Defeating the virtual reality argument

[ceptimus]

I suppose most of you know the 'proof' that we are all just figments of a virtual reality simulation?

The argument goes like this:

1. In the future, computers will become so powerful that they are able to simulate complex systems, right down to the sub-atomic level.

2. Future historians, interested in history, will run virtual reality simulations of the entire Earth to see how things might have evolved.

3. Because of the large numbers of such computers, and their ability to simultaneously render many different possible worlds, it follows that there will be millions or billions of such models rendered. Many of the rendered worlds will contain computers which are in turn rendering yet more worlds.

4. Since there are billions of artificial worlds, but only one real one, it follows that for every one real person, there are billions of rendered ones.

5. So the chances are that we in this world are rendered. It is vanishingly unlikely that we happen to be in the one 'real' world.

Now most people agree that this argument fails to pass the 'sniff test'. We know we are real, don't we? But where is the flaw in the argument? How can it be logically refuted, or if it can't, why don't we believe it to be true?

Perhaps some of you do believe this argument, and you're happy to believe that you are probably not real? Hey, if the rendering is so perfect, then it doesn't make any difference anyway!

[seebs]

I am not particularly convinced that it becomes possible to simulate a world in sufficient detail. Furthermore, for all we know, the simulations are necessarily imperfect; perhaps you cannot properly simulate people.

[Beth]

So, is "God's" world the real world and he, the creator?

[wade-w]

Seebs has an excellent point. Another problem is that there is an unstated assumption that even if such a simulation is possible, it is possible to know the exact state of the universe at some point in the past. Also, there is another unstated premise here that the universe is deterministic.

But there is another even more serious problem with the argument. How do we get from point 4 to point 5? It seems to assume that this has already happened.

[Godless Wonder]

Two problems.

1) Where do you get your initial conditions?
2) Any computer big enough to simulate the entire earth would have to be constructed of more atoms than the earth itself, unless you can figure out a way to convince one atom to simulate more than one atom (hint, you can't.)

I don't think there is any real reason why it shouldn't be possible to simulate for example, a 10x10x10 foot cube of matter given enough compute power and a good enough model (perhaps not perfect) if the model happened to contain the atoms of a person, I don't see any reason that simulated person shouldn't be able to function. Granted, I'm assuming that we'd at some point be able to understand physics enough to make a decent model, and that somehow you'd get the initial conditions into the computer which include all the atoms of a human (one hell of a feat). I just don't see any reason apart from the obvious practical considerations why it shouldn't work.

[Desert Dweller]

Oopps I just found out I'm a hologram.

[Farren]

Seebs makes an interesting point. Were this a virtual reality, we could assume glitches, which would amount to things like defiance of physics and logic that we don't actually observe.

Its concievable that one could build a glitch-free VR, but unlikely that the majority of world-VRs would have no way of communicating with the "outside world". So its fairly likely that, were this a VR, there'd be some interface, some means of detecting that theres another universe wrapped like a shell around our reality.

Another consideration is that one cannot simulate a particular reality without, of necessity, using vastly more matter than the "matter" in the simulated world. In modern computers, for instance, a single bit requires millions of atoms to store. Were there a one-to-one correlation between smallest units of matter in the simulation and smallest units of matter in the parent universe, the simulation wouldn't be a simulation.

So the simulator must always be significantly larger in apparent mass than the simulation. If one looks at our own situation (a minute speck in the universe where all of our history has played out) and assumes a similar parent universe, the universe in which the simulation is being run would be orders of magnitude larger than our own.

Its not unreasonable to envision such a situation, where tiny simulated universes exist within tiny simulated universes up through a procession much like a Russian doll, with some inconceivably huge universe at the top of the heirarchy, but, IMO, its an interesting consideration.

There's also the analog/digital consideration. There are scenarios (such as the three-body gravitational problem) which simply cannot be computed properly with digital symbolic logic. The simulator would have to be analog in order to simulate an analog reality.

Furthermore there's the distinct possibly that certain scenarios could not be compressed, even when made analogue. As far as I know the only way of completely accurately determining the history of three bodies exercising a gravitational influence on each other is to replicate the problem. In other words, make three bodies that are exactly the same as the originals and place them in an environment that is exactly the same.

