I don't think you understand the amounts of energy that are involved. If you are content with a crappy simulation, sure, then you can pay it out of your electricity bill. Ultimately though, high fidelity simulation of any system will require the same amount of energy as the information content of that system.
If the system you want to simulate is e.g. the solar system, my back-of-the -envelope calculation puts your energy consumption at 10^50 GWh (give or take a few orders of magnitude). Scaling back to just simulating Earth still leaved more gigawatt-hours than I can afford!
You're making the assumption that energy efficiency of computing systems will not also continue to increase while computing power also increases. I don't have time to look it up right now, but I'm pretty sure that an old Apple 2 used more electricity per unit of computational power than a Core 2 Duo. Also, a 500MB hard drive from back in the day uses roughly the same power as a 1.5TB drive today. At least it has the same power connector.
When you look at the kinds of things Google is doing to save energy in its data centers, I don't expect this to be a problem int he long-term.
The most powerful computers today are also some of the most efficient. Computing power requirements will be going down even as their computational abilitiy increases.
Well, there is the simulation paradox. If you can simulate something accurately, then you have effectively created said something. If your simulation uses less energy then the original, then you have discovered something amazing.
If you can simulate something accurately, then you have effectively created said something. If your simulation uses less energy then the original, then you have discovered something amazing.
I don't know if this is necessarily true. I mean, what of compression? Some things are incredibly complex, but often redundantly complex. That is to say they have many repeating and similar patterns.
In a simulation we represent a real world object with a model of said object. To perfectly simulate a tree, we would need to simluate carbon atoms. Rather than use the energy required to construct a hojillion carbon atoms, we would need to simulate just one carbon atom. With enough capacity, we could even store every possible behavior of a carbon atom in a database. Then we would never have to expend the energy to simulate one ever again, and could reduce the cost of further simulations to the cost of a database query.
All this, of course, assumes a deterministic finite universe. In a non-deterministic universe, accurate simulation is impossible given our deterministic computing methodologies.
If you can simulate something accurately, then you have effectively created said something.
I posit that in order to have effectively created something, it would have to obtain the ability to interact with the outside world in a manner that is almost indistinguishable from the original. Thus, the power requirements would have to nearly match the original.
I don't know if this is necessarily true. I mean, what of compression? Some things are incredibly complex, but often redundantly complex.
If the redundancy can have no effect in any situation, then it is unnecessary.
In a simulation we represent a real world object with a model of said object. To perfectly simulate a tree, we would need to simluate carbon atoms. Rather than use the energy required to construct a hojillion carbon atoms, we would need to simulate just one carbon atom. With enough capacity, we could even store every possible behavior of a carbon atom in a database. Then we would never have to expend the energy to simulate one ever again, and could reduce the cost of further simulations to the cost of a database query.
But the lookup costs of accessing this increasing dataset increase as the size increases. The only way around this is quantum computing, but you can see where that leads us.
Theoretically, but with significantly large data storage devices, you are going to start running into the I/O bottleneck issue.
We're already talking about computer technology that may well be in the realm of fantasy. If we have CPUs so powerful as to efficiently simulate individual atoms, why would we also have an I/O bottleneck? Also, we're assuming a computing model that takes our current architecture of processing and storage and increases its efficiency and power exponentially. More like there will be a completely new architecture where even the concept of I/O itself might be irrelevant.
More like there will be a completely new architecture where even the concept of I/O itself might be irrelevant.
You're talking a whole new model of computing. Even Turing machines and finite state automata have I/O. I'm not sure if you could even call it a computer if it didn't have I/O.
You're making the assumption that energy efficiency of computing systems will not also continue to increase while computing power also increases.
No Scott, as I said this is the energy necessary just for creating/arranging the bits of information assuming the process doesn't cost any energy otherwise. Information == energy.
It's all theoretical anyway. After all, we're assuming that perfect simulation is possible, which may not be the case.
The possibility of a perfect simulation (even at parity cost in energy with respect to the original thing) is the easier option regarding of your original argument. If it is not possible to make these theoretical media experiences / private universes perfect this means that you will likely spend resources at an exponential rate as you approach perfection. Effectively you will go way way past the energy parity with the simulation vs the real thing.
