Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.

Egan's Theorem?

13th Sep 2020

11Charlie Steiner

2johnswentworth

3Charlie Steiner

2johnswentworth

5Charlie Steiner

2johnswentworth

6shminux

2johnswentworth

5ryan_b

2johnswentworth

4romeostevensit

3digital_carver

New Answer

New Comment

1 Answers sorted by

The answer to the question you actually asked is no, there is no ironclad guarantee of properties continuing, nor any guarantee that there will be a simple mapping between theories. With some effort you can construct some perverse Turing machines with bad behavior.

But the answer the more generalized question is yes, simple properties can be expected (in a probabilistic sense) to generalize even if the model is incomplete. This is basically Minimum Message Length prediction, which you can put on the theoretical basis of the Solomonoff prior (It's somewhere in Li and Vitanyi - chapter 5?).

there is no ironclad guarantee of properties continuing

Properties *continuing* is not what I'm asking about. The example in the OP is relevant: even if the entire universe undergoes some kind of phase change tomorrow and the macroscopic physical laws change entirely, it would still be true that the old laws *did* work before the phase change, and any new theory needs to account for that in order to be complete.

nor any guarantee that there will be a simple mapping between theories

I do not know of any theorem or counterexample which actually says this. Do you?

...si

34y

If by "account for that" you mean not be in direct conflict with earlier sense data, then sure. All tautologies about the data will continue to be true. Suppose some data can be predicted by classical mechanics with 75% accuracy. This is a tautology given the data itself, and no future theory will somehow make classical mechanics stop giving 75% accurate predictions for that past data.
Maybe that's all you meant?
I'd sort of interpreted you as asking questions about properties of the theory. E.g. "this data is really well explained by the classical mechanics of point particles, therefore any future theory should have a particularly simple relationship to the point particle ontology." It seems like there shouldn't be a guaranteed relationship that's much simpler than reconstructing the data and recomputing the inferred point particles.
I spent a little while trying to phrase this in terms of Turing machines but I don't think I quite managed to capture the spirit.

24y

Yeah, I'm claiming exactly the opposite of this. When the old theory itself has some simple structure (e.g. classical mechanics), there should be a guaranteed relationship that's much simpler than reconstructing the data and recomputing the inferred point particles.
One possible formulation: if I find that a terabyte of data compresses down to a gigabyte, and then I find a different model which compresses it down to 500MB, there should be a relationship between the two models which can be expressed without expanding out the whole terabyte. (Or, if there isn't such a relationship, that means the two models are capturing different patterns from the data, and there should exist another model which compresses the data more than either by capturing the patterns found by both models.)

54y

Right, it's a little tricky to specify exactly what this "relationship" is. Is the notion that you should be able to compress the approximate model, given an oracle for the code of the best one (i.e. that they share pieces?). Because most Turing machines don't compress well, and so it's easy to find counterexamples (the most straightforward class is where the approximate model is already extremely simple).
Anyhow, like I said, hard to capture the spirit of the problem. But when I *do* try to formalize the problem, it tends to not have the property, which is definitely driving my intuition.

24y

I'd expect Turing machines to be a bad way to model this. They're inherently blackboxy; the only "structure" they make easy to work with is function composition. The sort of structures relevant here don't seem like they'd care much about function boundaries. (This is why I use models like these as my default model of computation these days.)
Anyway, yeah, I'm still not sure what the "relationship" should be, and it's hard to formulate in a way that seems to capture the core idea.

6 comments, sorted by Click to highlight new comments since: Today at 6:27 AM

When physicists were figuring out quantum mechanics, one of the major constraints was that it had to reproduce classical mechanics in all of the situations where we already knew that classical mechanics works well - i.e. most of the macroscopic world.