[wade-w]

I suppose I should have been more specific. Several of you have mentioned simulation of atoms. Yet the argument in the OP specifically says the simulation goes all the way to the subatomic level. If our understanding of Quantum Mechanics is reasonably accurate, specifically the Heisenberg Uncertainty Principle, then such a simulation is impossible even if we ignore the storage requirements. For that matter, it's also impossible to know the exact state at the beginning of such a simulation.

[ceptimus]

Quantum mechanics is deterministic until the point when the 'measurement is taken' (or waveform collapses, or whatever term you wish to use). Anyway, I assume that the guys who build these simulators will have a 'correct quantum theory' available to them by then. Not the messy hotch-potch we have today.

You don't have to know the starting conditions. That's the whole point of the simulation. You could begin with just a interstellar cloud of gas and dust, or you might want to make a primitive Earth model and maybe even seed it with a few RNA strands.

The Virtual Reality we happen to inhabit might be completely different from the 'real' one. There might be some bug eyed monsters examining our model earth right now, and commenting how unusual it is for the evolved dominant species to be a bipedal mammal.

Also, remember that everything you ever experience, you receive through nerve impulses entering your brain. To simulate just you would only require the accurate simulation of a couple of kilogrammes of matter (your brain) and the rest of the simulation can be much lower grade stuff - only a few orders of magnitude better than an airline simulator of today.

Also it doesn't have to be 'real time'. When there is some tricky rendering to be performed, the simulation can be slowed down as much as necessary. As your brain would be part of the simulation, you would necessarily be unaware of any speed changes.

[copiae]

Two problems.

1) Where do you get your initial conditions?


Assuming a VR simulation of an entire universe, the initial conditions could be based on the various models we have of how the universe began. If it is a simulation of its host universe (or some part of it), presumably, the initial conditions would be whatevers going on prior to when the simulation commences.

If neither of these are preferable, then we can aways use evolutionary engineering to come up with the intial conditions. Presumably a computer powerful enough to simulate a universe is capable of sifting through various permutations of conditions until one conducive to the end goal is found.


2) Any computer big enough to simulate the entire earth would have to be constructed of more atoms than the earth itself, unless you can figure out a way to convince one atom to simulate more than one atom (hint, you can't.)


Distributive computing means that space conditions wouldnt really be that much of a problem. Also, I am sure that clever optimising can remove much of the overhead (for instance, whats the point of figuring out something that isnt being perceived, if this was just a human-centric simulation, and assuming a certain environmental static-ness)?

edit: I see ceptimus beat me to it.


I don't think there is any real reason why it shouldn't be possible to simulate for example, a 10x10x10 foot cube of matter given enough compute power and a good enough model (perhaps not perfect) if the model happened to contain the atoms of a person, I don't see any reason that simulated person shouldn't be able to function. Granted, I'm assuming that we'd at some point be able to understand physics enough to make a decent model, and that somehow you'd get the initial conditions into the computer which include all the atoms of a human (one hell of a feat). I just don't see any reason apart from the obvious practical considerations why it shouldn't work.


If, however, consciousness was not reducible down to atoms and thier interactions, then all you would end up with is a simulated zombie.



There's also the analog/digital consideration. There are scenarios (such as the three-body gravitational problem) which simply cannot be computed properly with digital symbolic logic. The simulator would have to be analog in order to simulate an analog reality.

Furthermore there's the distinct possibly that certain scenarios could not be compressed, even when made analogue. As far as I know the only way of completely accurately determining the history of three bodies exercising a gravitational influence on each other is to replicate the problem. In other words, make three bodies that are exactly the same as the originals and place them in an environment that is exactly the same.


Hm. I must say Im having difficulties with this statement. Digital logic misses the nuances of many interactions, but why can we not use something like fuzzy logic to simulate said nuance?

[seebs]

Seebs makes an interesting point. Were this a virtual reality, we could assume glitches, which would amount to things like defiance of physics and logic that we don't actually observe.

Unless, of course, this is why miracles used to be more common. :P

[ceptimus]

"Whoops! Major glitch there on simulation A3923916-B28. The people in the simulation have figured out that something is wrong! Once the programming geeks have sorted out the problem, we'll restart the simulation from the last backup - it's only a month ago, in the simulation time-frame."

[Farren]

Hm. I must say Im having difficulties with this statement. Digital logic misses the nuances of many interactions, but why can we not use something like fuzzy logic to simulate said nuance?

