?

Log in

No account? Create an account
Poker, politics, philosophy
simulation commentary 
15th-Aug-2007 08:38 pm
poker
My last post inspired a good bit of conversation in the comments, nearly all of which thought the argument that we might all be in a simulation was pretty much bunk.

The argument itself held that either we are very likely to be in a simulation, or one of the following is not true:
  1. It is possible to have human intelligence without a brain by simulating a brain computationally
  2. Humanity will eventually develop the computational resources to easily simulate Earth's history at least through the present and near future
  3. Posthumans with access to the requisite computational power will never choose to run such a simulation


Interestingly, nobody argued with point 1. Apparently everyone agrees that you don't need a meat brain to have intelligence.

Several people objected to 2. The first objection seems to be that you can't simulate the history of humanity without simulating the history of the entire universe, which is computationally infeasible.

This seems so obviously untrue to me that I wonder if I'm misunderstanding the point. Your simulator needs to simulate the mental states of every human it cares about, plus all the sensory input those humans have... and that's it. For consistency's sake you probably need to keep a fairly detailed model of the things that people are likely to interact with, but you don't need to track the paths of every electron unless someone is setting up a cloud chamber.

Our sensory bandwidth is really low, and if you can model the mental states of a person completely, you presumably know what they are looking at and investigating. You don't really need to simulate the molecular behavior at the center of Jupiter until someone starts an experiment that looks at it -- something simpler that looks reasonable is all you need for the vast majority of the simulation.

The second objection seems to be that we may not ever get large enough computers to simulate even one person. It's true that Moore's law can't hold forever, since there's a physical upper limit to computation in the universe; presumably we'll hit the Moore's law wall well before then. However, we don't need exponential increases in computing power to get to the really impressive resources needed for whole-world simulations, we just need a lot of time plus any growth at all. I have no doubt that we'll have enough power to easily do such a thing if we want to, assuming society doesn't collapse before we get there.

Also brought up at least twice was that the programmers wouldn't be good enough to get such a simulation right. This is possible, I suppose, but we're very very early in the history and development of programming languages, and we can already do pretty amazing things. There's going to be a lot of progress between now and the postulated future of big chunks of computronium.

The most interesting objections came to point 3, though nobody made what I thought was the most basic observation about post-humans: by definition, we have no idea what they will or won't want to do. Maybe they won't be interested in historical simulations (or alternate world simulations) for reasons that wouldn't be obvious to us. If your greatest ambition is to keep the fire lit in your cave, or to find the secret of bronze that your enemies are using to drive your tribe away, are you equipped to understand the deflation of the credit bubble, or why people are simultaneously enthralled and disgusted at one man's ability to hit a ball with a stick, really hard. It's just unreasonable to try to predict what a post-human might want or not want to do.

Having said that, some of the objections are pretty questionable.

Will humans really evolve to the point that they would be willing to expend ludicrous amounts of resources just to achieve what is basically the ultimate game of SimCity... for the hell of it?

It shouldn't take much effort to come up with interesting research goals that could be evaluated with complete simulations. It's a much more plausible way for a future Hari Seldon to validate psychohistory, for instance. Really anyone interested in history, economics, political science, etc., should be interested in historical models. Unless, of course, those fields have been completely mined out by the time we have the resources to fully model human history.

Anyway, all that's required for the sim argument is for some researcher or hobbyist to take it up, for any reason at all. I can think of a lot of arguments why no particular group or individual would simulate history, but to claim that no member of humanity will ever do, well, anything, seems like a bit of a stretch. The arguments against computation are much more persuasive than the arguments against future behavior of post-humans.

