "real" time (seconds or w/e units you want).

The link in the "time" that I am describing is one where the game graphics are rendered in 60 frames per second. If the computation cannot be completed within 1/60th of a second, the frame is delayed because it took too long to render.

My comment was largely in response to your assertion

"If you have enough resources to lazily simulate the entire observed universe, you probably have enough resources to consistently simulate the entire observable universe."

First, assuming that it requires less resources to "lazily" simulate something than to fully simulate it, then no, I do not believe that just because you can do the first means that you can 'probably' do the second. If you had enough resources for the first and not the second, but the first was good enough, why not do the first?

In response to: "With that said, why would you simulate something so ridiculously large and detailed if you only had the resources to run a small portion of it at a time? It's like buying an 8K 3D monitor and hoping your 3DFx Voodoo 2 will do the trick."

The reference to animation was largely in response to this, because that is exactly what the purpose of animation is. They have enough resources to render the animation in whatever length of time it takes to render, but they can't do it quickly enough to just render it frame by frame for the viewer.

Ah, yes, I see where you're coming from.

My take on it is that it just fails occam's razor. As in, it turns the theory "The universe is a simulation", into "The universe is a simulation, and the simulation is imperfect and lazy-loaded". It's adding even more requirements to an unprovable theory.

Essentially, it turns "God exists" into "God exists, and he has a beard".

Ahhhh this makes more sense with your previous objections and I wasn't understanding it as well before. I always took little tidbits of the scenario that pointed to "lazy-eval" to be indicative of it being more likely to be a simulation because lazy eval is popular among programmers. Obviously when I write it out like that I can see where I lean on the same logic I argue against in other scenarios which is that we can't make any assumptions about intentions/motivations/etc. of the simulation creators, and the assertion that they would have similar programming paradigms is a pretty big assertion for sure.

Re: lazy simulation

Would the fact that our “observable universe” is limited by the speed of light and time since “creation”, be a kind of a “lazy” simulation?

i.e. the “servers” that process us would never need to know about anything in the region that will never interact with us.

I just realized I wanted to address this additional point:

> If you had enough resources for the first and not the second, but the first was good enough, why not do the first?

Why would you build something that is so immensely large, if you can only simulate a portion of it at a time? This hypothesis suggests that you would have to make the universe at least this large, or that lazy simulation is simpler.

Again thinking in terms of occam's razor and path of least resistance. It seems silly to make the unprovable theory even more complex and even harder to prove.

With that said I like Razengan's take on it; that is, c would be the speed at which the simulation computes new results (it also assumes the simulation isn't paused during computation).

Remember, even in simulation theory, there's nothing that suggests we humans are the intended product or object of observation; just like an ant farm that happens to be somewhere in some house in GTA5 isn't specifically the point of GTA5. So I wouldn't conflate "lazy loading" with "there are humans in this tiny corner of space, so we only need to simulate this tiny corner of space, and that makes a big difference in terms of computing power".

>Is it actually possible to simulate the entire universe?

For you to be a simulation doesn't require the entire universe be simulated. It only requires simulation of you and the inputs you receive. Scale that and you can simulate a population to see how they react, all without needing to simulate every unobserved event to the finest detail.

It's how we can simulate all sorts of things now that relate to the physical world without having to simulate the physical world in it's entirety.

you need to simulate unobserved events when they are observed. I'm not looking at the Nile river right now but when I look at in again in 50 years it better look correct for the 50 years that passed. For it to look correct requires simulating everything about it, everything feeding into it, everything it affects, from the moment I stopped looking at it until I looked at it again.

I'm not sure what you said refutes his point. It still doesn't require the entire universe to be simulated.

If you observe the Nile, then yes it becomes an input, so that then would need to be simulated. This could take however long is necessary to do.

It still does not necessitate simulating the entire universe down to the minutiae.

Perhaps if the people in the simulation purposefully set out to "observe" somehow their entire universe in one timestep, in such way that _necessitates_ the complete simulation of their universe (like, say, running an experiment that involves all atoms interacting with all other atoms in their universe), then maybe it will be a problem. Then again, maybe the parent universe just has more atoms & computing power to throw at it, or it will generate a simulation that is "close enough" to the real thing that the people in the simulation will not be able to distinguish.

>For it to look correct requires simulating everything about it, everything feeding into it, everything it affects, from the moment I stopped looking at it until I looked at it again.

No, it only needs to at the two points you take inputs provide answers that satisfy your understanding on what the inputs should be, and your understanding is formed by other inputs you've been fed, making the entire thing much easier. Satisfying you is probably much easier than satisfying someone who really studies rivers.

