I think the biggest reason Less Wrong seems like a cult is because there’s very little self-skepticism; people seem remarkably confident that their idiosyncratic views must be correct (if the rest of the world disagrees, that’s just because they’re all dumb). There’s very little attempt to provide any “outside” evidence that this confidence is correctly-placed (e.g. by subjecting these idiosyncratic views to serious falsification tests).
Instead, when someone points this out, Eliezer fumes “do you know what pluralistic ignorance is, and Asch’s conformity experiment? … your smart friends and favorite SF writers are not remotely close to the rationality standards of Less Wrong”.
What’s especially amusing is that EY is able to keep this stuff up by systematically ignoring every bit of his own advice: telling people to take the outside view and then taking the inside one, telling people to look into the dark while he studiously avoids it, emphasizing the importance of AI safety while he embarks on an extremely dangerous way of building AI—you can do this with pretty much every entry in the sequences.
These are the sorts of things that make me think LessWrong is most interesting as a study in psychoceramics.
There’s very little attempt to provide any “outside” evidence that this confidence is correctly-placed (e.g. by subjecting these idiosyncratic views to serious falsification tests).
Offhand, can you think of a specific test that you think ought to be applied to a specific idiosyncratic view?
My read on your comment is: LWers don’t act humble, therefore they are crackpots. I agree that LWers don’t always act humble. I think it’d be a good idea for them to be more humble. I disagree that lack of humility implies crackpottery. In my mind, crackpottery is a function of your reasoning, not your mannerisms.
Your comment is a bit short on specific failures of reasoning you see—instead, you’re mostly speaking in broad generalizations. It’s fine to have general impressions, but I’d love to see a specific failure of reasoning you see that isn’t of the form “LWers act too confident”. For example, a specific proposition that LWers are too confident in, along with a detailed argument for why. Or a substantive argument for why SI’s approach to AI is “extremely dangerous”. (I personally know pretty much everyone who works for SI, and I think there’s a solid chance that they’ll change their approach if your argument is good enough. So it might not be a complete waste of time.)
you can do this with pretty much every entry in the sequences
Now it sounds like you’re deliberately trying to be be inflammatory ಠ_ಠ
Offhand, can you think of a specific test that you think ought to be applied to a specific idiosyncratic view?
Well, for example, if EY is so confident that he’s proven “MWI is obviously true—a proposition far simpler than the argument for supporting SIAI”, he should try presenting his argument to some skeptical physicists. Instead, it appears the physicists who have happened to run across his argument found it severely flawed.
How rational is it to think that you’ve found a proof most physicists are wrong and then never run it by any physicists to see if you’re right?
My read on your comment is: LWers don’t act humble, therefore they are crackpots.
I do not believe that.
As for why SI’s approach is dangerous, I think Holden put it well in the most upvoted post on the site.
I’m not trying to be inflammatory, I just find it striking.
it appears the physicists who have happened to run across his argument found it severely flawed
The criticisms at those links have nothing to do with the argument for MWI. They are just about a numerical mistake in an article illustrating how QM works.
The actual argument for MWI that is presented is something like this: Physicists believe that the wavefunction is real and that it collapses on observation, because that is the first model that explained all the data, and science holds onto working models if they are falsified. But we can also explain all the data by saying that the wavefunction is real and doesn’t collapse, if we learn to see the wavefunction as containing multiple worlds that are equally real. The wavefunction doesn’t collapse, it just naturally spreads out into separate parts and what we see is one of those separate parts. A no-collapse theory is simpler than a collapse theory because it has one less postulate, so even though there are no new predictions, by Bayes (or is it Occam?) we can favor the no-collapse theory over the collapse theory. Therefore, there are many worlds.
This is informal reasoning about which qualitative picture of the world to favor, so it is not something that can be verified or falsified by a calculation or an experiment. Therefore, it’s not something that a hostile physicist could crisply debunk, even if they wanted to. In the culture of physics there are numerous qualitative issues where there is no consensus, and where people take sides on the basis of informal reasoning. Eliezer’s argument is on that level; it is an expression in LW idiom, of a reason for believing in MWI that quite a few physicists probably share. It can’t be rebutted by an argument along the lines that Eliezer doesn’t know his physics, because it is an argument which (in another form) a physicist might actually make! So if someone wants to dispute it, they’ll have to do so, just as if they were intervening in any of these informal professional disagreements which exist among physicists, by lines of argument about plausibility, future theoretical prospects, and so on.