My favorite comment comes from an anonymous poster (not Ming, the usual suspect) who posted a couple small comments, then the next day came back with this:

As written before, the universe can be seen as a quantum computer calculating every possibility simultaneously real time. The universe as we know it is a personal time bound interpretation of an approximate outcome of infinite (time)local calculations. The global universe as we know it is derived from cultural classifications of perceptions, shaped by interpretations within basic human experiences, and when science is involved ruled/written out by/in abstract mandatory formulas.
...
The next 30 to 50 years will be fussy. We'll need to adapt to complete new concepts of reality which will turn our lives upside down. We will see severe clashes between old world perceptions like religions and new insights which are proved by then everyday technology. Beyond that we will find ourselves in a supercultivated hyperfriendly universe in which human nature matches exactly with nature. And yes, probably we'll meet neighbours too. It's way too quiet out there, even if all clever species blew themselves up.

Happy landing.


I definitely don't agree with all of this -- for one thing, I think the universe is a fundamentally inhospitable place, and the next 50 years isn't going to change that. But his (or her) basic point stands: the future is a very very strange place. Our current ideas about reality are going to be rethought and turned around, only to require rethinking again. Quantum computation will rewrite everything we know about computers, and then nanotechnology will change everything we know about materials. In only a few decades, there will be more computational power in silicon than in the skulls of humanity -- unless we go into intelligence amplification, which is an even surer route to post-humanity.

The future, even the near future, will be more different from the present than the present is from the dawn of the industrial revolution. Happy landing, indeed.
Comments 
16th-Aug-2007 01:45 pm (UTC) - Simulation
I wish I could find the article, but someone did a back of the envelope to figure out how many particles would be required to build a quantum computer, starting from computing principles. It turns out that you need about as much stuff as there is in the universe to calculate the universe.

I have a more fundamental question. Why do you want to calculate someone's mental state? If, for example, you were going to simulate a full table game of hold em, would you take a whole deck, randomly shuffle it three times, then deal to all of the players, then deal the flop, turn, and river? You could easily simulate the same effect by taking an ordered deck, randomly selecting 25 cards, shuffle them once then play the game. By doing so, you cut down you computation. What I'm driving at is what are the effects that you want to simulate?
17th-Aug-2007 08:31 am (UTC) - Re: Simulation
Anonymous
Your shortcut is harder than the simulation itself.

In your example of a hold'em simulator, you use 25 cards instead of an entire deck. If you've programmed anything similar to this before, you would realize that it would take more CPU resources to run the 25 card hold'em game than the 52 card hold'em game. You must have the *potential* for all 52 cards, so at some point, they must be modeled in order to have a standard hold'em game. Later on, when you cut to 25, you must use the CPU in order to remove some of the memory allocated to all 52 cards.

In the case of cutting a 52 card game down to 25, you are likely to be trying to conserve memory at the expense of CPU. If memory/CPU has a conversion from matter/energy in the universe, I would suspect there is some sort of conservation of memory and CPU to matter and energy. Thus, you don't gain anything with a shortcut, but in fact lose some resources because you're just making things more complicated with your alleged shortcut.

Perhaps the post-humans are cleverer programmers than me, or they have vastly superior programming paradigms, languages or whatever. However, from a programmer's standpoint, I am unconvinced that the shortcuts actually are a net savings in computational cost.

Note that games are faster or slower than one another mostly due to how much we can pre-calculate on computers that are NOT running the simulation (ie. the developers' computers and not the customers'). In the most extreme case, such as an animated movie such as Ratatoille, all of the computation occurs ahead of time. The "simulation" is the fixed frames of the movie itself. At higher levels of interaction, more and more computation is put onto the client's computer. However, whether a game runs in real-time or is pre-calculated, those calculations are still within our universe of matter and energy. So offloading that work onto a source that is invisible to the client is not going to save us any matter/energy.

I will make a bold assertion and leave it up to someone else to prove me wrong. I proclaim that the universe is already in its most computationally efficient state. That is, any shortcut to create an illusion, facsimile, or facade of any part of the universe is more expensive in terms of matter and energy than merely having that part be real rather than simulated. This is only true if the observer of the universe and its facsimile has free will and whose actions are unknowable by the creator, but whose actions and capabilities are equal to the creators of the facsimile.

