The only way to recohere the system is if there's a boundary condition
Or if spacetime is compact in any way, or in fact if there is only finitely much negentropy. In any of these cases, your abstraction of a bunch of distinct but potentially interfering branches will break down, and you can pick up all of your old waste heat even if it is "running away" at light speed.
In the other universes, where somehow things can continue expanding at light speed indefinitely, you can recover perfect info by exploring possible physical theories until you find yourself.
But it seems like all of these considerations are far too frail to have much impact in themselves; they just serve to put a lower bound on how weird we can expect the far future to be.
In the other universes, where somehow things can continue expanding at light speed indefinitely, you can recover perfect info by exploring possible physical theories until you find yourself.
I don't immediately see how this gets around the problem; I'm probably just being stupid, but aren't you still left with a bunch of possible histories consistent with your current state/decisions only an unknown subset of which are real? ("Real" in the usual sense, i.e. can reliably be used to coordinate with other agents.)
I agree re lower bound on weirdness, I'd add it serves as a lower bound on how competent your decision theory has to be (which shouldn't be a problem).
This post may be interesting to some LWers.
In summary: it looks like our universe can support reversible computers which don't create entropy. Reversible computers can simulate irreversible computers, with pretty mild time and space blowup. So if moral value comes from computation, negentropy probably won't be such an important resource for distant future folks, and if the universe lasts a long time we may be able to simulate astronomically long-lived civilizations (easily 10^(10^25) clock cycles, using current estimates and neglecting other obstructions).
Has this been discussed before, and/or is there some reason that it doesn't work or isn't relevant? I suspect that this consideration won't matter in the long run, but it is at least interesting and seems to significantly deflate (long-run) concerns about entropy.