ETA One more comment about the argument for MWI as I have presented it. Physicists don’t agree that the wavefunction is real. The debate over whether it is real, goes all the way back to Schrodinger (it’s a real physical object or field) vs Heisenberg (it’s just a calculating device). The original Copenhagen interpretation was in Heisenberg’s camp: a wavefunction is like a probability distribution, and “collapse” is just updating on the basis of new experimental facts (the electron is seen at a certain location, so the wavefunction should be “collapsed” to that point, in order to reflect the facts). I think it’s von Neumann who introduced wavefunction realism into the Copenhagen interpretation (when he axiomatized QM), and thereby the idea of “observer-induced collapse of the wavefunction” as an objective physical process. Though wavefunction realism was always going to creep up on physicists, since they describe everything with wavefunctions (or state vectors) and habitually refer to these as “the state” of the object, rather than “the state of our knowledge” of the object; also because Copenhagen refused to talk about unobserved realities (e.g. where the electron is, when it’s not being seen to be somewhere), an attitude which was regarded as prim positivistic virtue by the founders, but which created an ontological vacuum that was naturally filled by the de-facto wavefunction realism of physics practice.
BTW, it’s important to note that by some polls an actual majority of theoretical physicists now believe in MWI, and this was true well before I wrote anything. My only contributions are in explaining the state of the issue to nonphysicists (I am a good explainer), formalizing the gross probability-theoretic errors of some critiques of MWI (I am a domain expert at that part), and stripping off a lot of soft understatement that many physicists have to do for fear of offending sillier colleagues (i.e., they know how incredibly stupid the Copenhagen interpretation appears nowadays, but will incur professional costs from saying it out loud with corresponding force, because there are many senior physicists who grew up believing it).
The idea that Eliezer Yudkowsky made up the MWI as his personal crackpot interpretation isn’t just a straw version of LW, it’s disrespectful to Everett, DeWitt, and the other inventors of MWI. It does seem to be a common straw version of LW for all that, presumably because it’s spontaneously reinvented any time somebody hears that MWI is popular on LW and they have no idea that MWI is also believed by a plurality and possibly a majority of theoretical physicists and that the Quantum Physics Sequence is just trying to explain why to nonphysicists / formalize the arguments in probability-theoretic terms to show their nonambiguity.
The original source for that “58%” poll is Tipler’s The Physics of Immortality, where it’s cited (chapter V, note 6) as “Raub 1991 (unpublished)”. (I know nothing about the pollster, L. David Raub, except that he corresponded with Everett in 1980.) Tipler says that Feynman, Hawking, and Gell-Mann answered “Yes, I think the MWI is true”, and he lists Weinberg as another believer. But Gell-Mann’s latest paper is a one-history paper, Weinberg’s latest paper is about objective collapse, and Feynman somehow never managed to go on record anywhere else about his belief in MWI.
It’s been suggested that I’m the one who invented the idea that it’s obviously true rather than just one more random interpretation; or even that I’m fighting a private war for some science-fiction concept, rather than being one infantry soldier in a long and distinguished battle of physicists. Certainly your remark to the extent that “he should try presenting his argument to some skeptical physicists” sounds like this. Any physicist paying serious attention to this issue (most people aren’t paying attention to most things most of the time) will have already heard many of the arguments, and not from me. It sounds like we have very different concepts of the state of play.
I just can’t ignore this.
If you take a minute to actually look at the talk section of that wikipedia page you will see those polls being thorn to pieces.
David Deutsch himself has stated that less than 10% of the people doing quantum fundamentals believe in MWI and then within that minority there are a lot of diverging views.
So this is still not by any means a “majority interpretation”.
As Mitchell_Porter has pointed out Gell-Mann certainly do not believe in MWI. Nor do Steven Weinberg, he denounced his ‘faith’ in it in a paper last year. Feynman certainly did never talk about it, which to me is more than enough indication that he did not endorse it.
Hawking is a bit harder, he is on record seemingly being pro and con it, so I guess he is a fence sitter.
But more importantly is the fact that none of the proponents agree on what MWI they support. (This includes you Eliezer)
Zurek is another fence sitter, partly pro-some-sort-of-MWI, partly pro-It-from-Bit. Also his way of getting the Born Rule in MWI is quite a bit different. From what I understand, only the worlds that are “persistent” are actualized.
This reminds me of Robin Hanson’s mangled worlds where only some worlds are real and the rest gets “cancelled” out somehow. Yet they are completley different ways of looking at MWI.
Then you got David Deutsch’s fungible worlds which is slightly different from David Wallace’s worlds.
