41 ms·

Very interesting to think about. Here's my interpretation of the argument: Suppose you have some system X whose dynamics are those of conscious agents in a worl

Very interesting to think about. Here's my interpretation of the argument: Suppose you have some system X whose dynamics are those of conscious agents in a world. For example, maybe X is the real world. Suppose that the agents in X really experience consciousness. Now suppose we apply some one-to-one function f to X, yielding Y = f(X). Then Y contains all the same information as X. So who's to say that the agents in X have consciousness but the agents in Y don't? Now by an inverse argument, suppose you randomly encounter some blob of random dust, Z. Then you can always come up with a one-to-one function g mapping Z to X, X = g(Z). So apparently the dust Z already contained all the conscious agents in the universe. Conclusion: The conscious experiences of all agents are present in all sufficiently complex blobs of random dust and all sufficiently complex networks of mathematical relations. One issue with this argument that hasn't been brought up yet is Kolmogorov complexity. Must random blobs of dust Z will be algorithmically random, not containing any real structure. In that case, the complexity of the function g that decodes Z into X will have to be huge. Essentially, the decoding function g itself will have to contain all the information that eventually ends up in X. So the information about conscious agents isn't really in the dust Z, it's in the decoder g.

ajconway 6y ago
If you believe in mathematical realism than the fact that there surely exists a mapping between a random set of data and a simulation with conscious agents in it, that should be enough.
- chr1 6y ago
  It's not enough if the mapping function does not use the information from the random set. For instance it is easy to construct a mapping from ordered digits of pi to ordered letters of your comment, but all of the information of your comment will be in the mapping.
  - canjobear 6y ago
    Exactly. The mapping function is almost always essentially a lookup table where you look up a dust configuration and get out a real-world configuration. The fact that you can construct this mapping doesn't mean much IMO.
  - ajconway 6y ago
    I believe the argument goes like this: assume that conscious agents can be simulated (that is, consciousness is computable). What is computation? If sudden wind arranges leaves on the ground into a pattern that looks like “2 + 2 = 4”, was there a computation? We would usually answer “no”, as computation, like many other things, requires causal connection. But Greg Egan states that “there is nothing more to causality than the correlations between states”. Now, a deterministic simulation can be represented as a series of states recorded on a medium. Then you, an observer, can rewind and inspect what’s going on there, similar to reading a book or watching a movie. But the inhabitants of the simulation couldn’t care less about you observing their states. They already “exist”. In that case, why do we need the states of the simulation be recorded on the “film” in the order that makes sense for an outside observer? Surely a random arrangement would not affect the “simulation”. The same goes for the encoding. Now, if the arrangement doesn’t matter, and the encoding doesn’t matter, why there needs to be a specific physical medium at all? Surely somewhere in the Universe you can find arrangements of “dust” that “encode” the same states (in no particular order).
    - chr1 6y ago
      What does “encode” mean here? If it is supposed to contain the whole state in one chunk, the space of possible deterministic algorithms is too huge compared to the amount of dust, so it can't contain that. If it is supposed to find small chunks and rearrange them, it will not be using any information from the dust, and with same success encoder could use just any set of 0s and 1s. To circumvent finiteness of dust One could use digits of an irrational number or simple counting, but that is a very strange definition of existence. By that logic every possible history is just a number, so everything exists. The Important question is how to construct these numbers, find them in the sea of other meaningless numbers. I think Stephen Wolfram's idea of computational irreducibility is very useful here. It says that most computations (even very simple systems like some of cellular automata) cannot be replaced by other computations, and even to verify that a number represents result of a computation, one have to carry out that same computation. So the part of the book where states of the simulation are computed out of order is simply wrong, the latter state of the simulation could not have been computed without computing the intermediate states, and all the methods of computing that state were equivalent to the person living through that interval of time.
- mcguire 6y ago
  Yours may be the finest argument for constructivism I have seen.
Schiphol 6y ago
I think the idea is thst you get to pick and choose from the 'dust': if you have a perfectly Kolmogorov-random string, it's still trivial to go jumping from one bit here to another it there like so: 10101010..., or in any other, however complicated pattern.
saeranv 6y ago
> One issue with this argument that hasn't been brought up yet is Kolmogorov complexity. Must random blobs of dust Z will be algorithmically random, not containing any real structure. This is my issue as well, and something Greg Egan brings up in his FAQ on Dust theory as well. If my state is simply the result of a sampling of domain Y, why is my continuous state so stable? If we're assuming I'm sampling from a uniform (maximally random) distribution of potential states, there are many more state configurations where (for example) the physical laws should break down, then a stable configuration. So there must be some missing element in Dust theory that explains away this randomness.