We could, but then progressiveluy more accurate digital descriptions wouldn't tend to corroborate the simulation (which in our world at least, they do). What I mean by this is that things like weather simulation (which is, IIRC, an NP-complete or irreducably complex problem) slowly but progressively get more acurrate over a longer period when we through more and more brute force computing power at them.

"Brute force" computing power here basically means calculate to more decimal places and model more atoms (in the programming, not physics, sense of "atoms", i.e. irreducable units). This uninterrupted trend to corroborate reality indicates either

1. If it is a simulation, the "atoms" of the simulation are simply smaller than our simulated ones and the number of decimal points higher than current supercomputing standards. If this scenario true a point will ultimately be reached where simulations in this "reality" yield results that conflict with observation in an irrational way.

2. That the reality is, indeed, reality and simulations in that reality will continue to corroborate that reality to a greater and greater extent.

If we presume (1) above, we must take a position akin to Karl Popper's. IIRC, he argued that, if you observe a phenomenon a trillion times and it doesn't deviate from a set behaviour, you still can't say its a fundamental truth. You just might not have observed the exception.

While such thinking is unassailable in terms of logic, it is problematic in terms of utility, inasmuch as presumption is the foundation of just about all useful science. Using such logic, one can dispute gravity, but its been far more useful to us to presume its existence and build aeroplanes and rockets to conquer it.

[Farren]

Seebs makes an interesting point. Were this a virtual reality, we could assume glitches, which would amount to things like defiance of physics and logic that we don't actually observe.

Unless, of course, this is why miracles used to be more common. :P

I know your post was probably as much jest as seriousness, but it brings me to a pet bugbear of mine.

Assuming we are, as we appear to be, a tiny civilisation on a speck of a planet orbiting a middling star in a medium sized galaxy in a small supercluster among billions in the observable universe in quite an advanced stage of that universe's evolution, wouldn't the glitches have long ago been worked out?

I mean if we take human existence, which is, according the Carl Sagan, in the last second of a normalised (to an earth year) cosmic calendar starting at the big bang, then thinking the system was adequately patched to make glitches disappear in the span of our history would be presume that 21st century civilisation and its understanding of concepts like VR had conveniently arrived in the last second of a year long project, as the last patch (that made everything work 100%) was applied.

One could, I suppose, presume that the apparent age of the universe and other internally consistent foundations of our current theories of the universe were simulated and in fact apparent history itself is simulated. BUT: Why would anyone "Back-date" a simulation? Its unlikely that anyone running a simulation would create some made-up history. It seems to go against the entire spirit of the venture.

I say its a bugbear of mine not because of considering this specific issue, but a similar one. There's a tendency among critics of evolutionary theory to use only the scope of the earth when calculating the chance of both abiogenesis and subsequent evolution of complex species. Most of these arbitrary and plainly wrong calculations fall apart when the entire universe is used as the framework for chance events.

[squian]

I'm not sure it's so implausible as suggested in previous posts. If we are in a simulation, we do not know what the "physics" of the "real" world is like. Perhaps it is perfectly feasible to conduct large scale, high-speed computing without a computer the size of the universe. For that matter, if we are in a simulation, we have no way of knowing it's high-speed. It may take years to compute each instant but we have no way of discerning the delay.

Alternatively, perhaps the simulation model is actually very simple. If there were a "Grand Unifying Theory", it might not be that hard to realize a simulation of everything.

Although others have suggested we should expect "glitches", I'm not so sure. As part of the simulation, how would we recognize irregularity? Perhaps we would even be programmed to ignore them (hmm, skepticism as a control mechanism?). To be sure, if the universe is a simulation it would explain a lot. For instance, the way abstract mathematics eerily predicts concrete physics. Hence, perhaps it is actually the regularity, not the glitches, that indicate a virtual reality.

All that said, the fundamental problem with this argument is the first point. Fundamentally, I see no reason that we can make such an assertion. Although organic chemistry is bound by the laws of inorganic chemistry, the interactions of organic chemicals is so complex that we need new models to explain and predict the emergent behavior. Likewise, as we go up or down the scale (size and complexity), there may not be much value in a model that "goes all the way to the sub-atomic level".

Moreover, any model so complete would need to contain itself. And that is so fundamentally paradoxical that I cannot accept the notion of such detailed simulations. Then again, maybe it's just paradoxical for our virtual reality but not so in real reality.

Mostly, I object to the notion of "odds" and "likely" for things that have never been observed. It's like asking, "What are the odds of Santa Claus?" or "How likely is the Easter Bunny?" No amount of deduction changes the fact that we have never seen a real simulation so complete that people in it are self-aware.