Tegmark got his own views etc.
There seems to be no single MWI and there has been no answer to the Born Rule.
So I want to know why you keep on talking about it as it is a slam dunk?
I think your use of “believe in” is a little suspect here. I’m willing to believe that more than half of all theoretical physicists believe some variant of the MWI is basically right (though the poll can’t have been that recent if Feynman was part of it, alas), but that’s different from the claim that there are no non-MWI interpretations worth considering, which is something a lot of people, including me, seem to be taking from the QP sequence. Do you believe that that’s a majority view, or anything close to one? My impression is that that view is very uncommon, not just in public but in private too...at least outside of Less Wrong.
That sounds correct to me. A physicist who also possesses probability-theory expertise and who can reason with respect to Solomonoff Induction and formal causal models should realize that single-world variants of MWI are uniformly unworkable (short of this world being a runtime-limited computer simulation); but such is rare (though not unheard-of) among professional physicists; and among the others, you can hardly blame them for trying to keep an open mind.
single-world variants of MWI are uniformly unworkable
The Penrose’sobjective collapse theory saying that the entanglement scale is limited by gravity, which results in the macroscopic objects remaining essentially classical, does not look all that unworkable.
It’d still be the only FTL discontinuous non-differentiable non-CPT-symmetric non-unitary non-local-in-the-configuration-space etc. etc. process in all of physics, to explain a phenomenon (why do we see only one outcome?) that doesn’t need explaining.
Well, one advantage of it is that it is testable, and so is not a mere interpretation, which holds a certain amount of appeal to the more old-fashioned of us.
I agree, and I myself was, and am still, sentimentally fond of Penrose for this reason, and I would cheer on any agency that funded a test. However and nonetheless, “testable” is not actually the same as “plausible”, scientifically virtuous as it may be.
Not if it doesn’t allow FTL communication, unless you want to argue that quantum entanglement is a FTL phenomenon, but that wouldn’t be an issue of the particular interpretation.
discontinuous non-differentiable
Not necessarily. Irreversible and stochastic quantum processes can be time-continuous and time-differentiable.
Consider the processes described by the Lindblad equation, for instance.
non-CPT-symmetric
CPT symmetry is a property of conventional field theories, not all quantum theories necessarily have it, and IIUC, there are ongoing experiments to search for violations. CPT symmetry is just the last of a series of postulated symmetries, the previous ones (C symmetry, P symmetry, T symmetry and CP symmetry) have been experimentally falsified.
non-unitary
Right, and that’s the point of objective collapse theories.
non-local-in-the-configuration-space
I’m not sure what you mean by that, but locality in physics is defined with respect to space and time, not to arbitrary configuration spaces.
to explain a phenomenon (why do we see only one outcome?) that doesn’t need explaining.
AFAIK, there have been attempts to derive the Born rule in Everett’s interpretation, but they didn’t lead to uncontroversial results.
Not necessarily. Irreversible and stochastic quantum processes can be time-continuous and time-differentiable.
I have never seen a proposed mechanism of ontological collapse that actually fits this, though.
Not if it doesn’t allow FTL communication
The inability to send a signal that you want, getting instead a Born-Rule-based pure random signal, doesn’t change that this Born-Rule-based pure random signal is, under ontological collapse distributed FTL.
I have never seen a proposed mechanism of ontological collapse that actually fits this, though.
AFAIK, Penrose’s interpretation doesn’t describe the details of the collapse process, it just says that above about the “one graviton” level of energy separation collapse will occur.
It doesn’t commit to collapse being instantaneous: It could be that the state evolution is governed by a non-linear law that approximates very well the linear Schrödinger equation in the “sub-graviton” regime and has a sharp, but still differentiable phase transition when approaching the “super-graviton” regime.
The GRW interpretation assumes instantaneous collapse, IIUC, but it would be a trivial modification to have fast, differentiable collapse.
My point is that non-differentiable collapse is not a requirement of objective collapse interpretations.
The inability to send a signal that you want, getting instead a Born-Rule-based pure random signal, doesn’t change that this Born-Rule-based pure random signal is, under ontological collapse distributed FTL.
But that’s an issue of QM, irrespective of the particular interpretation. Indeed the “spooky action at distance” bugged Einstein and many people of his time, but the modern view is that as long as you don’t have causal influences (that is, information transmission) propagating FTL, you don’t violate special relativity.
But that’s an issue of QM, irrespective of the particular interpretation.