Basically, if the viewers of the simulation have as much power as the creators of the simulation, then it will always require more energy/matter to create the simulation than to create the actual part of the universe that they are simulating.

-Ming
16th-Aug-2007 03:50 pm (UTC)
I have no doubt that we'll have enough power to easily do such a thing if we want to, assuming society doesn't collapse before we get there.

Yeah, it's that kind of logic that escapes me. Taking the above as an axiom seems sort of like cheating to me. Some have projected Moore's law as running into a final brick wall at about 10^100 or so in roughly 600 years. I think that's a bit optimistic, but it is a reasonable upper bound. Is that enough to run a reasonably complete simulation of planet earth, and all observable universal phenomenon? Personally, "I have no doubt that that is not enough computational power to do such a thing". I mean, if we get to pick our axioms...

You don't really need to simulate the molecular behavior at the center of Jupiter until someone starts an experiment that looks at it -- something simpler that looks reasonable is all you need for the vast majority of the simulation.

Yes and no. If they never look at it, then you may never need to simulate it. But if they do, then everything they observe must be consistent with all things observed in the past. From a more "grounded" perspective, you don't have to worry about evolution and fossils and whatnot until people start looking at them, but when they do you'll find that you have to run a complete pre-historic simulation just to get everything in a consistent state. If you are actually worried about things like physical laws, and universal consistency, you'll essentially have to simulate the entirity of all time at some resolution. Unfortunatly, the resolution that you have to simulate at will be determined by the depth of observations that your "sims" eventually make.

You realy do suffer from a form of the chicken and egg problem.

How deep do have to sim?
As deep as the sims eventually look.
How deep will the sims eventually look?
You won't know until you sim them.

It shouldn't take much effort to come up with interesting research goals that could be evaluated with complete simulations.

Not sure exactly what you are thinking of here, but I can't think of one reason to do a complete simulation. Any problem that you might like to investigate would be better served by a partial simulation. The complete simulation route really requires something on the order of unbounded resources, and for what pay off? 42?
16th-Aug-2007 04:08 pm (UTC)
Just to be clear here, I'm not arguing that we aren't in a simulation. Rather that if we are in a simulation, it's probably occuring in a universe that is unlike our own. Which of course means that the original navel gazers conclusion is probably false.
16th-Aug-2007 04:29 pm (UTC) - Maybe you can...
... make more sense out of this than I can.

And just another corollary to this, You don't really need to simulate the molecular behavior at the center of Jupiter until someone starts an experiment that looks at it smacks of "If a tree falls in the forest and nobody is there to hear it, does it make sound?"
16th-Aug-2007 04:47 pm (UTC) - Re: Maybe you can...
You gotta love a paper who's conclusion/discussion includes such eloquent statments as:

Is the universe a computer? The answer to this question depends both on the meaning of ‘computer’ and on the meaning of ‘is.’

Sort of a recursivly ill definded statement, when the first word of your sentence has an undefined meaning.

:)
17th-Aug-2007 05:11 am (UTC)
"Will humans really evolve to the point that they would be willing to expend ludicrous amounts of resources just to achieve what is basically the ultimate game of SimCity... for the hell of it?"

Erm... EverQuest? WoW? Second Life? It's just a question of scale, right?
16th-Aug-2007 11:27 pm (UTC)
Anyway, all that's required for the sim argument is for some researcher or hobbyist to take it up, for any reason at all. I can think of a lot of arguments why no particular group or individual would simulate history, but to claim that no member of humanity will ever do, well, anything, seems like a bit of a stretch.

Wow, that was EXACTLY my point: that no member of humanity will ever do anything.

Seriously, I really don't like having my words distorted to such an extreme, and it doesn't exactly make you look smart either. So please refrain from that. If you'd quoted the whole paragraph I wrote instead of just the last sentence, it would have been much clearer what I meant in context (and harder for you to distort).