Well, that's false. The details of quantum to classical transition are very much an open problem. Something happens after the decoherence process removes the off-diagonal elements from the density matrix, and before only a single eigenvalue remains; the mysterious projection postulate. We have no idea at what scales it becomes important and in what way. The original goal was to **explain new observations**, definitely. But it was not "to reproduce classical mechanics in all of the situations where we already knew that classical mechanics works well".

Your other examples is more in line with what was going on, such as

for special and general relativity - they had to reproduce Galilean relativity and Newtonian gravity, respectively, in the parameter ranges where those were known to work

That program worked out really well. But that is not a universal case by any means. Sometimes new models don't work in the old areas at all. The free will or the consciousness models do not reproduce physics or vice versa.

The way I understand the "it all adds up to normality" maxim (not a law or a theorem by any means), is that new models do not make your old models obsolete where the old models worked well, nothing more.

I have trouble understanding what you would want from what you dubbed the Egan's theorem. In one of the comment replies you suggested that the same set of observations could be modeled by two different models, and there should be a morphism between the two models, either directly or through a third model that is more "accurate" or "powerful" in some sense than the other two. If I knew enough category theory, I would probably be able to express it in terms of some commuting diagrams, but alas. But maybe I misunderstand your intent.

In one of the comment replies you suggested that the same set of observations could be modeled by two different models, and there should be a morphism between the two models, either directly or through a third model that is more "accurate" or "powerful" in some sense than the other two. If I knew enough category theory, I would probably be able to express it in terms of some commuting diagrams, but alas.

Yes, something like that would capture the idea, although it's not necessarily the only or best way to formulate it.

So in a very simple case, would something like a differential equation to which we later add a higher order term qualify?

It seems like if it is to be generally true, iterated refinements of "the same" model are really just a special case.

Sure. In that case, it would say something like "the higher order terms should be small in places where the lower-order equation was already accurate".

Sounds similar to Noether's Theorem in some ways when you take that theorem philosophically and not just mathematically.

The first paragraph reminded me of the Correspondence principle, which seems close to what you're looking for (if I'm understanding correctly). The Wikipedia article has an "Other scientific theories" section, indicating it does get used more generally than the quantum->classical correspondence Bohr had in mind (although that section doesn't have any citations unfortunately). Perhaps it's worth labeling it "General Correspondence principle" or "Generalized Correspondence principle" in practice, when using it outside of physics.

When physicists were figuring out quantum mechanics, one of the major constraints was that it had to reproduce classical mechanics in all of the situations where we already knew that classical mechanics works well - i.e. most of the macroscopic world. Likewise for special and general relativity - they had to reproduce Galilean relativity and Newtonian gravity, respectively, in the parameter ranges where those were known to work. Statistical mechanics had to reproduce the fluid theory of heat; Maxwell's equations had to agree with more specific equations governing static electricity, currents, magnetic fields and light under various conditions.

Even if the entire universe undergoes some kind of phase change tomorrow and the macroscopic physical laws change entirely, it would still be true that the old laws

didwork before the phase change. Any new theory and any new theory would still have to be consistent with the old laws working, where and when they actually did work.This is Egan's Law: it all adds up to normality. When new theory/data comes along, the old theories are still just as true as they always were. New models must reproduce the old in all the places where the old models worked; otherwise the new models are incorrect, at least in the places where the old models work and the new models disagree with them.

It really seems like this should be not just a

Law, but aTheorem.I imagine Egan's Theorem would go something like this. We find a certain type of pattern in some data. The pattern is highly unlikely to arise by chance, or allows significant compression of the data, or something along those lines. Then the theorem would say that, in any model of the data, either:

The meat of such a theorem would be finding classes of patterns which imply model-properties less trivial than just "the model must predict the pattern" - i.e. patterns which imply properties we actually care about. Structural properties like e.g. (approximate) conditional independencies seem particularly relevant, as well as properties involving abstractions/embedded submodels (in which case the theorem should tell how to find the abstraction/embedding).

Does anyone know of theorems like that? Maybe this is equivalent to some standard property in statistics and I'm just overthinking it?