No, it isn’t. QM is purely causal and relativistic. You can look into the equations and prove that nothing FTL is in there. The closest you get is accounting for the possibility of a vacuum bubble having appeared nearby a particle with exactly its energy, and the antimatter part of it the bubble then cancels with the particle. And that isn’t much like FTL.
When you do an EPR experiment, the appearance of FTL communication arises from the assumption that the knowledge you gain about what you’ll see if you go check the other branch of the experiment is something happens at the other end of the experiment, instead of locally, with the information propagating to the other end of the experiment as you go to check. The existence of nonlocal states does not imply nonlocal communication.
I’m puzzled.
What does Solomonoff Induction have to say about experimentally undistinguishable (as far as we can practically test, at least) interpretations of the same theory?
BTW, it’s important to note that by some polls an actual majority of theoretical physicists now believe in MWI,
But there is s case to be made for relatioal QM as superior to both MWI an collpase interpretations. I have metuioned it several times. I am still waiting to hear back.
It makes sense the way rQM says: there is no non-relational state, so there is not answer to “is the cat dead or
alive (absent an observer)”. Since rQM says there is no state, you don’t disprove it by insisting there is state.
OK, so suppose we have an observer. Now look at the cat. Is it alive or dead? If it is alive and only alive, well, we can affix the phrase “relative to the observer” but it doesn’t diminish the absoluteness of the cat’s being alive. But if the cat is alive “relative to one observer to which it is alive”, and dead “relative to another observer to which it is dead”, how can we possibly make sense of that except in many-worlds fashion, by saying there are two cats and two observers?
If two observers measure a cat, they will get compatible results. However one observer can have less complete information (“the cat collapsed”) and another more complete (“the cat is uncollapsed”). Observers can disagree about “collapse” because that is just an issue of their information, not an objective property.
“Relational interpretation
The relational interpretation makes no fundamental distinction between the human experimenter, the cat, or the apparatus, or between animate and inanimate systems; all are quantum systems governed by the same rules of wavefunction evolution, and all may be considered “observers.” But the relational interpretation allows that different observers can give different accounts of the same series of events, depending on the information they have about the system.[11] The cat can be considered an observer of the apparatus; meanwhile, the experimenter can be considered another observer of the system in the box (the cat plus the apparatus). Before the box is opened, the cat, by nature of it being alive or dead, has information about the state of the apparatus (the atom has either decayed or not decayed); but the experimenter does not have information about the state of the box contents. In this way, the two observers simultaneously have different accounts of the situation: To the cat, the wavefunction of the apparatus has appeared to “collapse”; to the experimenter, the contents of the box appear to be in superposition. Not until the box is opened, and both observers have the same information about what happened, do both system states appear to “collapse” into the same definite result, a cat that is either alive or dead.”—WP
In the interpretation of QM, one of the divides is between ontic and epistemic interpretations of the wavefunction. Ontic interpretations of the wavefunction treat it as a thing, epistemic interpetations as an incomplete description or a tabulation of uncertainty, just like a probability distribution.
In the relational interpretation of QM, are the states understood as ontic or as epistemic? The passage you quote makes them sound epistemic: the cat knows but the observer outside the box doesn’t, so the observer outside the box uses a different wavefunction. That undoubtedly implies that the wavefunction of the observer outside the box is epistemic, not ontic; the cat knows something that the outside observer doesn’t, an aspect of reailty which is already definite even though it is not definite in the outside observer’s description.
Or at least, this ought to imply that quantum state in the relational interpretation are epistemic. However, this is never explicitly stated, and instead meaningless locutions are adopted which make it sound as if the quantum states are to be regarded as ontic, but “relative”.
There are certain very limited senses in which it makes sense to say that the state of something is relative. For example, we may be floating in space, and what is up to you may be down to me, so whether one object is above another object may be relative to an observer. But clearly such a dodge will not work for something like Schrodinger’s cat. Either the cat is alive, dead, both, or neither. It can’t be “alive for one observer and dead for another” and still be just one cat. But that is the ontological implication one gets, if “relational QM’ is interpreted as an ontic interpertation.
On the other hand, if it is an epistemic interpretation, then it still hasn’t answered the question, “what is the nature of reality? what is the physical ontology behind the formalism and the instrumental success?”