But I ask, am I really that off-base for thinking that a flawless simulation of the entire known universe is not something that "some researcher or hobbyist" would be able to create in his basement in a few years, but would rather take a herculean effort of many people over many years to get right? Keeping in mind that each time a bug is found (assuming they have a way to detect it), the simulation needs to be shut down, recoded, and restarted (perhaps from the beginning, depending on the size of the bug).

For a task of this nature, infinitely fast and powerful computers are not enough. Let's say I handed you an infinitely fast computer today. Would that be enough for you to create a simulation of the known universe in your spare time? In your lifetime? How about with 1000 assistants?

Note that simulating the brains and sensory inputs of 7 billion people is just the tip of the iceberg (and even that assumes a near-perfect knowledge of the workings of the human brain). In addition, you need to create flawless simulations of all things that can affect each of those 7 billion humans, such as diseases, natural disasters, and the billions of animals, plants, and other organisms that share the Earth with us and affect our lives. You also need to create flawless simulations of all phenomena that humans can observe (from atoms to supernovas) with detail that stands up to our increasingly powerful instruments and exploratory vessels. And all of these must interact with each other in consistent and correct ways (even when they are not being directly observed), and adhere to all laws of science. To create such a simulation not only requires near-infinite computing power, it requires near-infinite knowledge of the universe.

So sure, we have learned a lot in the past ten thousand years, but fast computers can only help us so much when it comes to creating a complete knowledge of the universe. While computers are good at doing exactly what we tell them to do, they're not so good at taking the initiative to learn new things and make new deductions or even approximations about the laws of the universe all by themselves. Humans have to do that part.

And then there's the question as to whether humans are capable of attaining that level of knowledge and computing power without first destroying ourselves as a species (or knocking ourselves back into a dark age, eliminating most knowledge of the past few thousand years). The way things are going, I'd say nothing is certain. We live in a world where a huge percentage of the population of the planet believes (and even welcomes the idea) that humanity will cease to exist during their lifetimes, and an even larger percentage of humans actively disbelieve in evolution. In this context, what are the chances that our species will survive long enough to achieve scientific knowledge on the scale necessary to create a perfect simulation of the world?

On that note, if we are all running on a computer simulation somewhere, I'd actually be more inclined to believe that some non-human entity or entities somewhere created that computer simulation, rather than believing that future humans did.
17th-Aug-2007 12:17 am (UTC)
Seriously, I really don't like having my words distorted to such an extreme, and it doesn't exactly make you look smart either. So please refrain from that. If you'd quoted the whole paragraph I wrote instead of just the last sentence, it would have been much clearer what I meant in context (and harder for you to distort).

Apparently I wasn't clear in what I meant. I didn't mean that you were claiming that no person ever would run a simulation. But I thought that you and others failed to understand that arguments against future humans running simulations really did need to claim that nobody would ever do it, because just one person doing it is all the simulation argument needs.

In other words, I don't think you said something obviously dumb. I'm saying that the perfectly reasonable thing you did say misses the point.

But I ask, am I really that off-base for thinking that a flawless simulation of the entire known universe is not something that "some researcher or hobbyist" would be able to create in his basement in a few years, but would rather take a herculean effort of many people over many years to get right?

This is a fundamental disagreement between me and my commenters, I think. I don't think you need to simulate the entire known universe to any level of detail. You need to simulate only and exactly enough so that people see a consistent view of the universe. I think you can do that without simulating the entire thing, and without simulating even most of what happens right here!

For instance, if I open a box in the garage I haven't looked in for years, there's a chance that spiders and bugs will be in there. But there's no reason to actually generate the spiders until I look in, right? I don't think most people would notice if the garage ecosystem wasn't completely perfect, and for the few that would notice or care, you can simulate it just for them.

On that note, if we are all running on a computer simulation somewhere, I'd actually be more inclined to believe that some non-human entity or entities somewhere created that computer simulation, rather than believing that future humans did.

I think that post-humans are in fact "non-human entities" for any current meaning of human.
17th-Aug-2007 02:59 am (UTC)
This is a fundamental disagreement between me and my commenters, I think. I don't think you need to simulate the entire known universe to any level of detail. You need to simulate only and exactly enough so that people see a consistent view of the universe. I think you can do that without simulating the entire thing, and without simulating even most of what happens right here!