It is possible, however, to have procedural content generation that, while not perfect, is indistinguishable from human-created content, and which uses less total energy or time to create. Will artists one day have to compete in their rates with the electric bills of writeblockbustermovied?
No Scott, as I said this is the energy necessary just for creating/arranging the bits of information assuming the process doesn't cost any energy otherwise. Information == energy.
I don't get it. We use to have computers that used a whole bunch of energy to create and arrange just a few bits. Now computers create and arrange many more bits with far less energy. Also, matter contains a fuck ton of energy, far more energy than is required to flip a transistor on and off. So how could we not create a simulation of matter, in the form of flipping transistors that requires less energy than the matter itself?
It is possible, however, to have procedural content generation that, while not perfect, is indistinguishable from human-created content, and which uses less total energy or time to create. Will artists one day have to compete in their rates with the electric bills of writeblockbustermovied?
The artists will be the ones who create and maintain blockbustermovied itself. If blockbustermovied is itself intelligent enough to be considered a sentient being, then it is the artist. If it is an artist, I wouldn't worry too much about competing with it. Despite Elvis and the Beatles being #1, other people still sold records.
I don't get it. We use to have computers that used a whole bunch of energy to create and arrange just a few bits. Now computers create and arrange many more bits with far less energy. Also, matter contains a fuck ton of energy, far more energy than is required to flip a transistor on and off. So how could we not create a simulation of matter, in the form of flipping transistors that requires less energy than the matter itself?
This is a freshman physics fallacy of not understanding the abso-fucking-lutely HUUUUGE numbers involved. Sure flipping bits is cheap, but even if it is done for free at the transistor level there is a cost in the difference of entropy between before and after. In that sense a laptop with just zeroes on the hard drive actually for reals weighs less than a laptop full of information.
So the only question is, how much information is there in e.g. the solar system and how much energy does that correspond to? My quick calculation showed this to be 10^50 Jiggawatt-hours or thereabouts. As for the fuck ton of energy stored in matter, the c^2 in E=mc^2 gives you a rough conversion coefficient of 10^17 which doesn't help a lot. It takes 10^50 GWh to about 10^42 kg of matter. Earth weighs 10^25 kg, so that's 100000000000000000 Earths (give or take a few zeroes).
Edit: If there is someone on the forums versed in statistical entropy and information theory, they may care to check that i'm not completely talking out of my ass here.
In that sense a laptop with just zeroes on the hard drive actually for reals weighs less than a laptop full of information.
Do you mean this literally? I don't know about the other stuff, but this I find hard to believe. Any "information" that is contained in the zeros and ones of a hard drive is placed there by our understanding of it. There is no more or less information there at any time, just different information. A magnetic hard drive works by magnetizing or demagnetizing different parts of the drive. We the detect which parts are magnetized, and call those parts 1s and the other parts we call 0s.
If it weighs less when it is all zeros, it is because mass is added when the thing is magnetized. I'm guessing that's in the form of added electrons. We could, if we so chose, make a hard drive where the demagnetized areas counted as 1 and the magnetized areas counted as 0. Thus, a drive full of 0 would be completely magnetized and weigh more than a drive full of 1 which was completely demagnetized.
Even so, the quantity of information stored has little relation to the number of 1s and 0s on the drive. All 1s is the same amount of information as all 0s. A terabyte of information is still a terabyte whether it's all 1s or all 0s.
Also, techniques of lossless compression allow us to store the same quantity of information in a smaller number of ones and zeros. You know, zip files. Zip files work by replacing repeating patterns of data with shorter patterns, and then reversing the process at time of decompression. It does take CPU power to perform this compression and decompression, so total energy savings is not as big as it seems at first. However, if we're talking about simulating the natural world, I think the same principle would apply. There are lots and lots of similar repeating patterns in nature. Really, that's what science is all about, finding those patterns. By applying these sorts of techniques in our simulation, we can save energy by "compressing" these patterns and thus possibly create accurate simulations with less energy than the real deal.