It can’t be “alive for one observer and dead for another”
It can’t in rQM:
“However, the comparison does not lead to contradiction because the comparison is itself a physical process that must be understood in the context of quantum mechanics. Indeed, O′ can physically interact with the electron and then with the l.e.d. (or, equivalently, the other way around). If, for instance, he finds the spin of the electron up, quantum mechanics predicts that he will then consistently find the l.e.d. on (because in the first measurement the state of the composite system collapses on its [spin up/l.e.d. on] component). That is, the multiplicity of accounts leads to no contradiction precisely because the comparison between different accounts can only be a physical quantum interaction. This internal self-consistency of the quantum formalism is general, and it is perhaps its most remarkable aspect. This self consistency is taken in relational quantum mechanics as a strong indication of the relational nature of the world”—SEP
There were plenty of physicists reading those posts when they first came out on OB (the most famous name being Scott Aaronson). Some later readers have indeed asserted that there’s a problem involving a physically wrong factor of i in the first couple of posts (i.e. that’s allegedly not what a half-silvered mirror does to the phase in real life), which I haven’t yet corrected because I would need to verify with a trusted physicist that this was correct, and then possibly craft new illustrations instead of using the ones I found online, and this would take up too much time relative to the point that talking about a phase change of −1 instead of i so as to be faithful to real-world mirrors is an essentially trivial quibble which has no effect on any larger points. If anyone else wants to rejigger the illustration or the explanation so that it flows correctly, and get Scott Aaronson or another known trusted physicist to verify it, I’ll be happy to accept the correction.
Aside from that, real physicists haven’t objected to any of the math, which I’m actually pretty darned proud of considering that I am not a physicist.
There were plenty of physicists reading those posts when they first came out on OB (the most famous name being Scott Aaronson)
As Scott keeps saying, he’s not a physicist! He’s a theoretical computer scientist with a focus on quantum computing. He clearly has very relevant expertise, but you should get his field right.
I still wonder why you haven’t written a update in 4 years regarding this topic. Especially in regards to the Born Rule probability not having a solution yet + the other problems.
You also have the issue of overlap vs non-overlapping of worlds, which again is a relevant issue in the Many Worlds interpretation. Overlap = the typical 1 world branching into 2 worlds. Non-overlap = 2 identical worlds diverging (Saunders 2010, Wilson 2005-present)
Also I feel like the QM sequence is a bit incomplete when you do not give any thought to things like Gerard ’t Hoofts proposal of a local deterministic reality giving rise to quantum mechanics from a cellular automaton at the planck scale?
It’s misleading to say the MWI is “a slam dunk” winner when there are so many unanswered questions. Mitchell Porter is one of the few persons here who seem to have a deep understanding of the subject before reading your sequence, so he has raised some interesting points...
I agree that EY is probably overconfident in MWI, although I’m uniformed about QM so I can’t say much with confidence. I don’t think it’s accurate to damn all of Less Wrong because of this. For example, this post questioning the sequence was voted up highly.
I don’t think EY claims to have any original insights pointing to MWI. I think he’s just claiming that the state of the evidence in physics is such that MWI is obviously correct, and this is evidence as to the irrationality of physicists. I’m not too sure about this myself.
As for why SI’s approach is dangerous, I think Holden put it well in the most upvoted post on the site.
Well there have been responses to that point (here’s one). I wish you’d be a bit more self-skeptical and actually engage with that (ongoing) debate instead of summarizing your view on it and dismissing LW because it largely disagrees with your view.
EY is able to keep this stuff up by systematically ignoring every bit of his own advice
Your examples seem … how do I put this … unreliable. The first two are less examples and more insults, since you do not provide any actual examples of these tendencies; the last one would be more serious, if he hadn’t written extensively on why he believes this to be the safest way—the only way that isn’t suicidal—or if you had provided some evidence that his FAI proposals are “extremely dangerous”. And, of course, airily proclaiming that this is true of “pretty much every entry in the sequences” seems, in the context of these examples, like an overgeneralization at best and … well, I’m not going to bother outlining the worst possible interpretation for obvious reasons.
I think the biggest reason Less Wrong seems like a cult is because there’s very little self-skepticism; people seem remarkably confident that their idiosyncratic views must be correct (if the rest of the world disagrees, that’s just because they’re all dumb). There’s very little attempt to provide any “outside” evidence that this confidence is correctly-placed (e.g. by subjecting these idiosyncratic views to serious falsification tests).
Instead, when someone points this out, Eliezer fumes “do you know what pluralistic ignorance is, and Asch’s conformity experiment? … your smart friends and favorite SF writers are not remotely close to the rationality standards of Less Wrong”.
What’s especially amusing is that EY is able to keep this stuff up by systematically ignoring every bit of his own advice: telling people to take the outside view and then taking the inside one, telling people to look into the dark while he studiously avoids it, emphasizing the importance of AI safety while he embarks on an extremely dangerous way of building AI—you can do this with pretty much every entry in the sequences.