Whether it's likely or not, what difference does it make? Does it matter that we are virtual or real?

[copiae]

We could, but then progressiveluy more accurate digital descriptions wouldn't tend to corroborate the simulation (which in our world at least, they do). What I mean by this is that things like weather simulation (which is, IIRC, an NP-complete or irreducably complex problem) slowly but progressively get more acurrate over a longer period when we through more and more brute force computing power at them.

"Brute force" computing power here basically means calculate to more decimal places and model more atoms (in the programming, not physics, sense of "atoms", i.e. irreducable units). This uninterrupted trend to corroborate reality indicates either

1. If it is a simulation, the "atoms" of the simulation are simply smaller than our simulated ones and the number of decimal points higher than current supercomputing standards. If this scenario true a point will ultimately be reached where simulations in this "reality" yield results that conflict with observation in an irrational way.

2. That the reality is, indeed, reality and simulations in that reality will continue to corroborate that reality to a greater and greater extent.

If we presume (1) above, we must take a position akin to Karl Popper's. IIRC, he argued that, if you observe a phenomenon a trillion times and it doesn't deviate from a set behaviour, you still can't say its a fundamental truth. You just might not have observed the exception.

While such thinking is unassailable in terms of logic, it is problematic in terms of utility, inasmuch as presumption is the foundation of just about all useful science. Using such logic, one can dispute gravity, but its been far more useful to us to presume its existence and build aeroplanes and rockets to conquer it.

Farren,

Why would progressively more accurate digital descriptions tend not to corroborate the simulation, if you don't presuppose 1. and all of its implicit assumptions - after all, the more accurate something is, the more corroborative it is?

Regards,

[Godless Wonder]

[i]Godless Wonder wrote:
2) Any computer big enough to simulate the entire earth would have to be constructed of more atoms than the earth itself, unless you can figure out a way to convince one atom to simulate more than one atom (hint, you can't.)


Distributive computing means that space conditions wouldnt really be that much of a problem. Also, I am sure that clever optimising can remove much of the overhead (for instance, whats the point of figuring out something that isnt being perceived, if this was just a human-centric simulation, and assuming a certain environmental static-ness)?
If you're starting from Big Bang, there aren't going to be any humans. There won't be anything that is perceived, so if you think you can get away without simulating things which aren't perceived, you might as well not simulate anything. You have to simulate everything because some of the things you're supposedly looking for are so low probability (e.g. beginning of life) that you need gobs of space and gobs of matter to bang around randomly for eons to get started. You can't optimize that away, because you don't know ahead of time which parts are the important parts. Well, you might be able to optimize away, say the earth's core into some simplified form, and the same for the sun and outerspace, but then you've still got to simulate the entire earths surface, atom by atom, and then you've only got one planet to get your life going, which isn't going to be enough. You'll be waiting around a long long time, and your computer will still be gigantic (composed of at least as many atoms as you are bothering to simulate. Even if you don't simulate all the atoms simultaneously (timeslice the problem) you still have to store all the information about all the atoms you aren't currently processing, which will take at least one atom per atom simulated. Going subatomic only makes that problem worse.


If, however, consciousness was not reducible down to atoms and thier interactions, then all you would end up with is a simulated zombie. a) there's no reason to think that is the case, and b) what makes you think you aren't exactly such a "zombie" already? I contend there's no difference between the consciousness of a "real" human and the consciousness of a simulated one, provided the simulation is good enough. (e.g. imagine a 3d atomic xerox machine that zaps instantly the information about all the atoms in a 10x10x10 cube in which you, a video camera and a television sit. By connecting the output of the real video camera to the simulated inputs of the TV and the simulated video camera up to the real TV, you could then communicate with your simulated self, and ask yourself questions to which only you know the answer. So far, every bit of evidence we have suggests that consciousness is reducible to ordinary chemistry.

[copiae]

If you're starting from Big Bang, there aren't going to be any humans. There won't be anything that is perceived, so if you think you can get away without simulating things which aren't perceived, you might as well not simulate anything.


I guess I should have elaborated on human-centric (The focus of the simulation is on humans, and thier interactions) and environmental static-ness (No major changes to the environment are necessary). A good example of such a world is the 'fake' world in the matrix, even though the consciousness present in it is seperate from the simulation.