I think you're getting "information" wrong here. All 0's and all 1's have the same (minimum) amount of information: there is only one way to arrange a string of N bits of 0's or 1's. When you have N/2 0's and N/2 1's, that is the maximum information: there are (N choose N/2) ways to arrange those N bits.
If it weighs less when it is all zeros, it is because mass is added when the thing is magnetized. I'm guessing that's in the form of added electrons. We could, if we so chose, make a hard drive where the demagnetized areas counted as 1 and the magnetized areas counted as 0. Thus, a drive full of 0 would be completely magnetized and weigh more than a drive full of 1 which was completely demagnetized.
Isn't a hard drive sealed? Where is this extra mass coming from if not from something already present within the drive casing? The idea that a sealed drive would change in weight due to data being written or removed does not make sense. It may be that the platter itself gains or loses mass but the device as a whole should not be losing or gaining mass.
Isn't a hard drive sealed? Where is this extra mass coming from if not from something already present within the drive casing? The idea that a sealed drive would change in weight due to data being written or removed does not make sense. It may be that the platter itself gains or loses mass but the device as a whole should not be losing or gaining mass.
I think the issue can be succinctly summed up as such:
It will, by definition, require more energy to wholly describe the state of something than is contained in said thing.
Consider that anything but the actual information itself is extra metadata, requiring extra energy. Even a nomenclature requires information, and therefore energy, beyond what is contained in the thing. The only way to efficiently describe something is to do so with an inherent schema consisting only of the object in question. This can otherwise be called the object itself.
Is that crazy? It's as close to defining it as I can get.
Even if no electrons enter or leave, it is possible that it could lose or gain mass if one of the states is at higher energy than the other. E=mc2.
So magnetized vs. non-magnetized. Still, it does not relate to how much information is stored. Also, there are other types of data storage besides magnetic disks.
Let's say I have 8 coins on a table. We decide that heads is zero and tails is 1. I have 8 bits of information in this storage device. That's one byte. It does take energy to flip the coins. However, the mass of the coins does not change no matter what the configuration. The amount of information also does not change unless I add or remove coins. The addition of mass by my hand oils rubbing off onto the coins or dust landing on them does not count in the same way that physics teachers never make you count air resistance when you calculate how long it takes the ball to roll down the ramp.
However, the mass of the coins does not change no matter what the configuration.
Okay, when you say configuration, you mean the string of 1's and 0's, right? When I say configuration, I mean how the coin is physically positioned with respect to the Earth. That does change their mass. A coin at sea level has less mass than a coin 1000 miles up. It is a minuscule difference, but real. Whatever you define your 1 and 0 to be, if they are at different energy levels, that will correspond to different masses.
I said it is possible a hard drive could gain or lose mass without any electrons entering or leaving. I don't know enough about the details of how hard drives work to tell you for sure. It might depend on the orientation of the hard drive with respect to the Earth's magnetic field as well. In any case, it would depend on the energy levels of the states, and not on the information contained therein. That is, all 1's would weigh the most if 1 had a higher energy than 0, regardless of how much "information" is on the drive.
I realize in my earlier post I was conflating entropy and information. What I said holds regarding entropy, but I'm not sure about information.
That is, all 1's would weigh the most if 1 had a higher energy than 0, regardless of how much "information" is on the drive.
You'd require energy to change any of those bits, or to create a nomenclature to make them useful. Unless you change them or create a nomenclature for understanding them, you have not actually stored any information.
Again, irrespective of any information that may or may not be stored, it is possible that the hard drive could change mass. If 1 (for example) is at a higher energy than 0, a drive full of 1's will weigh more than a drive full of 0's. Literally. It has more mass. That is all I am saying. Anything else is Timo.
When I say configuration, I mean how the coin is physically positioned with respect to the Earth. That does change their mass. A coin at sea level has less mass than a coin 1000 miles up.
Uh, you are confusing mass and weight. You have the same mass no matter where you are in the universe, but your weight is relative to gravity. They are completely different.
Uh, you are confusing mass and weight. You have the same mass no matter where you are in the universe, but your weight is relative to gravity. They are completely different.
Silly Scott, I do not mean weight. I mean mass.