These are the sorts of things that make me think LessWrong is most interesting as a study in psychoceramics.
Offhand, can you think of a specific test that you think ought to be applied to a specific idiosyncratic view?
My read on your comment is: LWers don’t act humble, therefore they are crackpots. I agree that LWers don’t always act humble. I think it’d be a good idea for them to be more humble. I disagree that lack of humility implies crackpottery. In my mind, crackpottery is a function of your reasoning, not your mannerisms.
Your comment is a bit short on specific failures of reasoning you see—instead, you’re mostly speaking in broad generalizations. It’s fine to have general impressions, but I’d love to see a specific failure of reasoning you see that isn’t of the form “LWers act too confident”. For example, a specific proposition that LWers are too confident in, along with a detailed argument for why. Or a substantive argument for why SI’s approach to AI is “extremely dangerous”. (I personally know pretty much everyone who works for SI, and I think there’s a solid chance that they’ll change their approach if your argument is good enough. So it might not be a complete waste of time.)
Now it sounds like you’re deliberately trying to be be inflammatory ಠ_ಠ
Well, for example, if EY is so confident that he’s proven “MWI is obviously true—a proposition far simpler than the argument for supporting SIAI”, he should try presenting his argument to some skeptical physicists. Instead, it appears the physicists who have happened to run across his argument found it severely flawed.
How rational is it to think that you’ve found a proof most physicists are wrong and then never run it by any physicists to see if you’re right?
I do not believe that.
As for why SI’s approach is dangerous, I think Holden put it well in the most upvoted post on the site.
I’m not trying to be inflammatory, I just find it striking.
The criticisms at those links have nothing to do with the argument for MWI. They are just about a numerical mistake in an article illustrating how QM works.
The actual argument for MWI that is presented is something like this: Physicists believe that the wavefunction is real and that it collapses on observation, because that is the first model that explained all the data, and science holds onto working models if they are falsified. But we can also explain all the data by saying that the wavefunction is real and doesn’t collapse, if we learn to see the wavefunction as containing multiple worlds that are equally real. The wavefunction doesn’t collapse, it just naturally spreads out into separate parts and what we see is one of those separate parts. A no-collapse theory is simpler than a collapse theory because it has one less postulate, so even though there are no new predictions, by Bayes (or is it Occam?) we can favor the no-collapse theory over the collapse theory. Therefore, there are many worlds.
This is informal reasoning about which qualitative picture of the world to favor, so it is not something that can be verified or falsified by a calculation or an experiment. Therefore, it’s not something that a hostile physicist could crisply debunk, even if they wanted to. In the culture of physics there are numerous qualitative issues where there is no consensus, and where people take sides on the basis of informal reasoning. Eliezer’s argument is on that level; it is an expression in LW idiom, of a reason for believing in MWI that quite a few physicists probably share. It can’t be rebutted by an argument along the lines that Eliezer doesn’t know his physics, because it is an argument which (in another form) a physicist might actually make! So if someone wants to dispute it, they’ll have to do so, just as if they were intervening in any of these informal professional disagreements which exist among physicists, by lines of argument about plausibility, future theoretical prospects, and so on.
ETA One more comment about the argument for MWI as I have presented it. Physicists don’t agree that the wavefunction is real. The debate over whether it is real, goes all the way back to Schrodinger (it’s a real physical object or field) vs Heisenberg (it’s just a calculating device). The original Copenhagen interpretation was in Heisenberg’s camp: a wavefunction is like a probability distribution, and “collapse” is just updating on the basis of new experimental facts (the electron is seen at a certain location, so the wavefunction should be “collapsed” to that point, in order to reflect the facts). I think it’s von Neumann who introduced wavefunction realism into the Copenhagen interpretation (when he axiomatized QM), and thereby the idea of “observer-induced collapse of the wavefunction” as an objective physical process. Though wavefunction realism was always going to creep up on physicists, since they describe everything with wavefunctions (or state vectors) and habitually refer to these as “the state” of the object, rather than “the state of our knowledge” of the object; also because Copenhagen refused to talk about unobserved realities (e.g. where the electron is, when it’s not being seen to be somewhere), an attitude which was regarded as prim positivistic virtue by the founders, but which created an ontological vacuum that was naturally filled by the de-facto wavefunction realism of physics practice.