You have to simulate everything because some of the things you're supposedly looking for are so low probability (e.g. beginning of life) that you need gobs of space and gobs of matter to bang around randomly for eons to get started.

I don't know if your familiar with adapative processes and evolutionary engineering, but this problem is not as insurmountable as it may seem. Basically, evolutionary engineering is the application of 'natural selection' within the digitial environment. What I mean by that is this: We have a thwopping powerful computer thats capable of simulating a universe. If we don't know the initial conditions, we can just throw pseudo-random permutations of conditions in, and see what happens. If its totally wrong, we cull it, but if its close(r) to what we want to happen, we can save it, and base future permutations upon just it, and other close(r) permutations.

Manually, this would be tedious, to say the least, but luckily, we can automate the process.




a) there's no reason to think that is the case,


See below.



and b) what makes you think you aren't exactly such a "zombie" already?


By zombie, I meant the human equivalent of a vegetable: Perhaps with basic physiological functionality, but with no higher-order intellectual capabilities. At this point, many of you may be inclined to re-ask the original question to me verbatim, but nonetheless... :)



So far, every bit of evidence we have suggests that consciousness is reducible to ordinary chemistry.

Cool. If you have the time, what evidence is that? Or, are you just talking about our inability to verify the existence of a soul?

[wade-w]

But there is another even more serious problem with the argument. How do we get from point 4 to point 5? It seems to assume that this has already happened.

Well? As far as I can see, the argument goes from "this will someday be possible" to "This has already happened."

[Farren]

Farren,

Why would progressively more accurate digital descriptions tend not to corroborate the simulation, if you don't presuppose 1. and all of its implicit assumptions - after all, the more accurate something is, the more corroborative it is?

Regards,

I'm unclear as to what the question is here but I'll reframe the argument and perhaps that will answer you. Otherwise, you might need to reframe the question so that I understand it clearly.

There are physical scenarios that are chaotic and complex: Fluid dynamics, multiple body gravitational interactions etc etc. Its not our understanding of them that makes it difficult to compute outcomes. Its there chaotic nature. Because the tiniest variance in initial conditions can massively affect the outcome, we have to model them down to the smallest subatomic unit to get an accurate picture.

Weather systems are an example. Since we're incapable of doing this in reasonable time without, ultimately, having a model as large as the thing we're modelling, we constantly get more accurate forecasts over longer periods, but we can never determine the outcome for any arbitrary future point in time.

What we can do, with enough brute force and sufficient understanding, is determine limits to the possible over limited periods of time. IOW, within limited periods of time we can determine what cannot happen. Were the presumed universe simulation we're in resort to fuzzy math at some point, we would eventually detect anomalies because an impossible situation would be detected.

It wouldn't necessarily be an impossible situation in the phase space of all possible weather conditions on earth, but it would be impossible given the initial conditions and the period of elapsed time. With adequate brute force we could detect such anomalies.

[Farren]

I've just realised that the argument in the OP fails if a number of reasonable presumptions are made

1) That there is a minimum irreducable unit of energy in the ultimate parent universe. As argued by myself and others, any child simulation universe by default would be tiny compared to the universe it is simulated in.
2) That most beings would not want to utilise all the matter in their universe for simulations. As squian pointed out, ascribing chance to these things is a fairly arbitrary exercise, but since we don't know, 50/50 seems fair, in which case theres a high chance this is true.
3) That most beings will be incapable of using up all the matter in their universe, or even a large fraction of it. I don't have a formal argument to establish this but it seems reasonable to assume that, for any universe in which beings evolve, there would be a great many of them wanting to use the same energy for various purposes and that, like our own universe, the constraints of their natural laws might prohibit anything approaching universal consumption.

If various combinations of the above are assumed, then its fair to assume that there's a parent universe containing most of the energy in the entire Russian-doll like system and most of that energy isn't engaged in simulating things.

This is a rough proposal because I'm stuffed and must sleep now so forgive any crappy logic.

Good night :)

[seebs]

Maybe that's what dark matter is; matter we can't afford to simulate, but which is required to make the other equations line up.

[Desert Dweller]

Tttttthhere's a .................................... glitch in .............the fabric............
I just had this moment before and am experiencing replay............ now it's all gone the color of the blue screen of death...........waiting...........it's coming back...........ah, that's it. I'm back now and up with present time again. Damn machine I wish they'd fix it. I particularly hate having to have the same nightmare more than once!