At an infinite distance from the Earth (pretend there is nothing in the universe besides you and Earth), your body has some given mass. You also have 0 potential energy. As you move towards the Earth, your potential energy goes negative. This energy gets released, in the form of you losing mass, because energy is conserved. A carbon-12 atom has less mass than 6 protons + 6 neutrons + 12 electrons. Science.
Comments
When you look at the kinds of things Google is doing to save energy in its data centers, I don't expect this to be a problem int he long-term.
In a simulation we represent a real world object with a model of said object. To perfectly simulate a tree, we would need to simluate carbon atoms. Rather than use the energy required to construct a hojillion carbon atoms, we would need to simulate just one carbon atom. With enough capacity, we could even store every possible behavior of a carbon atom in a database. Then we would never have to expend the energy to simulate one ever again, and could reduce the cost of further simulations to the cost of a database query.
All this, of course, assumes a deterministic finite universe. In a non-deterministic universe, accurate simulation is impossible given our deterministic computing methodologies.
So the only question is, how much information is there in e.g. the solar system and how much energy does that correspond to? My quick calculation showed this to be 10^50 Jiggawatt-hours or thereabouts. As for the fuck ton of energy stored in matter, the c^2 in E=mc^2 gives you a rough conversion coefficient of 10^17 which doesn't help a lot. It takes 10^50 GWh to about 10^42 kg of matter. Earth weighs 10^25 kg, so that's 100000000000000000 Earths (give or take a few zeroes).
Edit: If there is someone on the forums versed in statistical entropy and information theory, they may care to check that i'm not completely talking out of my ass here.
If it weighs less when it is all zeros, it is because mass is added when the thing is magnetized. I'm guessing that's in the form of added electrons. We could, if we so chose, make a hard drive where the demagnetized areas counted as 1 and the magnetized areas counted as 0. Thus, a drive full of 0 would be completely magnetized and weigh more than a drive full of 1 which was completely demagnetized.
Even so, the quantity of information stored has little relation to the number of 1s and 0s on the drive. All 1s is the same amount of information as all 0s. A terabyte of information is still a terabyte whether it's all 1s or all 0s.
Also, techniques of lossless compression allow us to store the same quantity of information in a smaller number of ones and zeros. You know, zip files. Zip files work by replacing repeating patterns of data with shorter patterns, and then reversing the process at time of decompression. It does take CPU power to perform this compression and decompression, so total energy savings is not as big as it seems at first. However, if we're talking about simulating the natural world, I think the same principle would apply. There are lots and lots of similar repeating patterns in nature. Really, that's what science is all about, finding those patterns. By applying these sorts of techniques in our simulation, we can save energy by "compressing" these patterns and thus possibly create accurate simulations with less energy than the real deal.
Maybe?
Fixed my math.
Is that crazy? It's as close to defining it as I can get.
Let's say I have 8 coins on a table. We decide that heads is zero and tails is 1. I have 8 bits of information in this storage device. That's one byte. It does take energy to flip the coins. However, the mass of the coins does not change no matter what the configuration. The amount of information also does not change unless I add or remove coins. The addition of mass by my hand oils rubbing off onto the coins or dust landing on them does not count in the same way that physics teachers never make you count air resistance when you calculate how long it takes the ball to roll down the ramp.
I said it is possible a hard drive could gain or lose mass without any electrons entering or leaving. I don't know enough about the details of how hard drives work to tell you for sure. It might depend on the orientation of the hard drive with respect to the Earth's magnetic field as well. In any case, it would depend on the energy levels of the states, and not on the information contained therein. That is, all 1's would weigh the most if 1 had a higher energy than 0, regardless of how much "information" is on the drive.
I realize in my earlier post I was conflating entropy and information. What I said holds regarding entropy, but I'm not sure about information.
At an infinite distance from the Earth (pretend there is nothing in the universe besides you and Earth), your body has some given mass. You also have 0 potential energy. As you move towards the Earth, your potential energy goes negative. This energy gets released, in the form of you losing mass, because energy is conserved. A carbon-12 atom has less mass than 6 protons + 6 neutrons + 12 electrons. Science.