BTW, it’s important to note that by some polls an actual majority of theoretical physicists now believe in MWI, and this was true well before I wrote anything. My only contributions are in explaining the state of the issue to nonphysicists (I am a good explainer), formalizing the gross probability-theoretic errors of some critiques of MWI (I am a domain expert at that part), and stripping off a lot of soft understatement that many physicists have to do for fear of offending sillier colleagues (i.e., they know how incredibly stupid the Copenhagen interpretation appears nowadays, but will incur professional costs from saying it out loud with corresponding force, because there are many senior physicists who grew up believing it).
The idea that Eliezer Yudkowsky made up the MWI as his personal crackpot interpretation isn’t just a straw version of LW, it’s disrespectful to Everett, DeWitt, and the other inventors of MWI. It does seem to be a common straw version of LW for all that, presumably because it’s spontaneously reinvented any time somebody hears that MWI is popular on LW and they have no idea that MWI is also believed by a plurality and possibly a majority of theoretical physicists and that the Quantum Physics Sequence is just trying to explain why to nonphysicists / formalize the arguments in probability-theoretic terms to show their nonambiguity.
The original source for that “58%” poll is Tipler’s The Physics of Immortality, where it’s cited (chapter V, note 6) as “Raub 1991 (unpublished)”. (I know nothing about the pollster, L. David Raub, except that he corresponded with Everett in 1980.) Tipler says that Feynman, Hawking, and Gell-Mann answered “Yes, I think the MWI is true”, and he lists Weinberg as another believer. But Gell-Mann’s latest paper is a one-history paper, Weinberg’s latest paper is about objective collapse, and Feynman somehow never managed to go on record anywhere else about his belief in MWI.
I trust Tipler as far as I can throw his book.
(It’s a large book, and I’m not very strong.)
Has anyone seriously suggested you invented MWI? That possibility never even occurred to me.
It’s been suggested that I’m the one who invented the idea that it’s obviously true rather than just one more random interpretation; or even that I’m fighting a private war for some science-fiction concept, rather than being one infantry soldier in a long and distinguished battle of physicists. Certainly your remark to the extent that “he should try presenting his argument to some skeptical physicists” sounds like this. Any physicist paying serious attention to this issue (most people aren’t paying attention to most things most of the time) will have already heard many of the arguments, and not from me. It sounds like we have very different concepts of the state of play.
Can’t help but compare this to the Swiftian battle of big-endians and little-endians, only the interpretational war makes even less sense.
I just can’t ignore this. If you take a minute to actually look at the talk section of that wikipedia page you will see those polls being thorn to pieces.
David Deutsch himself has stated that less than 10% of the people doing quantum fundamentals believe in MWI and then within that minority there are a lot of diverging views. So this is still not by any means a “majority interpretation”.
As Mitchell_Porter has pointed out Gell-Mann certainly do not believe in MWI. Nor do Steven Weinberg, he denounced his ‘faith’ in it in a paper last year. Feynman certainly did never talk about it, which to me is more than enough indication that he did not endorse it. Hawking is a bit harder, he is on record seemingly being pro and con it, so I guess he is a fence sitter.
But more importantly is the fact that none of the proponents agree on what MWI they support. (This includes you Eliezer)
Zurek is another fence sitter, partly pro-some-sort-of-MWI, partly pro-It-from-Bit. Also his way of getting the Born Rule in MWI is quite a bit different. From what I understand, only the worlds that are “persistent” are actualized. This reminds me of Robin Hanson’s mangled worlds where only some worlds are real and the rest gets “cancelled” out somehow. Yet they are completley different ways of looking at MWI. Then you got David Deutsch’s fungible worlds which is slightly different from David Wallace’s worlds. Tegmark got his own views etc.
There seems to be no single MWI and there has been no answer to the Born Rule.
So I want to know why you keep on talking about it as it is a slam dunk?
Good question.
I think your use of “believe in” is a little suspect here. I’m willing to believe that more than half of all theoretical physicists believe some variant of the MWI is basically right (though the poll can’t have been that recent if Feynman was part of it, alas), but that’s different from the claim that there are no non-MWI interpretations worth considering, which is something a lot of people, including me, seem to be taking from the QP sequence. Do you believe that that’s a majority view, or anything close to one? My impression is that that view is very uncommon, not just in public but in private too...at least outside of Less Wrong.
That sounds correct to me. A physicist who also possesses probability-theory expertise and who can reason with respect to Solomonoff Induction and formal causal models should realize that single-world variants of MWI are uniformly unworkable (short of this world being a runtime-limited computer simulation); but such is rare (though not unheard-of) among professional physicists; and among the others, you can hardly blame them for trying to keep an open mind.
