= 27d567bc2d48d9f40e9ccc20ea370b15)
Eliezer_Yudkowsky comments on If Many-Worlds Had Come First - Less Wrong
You are viewing a comment permalink. View the original post to see all comments and the full post content.
= 783df68a0f980790206b9ea87794c5b6)
Loading…= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Powered by Reddit
= f037147d6e6c911a85753b9abdedda8d)
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (179)
Basic question I probably should've asked earlier: Does shminux::RQM entail not-MWI?
If the answer is "no" then shminux::RQM is indeed plausibly shutting up, since by adding further information we can arrive at MWI. I plead guilty to failing to ask this question, note that shminux failed to volunteer the information, and finally plead that I think most RQMers would claim that theirs is an alternative to MWI.
MWI=universal state
Rovelli-rQM=no universal state
Can you describe in more detail what you mean by 'no universal state'?
By "state" I mean information physically embodied in a non relational way.
By "universal" I mean the maximal ensemble: universe, multiverse, cosmos, whatever.
(I think you might have been hearing "the universe does not have a state" as "nothing is real" or "nothing is out there". There is something out there, but it is not anything that can even be conceived as existing in a classical view-from nowhere style. "Following the idea of relational networks above, an RQM-oriented cosmology would have to account for the universe as a set of partial systems providing descriptions of one another. The exact nature of such a construction remains an open question."--WP)
To the extent that this seems to be meaningful at all, this would seem to imply that not only is the universe mysterious and ineffable, it's also uncomputable - since anything you can calculate in a turing machine (or even a few kinds of hypercomputers) can be "conceived of as existing in a classical view-from nowhere style" (it's just a list of memory states, together with the program). That's a lot of complexity just to be able to deny the idea of objective reality!
QM is computable. rQM doesnt change that. If an observer wants to do quantum cosmology, they can observe the universe, not from nowhere, but from their perspective, store observations and compute with them. Map-wise, nothing much has changed.
Territory-wise, it looks like the universe can't be a (classical) computer. Is that a problem?
As I understand it, any quantum computer can be modeled on a classical one, possibly with exponential slowdown.
Be modeled doesn't mean be.
I guess that's the root of our disagreement about instrumentalism.
The dictionary seems to be on my side.
I can see how your conclusion follows from that assumption, but the assumption is as strange as the conclusion. Ideally, an argument should proceed from plausible premises.
"The universe is not anything that can even be conceived as existing in a classical view-from nowhere style" also means that the universe can't be modeled on a computer (classical or otherwise). From a complexity theory point of view, this makes the rQM cosmology an exceptionally bad one, since you must have to add something uncomputable to QM to make this true (if there is even any logical model that makes this true at all).
The fact that you can still computably model a specific observer's subjective perspective isn't really relevant.
Out of the box, a classical computer doesn't represent the ontology of rQM because all information has an observer-independent representation, but s software layer can hide literal representations in the way a LISP gensym does. Uncomputability is not required.
In any case, classical computability isn't a good index of complexity. It's an index of how close something is to a classical computer. Problems are harder or easier to solve according to the technology used to solve them. That's why people don't write device drivers in LISP.
Um, computability has very little to do with "classical" computers. It's a very general idea relating to the existence of any algorithm at all.
Uncomputability isn't needed to model the ontology of rQM,
Well, general relativity, while descriptively very simple, is awfully complex if you measure complexity by the length of a simulator program, so perhaps in the interest of consistency you should join the anti Einsteinian crank camp first.
Those incredibly successful theories were based entirely on the notion of complexity in a more abstract language where things like having no outside view and no absolute spacetime are simpler than having outside view.
Nice non-sequitor you've got there. Newtonian mechanics is simpler than general relativity. It also happens to be wrong, so there's no point going back to it. But GR is not even that complex relative to a theory that claims that the cosmos is an ineffable mystery - GR has well defined equations, and takes place in a fixed riemannian manifold. You can in fact freely talk about the objective spacetime location of events in GR, using whatever coordinate system you like. This is because it is a good theory.
Actually GR shows the advantage of having an outside view and being able to fit things into a comprehensive picture. If my graduate GR course had refused to talk about manifolds and tensors and insisted that you could only measure "lengths relative to specific observers", and shown us a bunch of arcane equations for converting measurements between different observers' realties, I imagine it wouldn't have been half as fun.
(Although the fact that certain solutions to the GR equations allow closed timelike curves and thereby certain kinds of hypercomputation is less than ideal -- hopefully future unified theories will conspire to eliminate such shenanigens.)
The point is that absence of the absolute time really gets in the way of implementing a naive simulator, the sort that just updates per timestep. Furthermore, there is no preferred coordinate frame in GR, but there is a preferred coordinate frame in a simulator.
Ultimately, a Turing machine is highly arbitrary and comes with a complex structure, privileging implementations that fit into that structure, over conceptually simpler theories which do not.
But it's no problem for a simulator that derives a proof of the solution to the equations, such as a SAT solver. Linear time is not neccesary for simulation, just easier for humans to grasp.
Even if this is true, if the simulation is correct, the existence of such a preferred reference frame is unobservable to any observer inside the simulation, and therefore makes no difference. A simulation that does GR calculations in a particular coordinate system, still does GR calculations.
How are you even going to do those calculations exactly? If you approximate, itll be measurable.
Ultimately there is this minimal descriptive complexity approach that yields things like GR based on assumptions of as few absolutes as possible, and then theres this minimal complexity of implementation on a very specific machine approach, which would yield a lot of false predictions had anybody bothered to try to use it as the measurements improved.
edit: also under an ontology where invariants and relationals with no absolutes are not simpler its awfully strange to find oneself in an universe wheich looks like ours. The way i see it, there are better and worse ways to assign priors, and if you keep making observations with very low priors under one assignment but not other, you should consider the prioes scheme where you keep predicting wrong to be worse.