3 ms·

I suppose it could in theory, but it probably wouldn't do so very well. In general, you can't simulate physics on physics faster than physics, so non-approximat

I suppose it could in theory, but it probably wouldn't do so very well. In general, you can't simulate physics on physics faster than physics, so non-approximate simulation of the universe would probably run several orders of magnitude slower than the universe does. You would have to approximate massively, but that means a lot of things that are physically possible in the universe would become impossible in the simulation, especially systems that require a lot of precision and optimal behavior such as computers. Honestly, even if you simplified everything to the max except for conscious agents in the simulation, it's not even clear to me that we could simulate human brains faster than actual human brains.

vidarh 3y ago
> In general, you can't simulate physics on physics faster than physics You assume that the physics of the simulation are equivalent to the physics outside it. We simulate simpler environments than our own all the time. Maybe the constraints of our universe are the way they are because it reflects the tradeoffs required to be able to simulate it given the conditions the simulation runs in.
- breuleux 3y ago
  Maybe, but that's not interesting. It is possible that the physics inside are different from the physics outside, but that tells us absolutely nothing of value about whether we are in a simulation or not. If we do assume that an observer cannot differentiate the simulation from the outside world, then we can make an interesting argument: if the rules of physics in some universe entails that this universe will simulate many other universes with the same apparent rules, then an observer of these rules is probably simulated regardless of what the root universe happens to be. This is the only way you can ground the probability calculation. A lot of things are possible. Not many things are probable.
  - vidarh 3y ago
    For an observer to be unable to differentiate the simulation from the outside world if put in the simulation is a very low bar, and the one that matters. It's a low bar because you don't need to presume the simulated entity gets the benefit of the passage of time. For it to be likely for us to be in a simulation, it's sufficient for it to be likely for the total runtime of "simulated us" to be higher than "real us", and our only knowledge of the passage of time are the momentary memories and sensory inputs right now, and we can't validate their truthfulness. In other words: you don't need to simulate any physics to prevent a mind from differentiating. You just need to limit the scope of your simulation to a sufficiently narrow slice of time. Personally, I tend to think that if simulations become widespread, test runs, debugging and development, research runs, or application specific runs of small slices are likely to far outpace full scale simulations in runtime, and so if simulations are probable, I'd be inclined to think we're most likely in a (very) short-lived, scope-limited simulation in a horrible groundhog day like setting.
    - breuleux 3y ago
      > In other words: you don't need to simulate any physics to prevent a mind from differentiating. But if your simulator is not simulating physics... what is it simulating? The mind? If you are simulating a human brain, that IS a physics simulation, and it is probably on the harder side of the scale, so we cannot readily assume it's going to be easy or fast. And if you are running some different kind of mind based on an abstract neural network or something like that, well, it's not a simulated mind, it's just a mind.
FeepingCreature 3y ago
When I think about entities simulating our universe, I certainly do not assume that they're inside our universe at the time! That would be very difficult, though it may work if you assume that most of the universe at the moment is computationally "dead" and can be optimized out. But IMO, this sort of theory makes a lot more sense in a "large" cosmology like Tegmark IV.
- breuleux 3y ago
  > I certainly do not assume that they're inside our universe at the time! I'm not sure what you mean by that, but if the simulating universe is not qualitatively identical to the simulated universe, you lose the mathematical properties that make the simulation argument interesting. > it may work if you assume that most of the universe at the moment is computationally "dead" and can be optimized out. Our universe is chaotic and is largely defined by its edge cases sometimes leading to a global phase change (e.g. evolution). If one wishes to preserve these properties, I would argue it is not optimizable at all. If you are willing to lose these edge cases, maybe you can optimize certain things away, but it's quite possible you will lose more than you bargain for (basically nothing is computationally inert in a chaotic system). Your system will be bottlenecked by the most demanding parts of the simulation anyway. > But IMO, this sort of theory makes a lot more sense in a "large" cosmology like Tegmark IV. I tried to work it out mathematically a while ago, but I don't think it really works out (I forget the details, I should dig it up sometime). If the infinity of possible mathematical structures actually exist, there's still an infinity of possible mathematical structures to simulate. And if, to simplify, 1% of mathematical structures are conformant with our universe, by default you would expect 1% of simulations (across the multiverse) to be simulations of our universe (unless you think we are somehow special and other universes specifically want to simulate us). So you would need over 50% of the multiverse to be simulation fodder in order for a given observer to probably be in a simulation, and that seems... kinda really high?