Neither you nor an expert needs simulating it much at all while you're not looking at it. And I doubt there's any case to simulate to any atomic or subatomic level. The gravitational pull of Neptune can likely be ignored. Neutrinos that may scatter from Antartic ice may hit it, but can probably ignored. And on and on. You're inspection is no where detailed enough to check if these things were taken into account for a decade while you were away. All of this detail is irrelevant.

It may well be possible to do a good job with a very rough simulation of water and time, all of which could be computed on demand.

So really all that needs stored is enough state to simulate in case you look at it again, and no work needs done on it between those looks. And this is a trivial amount of storage and computation to satisfy your two causal glances when compared to needing to simulate an entire universe down to all 10^120 subatomic particles.

Please note how the simulated world in GTA5 is based not on atoms, but on entirely different concepts which allow modeling "large" amounts of in-world physics using a "small" amount of out-of-world atoms.

As long as we don't suppose that the simulated world copies all the features of the simulator-running world, there should not be a problem of needing more outer-world stuff to simulate a given amount of in-simulation stuff. They can just be quite different.

There's still an information-theoretic argument that the complexity of a simulated world cannot exceed the complexity of the outer "true real" simulator-running world. We still could see "more complex" activity of a simulated world by spending more "real" time to model "simulated" time, and having the simulated world to more often and more evenly fill various parts of its phase space, as opposed to the outer, simulator-running world.

> note how the simulated world in GTA5 is based not on atoms

That's actually my point. GTA5 doesn't work on atoms but the simulation of our actual universe would have to. Any NPC in our simulated universe can turn any building into rubble, turn any sand into glass, empty any lake, divert any river, put any drop of water under a microscope, mix any chemicals and watch their reactions, put any atoms in a particle accelerator, draw an animated stick figure with an electron microscope.

I'm arguing we can never simulate a universe because we'd need a computer larger than the entire universe to do it. I'm arguing that the thought experiment of "simulations will continue to get better until we can put A.I NPCs in a simulation of our universe and it will be impossible for them to tell the difference" is itself impossible.

The argument for why it's "likely" we are in a simulation is because it's argued it's inevitable we'll create such simulations ourselves. In doing that there will be more simulated people than not. It a huge assumption that "it's inevitable we'll create such simulations ourselves". If that is not inevitable then the likelihood we're in one drops dramatically.

A counter argument is we'll simulate a smaller universe for our NPCs but that argument doesn't really hold as it would argue that every universe is a simulation running in a larger universe to infinity. That would require infinite complexity

Why would a simulated being expect to see atoms in their universe? Their physical theories would be a result of whatever experiments they can do from where they're sitting.

What's so special about our physics that we would expect any possible consistent universe to be built this way? Maybe we're being simulated in a vastly different parent universe with more available computational power.

You're now basically saying "because magic!". Sure the outer universe could be based on something else but the thought experiment is that we'd be doing this in our universe. It doesn't require some other universe with magic unknown properties, it requires first that we can do this in our universe to a sufficient level that NPCs in simulations in our universe can't easily see the issues.

Sure, but why should we expect any 'host' universe to be based on such things as atoms, either? Perhaps our atoms are as coarse to them as models in gta are to us.

I don't think the argument depends on simulations being able to infinitely and recursively create sub-simulations. You only need one layer of simulations to make the odds work.

> Is it actually possible to simulate the entire universe?

In quantum mechanics there is something called collapse of wave function, which takes place when observer is making an observation on certain object. When there is no observer state goes into superposition which means it is one and zero at same time. So basically this means that object can exist and doesn't exist at same time.

That's why Einstein said he likes to think that moon exist even if no one is watching.

This looks similar to mechanism we use in rendering games today, you only load and render the view that players is looking at and the stage/level that player is in. Of course it's a lot more simplified in case of games. You can save resources using those methods while making huge scale simulations.

I wonder if by the time we solve a 'theory of everything' (or understand by some other means the basic mechanisms of our universe), we'll be able to efficiently, procedurally, generate a simulated universe.

Also, the Fermi paradox could be a solution - we don't see any other alien life forms because we're running in a computationally limited simulator, and everything outside our (relatively easy to simulate) solar system is running on minimum resolution (so to speak).

But we could simulate (in theory) a smaller universe than our own. Maybe that is what is happening. We can also simulate alternate physical properties. Maybe the simulating universe has vastly different rules/requirements.

Have you counted into your math that we're close (within 30 years I'd say) to more than 5 different computing revolutions?

to simulate our universe perfectly and in real time would take a computer the same mass as the universe.