ajconway 6y ago

If you believe in mathematical realism than the fact that there surely exists a mapping between a random set of data and a simulation with conscious agents in it, that should be enough.

chr1 6y ago
It's not enough if the mapping function does not use the information from the random set. For instance it is easy to construct a mapping from ordered digits of pi to ordered letters of your comment, but all of the information of your comment will be in the mapping.
- canjobear 6y ago
  Exactly. The mapping function is almost always essentially a lookup table where you look up a dust configuration and get out a real-world configuration. The fact that you can construct this mapping doesn't mean much IMO.
- ajconway 6y ago
  I believe the argument goes like this: assume that conscious agents can be simulated (that is, consciousness is computable). What is computation? If sudden wind arranges leaves on the ground into a pattern that looks like “2 + 2 = 4”, was there a computation? We would usually answer “no”, as computation, like many other things, requires causal connection. But Greg Egan states that “there is nothing more to causality than the correlations between states”. Now, a deterministic simulation can be represented as a series of states recorded on a medium. Then you, an observer, can rewind and inspect what’s going on there, similar to reading a book or watching a movie. But the inhabitants of the simulation couldn’t care less about you observing their states. They already “exist”. In that case, why do we need the states of the simulation be recorded on the “film” in the order that makes sense for an outside observer? Surely a random arrangement would not affect the “simulation”. The same goes for the encoding. Now, if the arrangement doesn’t matter, and the encoding doesn’t matter, why there needs to be a specific physical medium at all? Surely somewhere in the Universe you can find arrangements of “dust” that “encode” the same states (in no particular order).
  - chr1 6y ago
    What does “encode” mean here? If it is supposed to contain the whole state in one chunk, the space of possible deterministic algorithms is too huge compared to the amount of dust, so it can't contain that. If it is supposed to find small chunks and rearrange them, it will not be using any information from the dust, and with same success encoder could use just any set of 0s and 1s. To circumvent finiteness of dust One could use digits of an irrational number or simple counting, but that is a very strange definition of existence. By that logic every possible history is just a number, so everything exists. The Important question is how to construct these numbers, find them in the sea of other meaningless numbers. I think Stephen Wolfram's idea of computational irreducibility is very useful here. It says that most computations (even very simple systems like some of cellular automata) cannot be replaced by other computations, and even to verify that a number represents result of a computation, one have to carry out that same computation. So the part of the book where states of the simulation are computed out of order is simply wrong, the latter state of the simulation could not have been computed without computing the intermediate states, and all the methods of computing that state were equivalent to the person living through that interval of time.
mcguire 6y ago
Yours may be the finest argument for constructivism I have seen.

Schiphol 6y ago

I think the idea is thst you get to pick and choose from the 'dust': if you have a perfectly Kolmogorov-random string, it's still trivial to go jumping from one bit here to another it there like so: 10101010..., or in any other, however complicated pattern.

saeranv 6y ago

> One issue with this argument that hasn't been brought up yet is Kolmogorov complexity. Must random blobs of dust Z will be algorithmically random, not containing any real structure. This is my issue as well, and something Greg Egan brings up in his FAQ on Dust theory as well. If my state is simply the result of a sampling of domain Y, why is my continuous state so stable? If we're assuming I'm sampling from a uniform (maximally random) distribution of potential states, there are many more state configurations where (for example) the physical laws should break down, then a stable configuration. So there must be some missing element in Dust theory that explains away this randomness.