I agree this is the fundamental disagreement, but I think your version breaks down upon closer analysis and becomes our version.

You're saying that you don't actually need to simulate every observable object or phenomenon in the universe except for when it's actively being observed by a human. If that's the case, in order to create a consistent world, you still need to somehow keep track of what it was they observed, so that the next time they see it, or another human being sees it, it will match expectations. And for that matter, you have to keep track of all human recording devices and observational instruments as well, for the same reason that at any given point, a human could look at one of these and would have to see something that made sense. Plus, you of course need to create behavioral models for how various objects or phenomena will behave, how they will interact with their surroundings (and other objects and phenomena), and how they will change/evolve/decay over time. Is this still sounding easy?

So let's say a human measures the effect of a butterfly's wings flapping, do you have to simulate whatever chaotic effects that might cause, up to and including causing a cyclone on a distant continent? And if that cyclone then hits an island and causes a tree to fall in a forest where there is no human to observe it, does it make a sound? What if the sound of the falling tree startles a monkey, who bites another monkey, infecting it with a virus in the process, and the virus later mutates in the new monkey, and becomes HIV. How much of this process needs to be simulated, and how much can we ignore?

In short, if you believe that the universe is a huge interconnected chaotic system (which I do), with all sorts of objects and phenomena constantly and independently interacting with and affecting one another, it's pretty much impossible to create a simulation of this system that only consists of human observations, since we are constantly being affected by things we have not directly observed.
17th-Aug-2007 08:50 am (UTC) - Yes, my thoughts exactly...
Anonymous
Yes, Nat echoed my point. In fact, if there were any mistake in your simulation, I would immediately believe in "intelligent design"--- which would be a rather poor simulation of human evolution if others like me changed beliefs en masse.

Basically, your "just in time" observer illusion is a similar argument to intelligent design. That is, if someone finds a dinosaur bone and we carbon date it, or maybe we discover even newer ways to understand it, yet all of that information remains consistent, not only with that single stupid dinosaur bone, but all of science and its observations up to that point, then which is more likely: 1) Until observed, that bone did not exist. Instead, Someone placed that bone there for you to find and come up with those conclusions "just in time" or 2) It was sitting there and following the rules of the universe that we puny humans are uncovering very slowly. And with each test thoughout the eons of humanity, that bone answers the same stupid questions that we humans like to pose and that the only thing that ever changes is our human knowledge of the universe and not the universe itself.

Not only which is more likely, but which is more computationally efficient? As in my other post, I contend that it is simply more efficient to have reality than to calculate some illusion of reality or to predict when some human with free will would ask some stupid question you might not have considered.

The latter is the bane of game programmers.

-Ming
17th-Aug-2007 09:01 am (UTC) - Yes, my thoughts exactly...
Anonymous
Yes, Nat echoed my point. In fact, if there were any mistake in your simulation, I would immediately believe in "intelligent design"--- which would be a rather poor simulation of human evolution if others like me changed beliefs en masse.

Basically, your "just in time" observer illusion is a similar argument to intelligent design. That is, if someone finds a dinosaur bone and we carbon date it, or maybe we discover even newer ways to understand it, yet all of that information remains consistent, not only with that single stupid dinosaur bone, but all of science and its observations up to that point, then which is more likely: 1) Until observed, that bone did not exist. Instead, Someone placed that bone there for you to find and come up with those conclusions "just in time" or 2) It was sitting there and following the rules of the universe that we puny humans are uncovering very slowly. And with each test thoughout the eons of humanity, that bone answers the same stupid questions that we humans like to pose and that the only thing that ever changes is our human knowledge of the universe and not the universe itself.

Not only which is more likely, but which is more computationally efficient? As in my other post, I contend that it is simply more efficient to have reality than to calculate some illusion of reality or to predict when some human with free will would ask some stupid question you might not have considered.