vidarh 3y ago

> In general, you can't simulate physics on physics faster than physics You assume that the physics of the simulation are equivalent to the physics outside it. We simulate simpler environments than our own all the time. Maybe the constraints of our universe are the way they are because it reflects the tradeoffs required to be able to simulate it given the conditions the simulation runs in.

breuleux 3y ago
Maybe, but that's not interesting. It is possible that the physics inside are different from the physics outside, but that tells us absolutely nothing of value about whether we are in a simulation or not. If we do assume that an observer cannot differentiate the simulation from the outside world, then we can make an interesting argument: if the rules of physics in some universe entails that this universe will simulate many other universes with the same apparent rules, then an observer of these rules is probably simulated regardless of what the root universe happens to be. This is the only way you can ground the probability calculation. A lot of things are possible. Not many things are probable.
- vidarh 3y ago
  For an observer to be unable to differentiate the simulation from the outside world if put in the simulation is a very low bar, and the one that matters. It's a low bar because you don't need to presume the simulated entity gets the benefit of the passage of time. For it to be likely for us to be in a simulation, it's sufficient for it to be likely for the total runtime of "simulated us" to be higher than "real us", and our only knowledge of the passage of time are the momentary memories and sensory inputs right now, and we can't validate their truthfulness. In other words: you don't need to simulate any physics to prevent a mind from differentiating. You just need to limit the scope of your simulation to a sufficiently narrow slice of time. Personally, I tend to think that if simulations become widespread, test runs, debugging and development, research runs, or application specific runs of small slices are likely to far outpace full scale simulations in runtime, and so if simulations are probable, I'd be inclined to think we're most likely in a (very) short-lived, scope-limited simulation in a horrible groundhog day like setting.
  - breuleux 3y ago
    > In other words: you don't need to simulate any physics to prevent a mind from differentiating. But if your simulator is not simulating physics... what is it simulating? The mind? If you are simulating a human brain, that IS a physics simulation, and it is probably on the harder side of the scale, so we cannot readily assume it's going to be easy or fast. And if you are running some different kind of mind based on an abstract neural network or something like that, well, it's not a simulated mind, it's just a mind.

FeepingCreature 3y ago

When I think about entities simulating our universe, I certainly do not assume that they're inside our universe at the time! That would be very difficult, though it may work if you assume that most of the universe at the moment is computationally "dead" and can be optimized out. But IMO, this sort of theory makes a lot more sense in a "large" cosmology like Tegmark IV.

breuleux 3y ago
> I certainly do not assume that they're inside our universe at the time! I'm not sure what you mean by that, but if the simulating universe is not qualitatively identical to the simulated universe, you lose the mathematical properties that make the simulation argument interesting. > it may work if you assume that most of the universe at the moment is computationally "dead" and can be optimized out. Our universe is chaotic and is largely defined by its edge cases sometimes leading to a global phase change (e.g. evolution). If one wishes to preserve these properties, I would argue it is not optimizable at all. If you are willing to lose these edge cases, maybe you can optimize certain things away, but it's quite possible you will lose more than you bargain for (basically nothing is computationally inert in a chaotic system). Your system will be bottlenecked by the most demanding parts of the simulation anyway. > But IMO, this sort of theory makes a lot more sense in a "large" cosmology like Tegmark IV. I tried to work it out mathematically a while ago, but I don't think it really works out (I forget the details, I should dig it up sometime). If the infinity of possible mathematical structures actually exist, there's still an infinity of possible mathematical structures to simulate. And if, to simplify, 1% of mathematical structures are conformant with our universe, by default you would expect 1% of simulations (across the multiverse) to be simulations of our universe (unless you think we are somehow special and other universes specifically want to simulate us). So you would need over 50% of the multiverse to be simulation fodder in order for a given observer to probably be in a simulation, and that seems... kinda really high?