The latter is the bane of game programmers.

-Ming
17th-Aug-2007 03:02 am (UTC)
This is a fundamental disagreement between me and my commenters, I think. I don't think you need to simulate the entire known universe to any level of detail. You need to simulate only and exactly enough so that people see a consistent view of the universe.

Herein lies your paradox. Since your sims will be able to potentially view any subset of the universe which we can, you must simulate any level of detail which they can view.

In simple terms; you can't simulate the universe without simulating the universe. And you can't simulate the universe from within the same universe, by simple recursion.
17th-Aug-2007 07:58 am (UTC) - Observer fallacy
Anonymous
I agree with prock on this.

It is not enough to simulate the sensory input of the sims as they happen. As a game programmer, we continually worry about the potentially visible set and potential actions of our game user. If the user *may* do something, it must be there and programmed for it to be experienced. That is, if our game is to have ragdoll physics, then it someone must program it before-hand in order for it to occur when that situation arises.

In your example, what happens inside of Jupiter only needs to be modeled when someone observes it. But if you give your sims free will, how will you know when someone observes it? By what rules should it behave? And how deep will they want to look into it? An annoying trait of human curiosity for "world designers" is that people immediately want to investigate inconsistencies in order to learn "the truth". When does the reality end and the cheaper computational facade begin?

Furthermore, there are events which require more than an observer to be modeled. The events that require modelling go far beyond what the sim observers witness. Beginning from your conception, the physics of the sperm and fertilization of the egg need to be modeled if genetics is to be a proper simulation of our own earthly evolution. Which strands of DNA you receive from your father and which from your mother need to be modeled in order to determine to a very large degree how you will turn out. Are you saying that these things, the cornerstones of evolution, should be abstracted away when no one is looking? And that somehow, you know when someone is looking and not looking? That we could reduce DNA errors, evolution, cosmic rays damaging DNA as mothers age, and all the rest as a probability function? And are we certain that the processing power used to determine when you have an observer or not in ALL of these varied cases is actually cheaper than doing the computation in the first place?

A simulation like the Simnasium baseball game that you and Mike Lewis played (and I still play) is very static. The virtual ball players may have new accomplishments such as break the home run record or strike out record. However, the players themselves will never create a new baseball strategy or find a way to cheat or invent the gyroball pitch. There is nothing in that simulation which the programmers did not explicitly put in. There is no discovery, no invention, no curiosity, and no free will.

If we do not have free will, then I will concede that we could be in a simulation. However, that is a different philosophical debate. Assuming that we do have free will, I would argue that no designer could abstract any meaningful amount of the universe away if we must model what each observer may potentially investigate. If we take all of the energy of the next galaxy to simulate our own galaxy, who is to say that some future human won't wander over there is a space ship? If you say that you design the rules of the world so that the sims may never go there (such as a speed of light limit), then either we do not entirely have free will or else our universe is not an accurate simulation of the original universe.

If the actual world is vastly different than our own world, then I will concede that we could be in a simulation. Simply, if the rules of energy and information are different or simply not conserved as ours are, then all bets are off.

Anyway, I'm just trying to illuminate what my position is. I'm curious what your thoughts on this are, because we seem to disagree on point #2.

-Ming
17th-Aug-2007 05:12 am (UTC)
Keeping in mind that each time a bug is found (assuming they have a way to detect it), the simulation needs to be shut down, recoded, and restarted (perhaps from the beginning, depending on the size of the bug).

If you're in the simulation, you have no memory of that across runs. The people existing in the current execution have no memory of when they were in the previous one. Like reincarnation with no memory of past lives :-)

I'm too tired to really think about this right now but it seems like there could be some tie-in to at least the weak anthropomorphic principle here.
24th-Aug-2007 08:08 pm (UTC) - Moore's law on computing power
Anonymous
...is flawed due to static analysis. http://en.wikipedia.org/wiki/Static_analysis

This page was loaded Nov 22nd 2017, 7:55 pm GMT.