The Penrose’s objective collapse theory saying that the entanglement scale is limited by gravity, which results in the macroscopic objects remaining essentially classical, does not look all that unworkable.
It’d still be the only FTL discontinuous non-differentiable non-CPT-symmetric non-unitary non-local-in-the-configuration-space etc. etc. process in all of physics, to explain a phenomenon (why do we see only one outcome?) that doesn’t need explaining.
Well, one advantage of it is that it is testable, and so is not a mere interpretation, which holds a certain amount of appeal to the more old-fashioned of us.
I agree, and I myself was, and am still, sentimentally fond of Penrose for this reason, and I would cheer on any agency that funded a test. However and nonetheless, “testable” is not actually the same as “plausible”, scientifically virtuous as it may be.
Not if it doesn’t allow FTL communication, unless you want to argue that quantum entanglement is a FTL phenomenon, but that wouldn’t be an issue of the particular interpretation.
Not necessarily. Irreversible and stochastic quantum processes can be time-continuous and time-differentiable.
Consider the processes described by the Lindblad equation, for instance.
CPT symmetry is a property of conventional field theories, not all quantum theories necessarily have it, and IIUC, there are ongoing experiments to search for violations. CPT symmetry is just the last of a series of postulated symmetries, the previous ones (C symmetry, P symmetry, T symmetry and CP symmetry) have been experimentally falsified.
Right, and that’s the point of objective collapse theories.
I’m not sure what you mean by that, but locality in physics is defined with respect to space and time, not to arbitrary configuration spaces.
AFAIK, there have been attempts to derive the Born rule in Everett’s interpretation, but they didn’t lead to uncontroversial results.
I have never seen a proposed mechanism of ontological collapse that actually fits this, though.
The inability to send a signal that you want, getting instead a Born-Rule-based pure random signal, doesn’t change that this Born-Rule-based pure random signal is, under ontological collapse distributed FTL.
AFAIK, Penrose’s interpretation doesn’t describe the details of the collapse process, it just says that above about the “one graviton” level of energy separation collapse will occur.
It doesn’t commit to collapse being instantaneous: It could be that the state evolution is governed by a non-linear law that approximates very well the linear Schrödinger equation in the “sub-graviton” regime and has a sharp, but still differentiable phase transition when approaching the “super-graviton” regime.
The GRW interpretation assumes instantaneous collapse, IIUC, but it would be a trivial modification to have fast, differentiable collapse.
My point is that non-differentiable collapse is not a requirement of objective collapse interpretations.
But that’s an issue of QM, irrespective of the particular interpretation. Indeed the “spooky action at distance” bugged Einstein and many people of his time, but the modern view is that as long as you don’t have causal influences (that is, information transmission) propagating FTL, you don’t violate special relativity.
No, it isn’t. QM is purely causal and relativistic. You can look into the equations and prove that nothing FTL is in there. The closest you get is accounting for the possibility of a vacuum bubble having appeared nearby a particle with exactly its energy, and the antimatter part of it the bubble then cancels with the particle. And that isn’t much like FTL.
When you do an EPR experiment, the appearance of FTL communication arises from the assumption that the knowledge you gain about what you’ll see if you go check the other branch of the experiment is something happens at the other end of the experiment, instead of locally, with the information propagating to the other end of the experiment as you go to check. The existence of nonlocal states does not imply nonlocal communication.
I’m not sure what we are disagreeing about.
My point is that objective collapse is FTL only in the same sense that QM is. That is, if QM isn’t FTL, then collapse isn’t.
I’m puzzled. What does Solomonoff Induction have to say about experimentally undistinguishable (as far as we can practically test, at least) interpretations of the same theory?
But there is s case to be made for relatioal QM as superior to both MWI an collpase interpretations. I have metuioned it several times. I am still waiting to hear back.
Relational QM is gibberish. Whether the cat is dead or alive is “relative to the observer”. How could that make sense except via many worlds?
It makes sense the way rQM says: there is no non-relational state, so there is not answer to “is the cat dead or alive (absent an observer)”. Since rQM says there is no state, you don’t disprove it by insisting there is state.
BTW, there is no simultaneity either.
OK, so suppose we have an observer. Now look at the cat. Is it alive or dead? If it is alive and only alive, well, we can affix the phrase “relative to the observer” but it doesn’t diminish the absoluteness of the cat’s being alive. But if the cat is alive “relative to one observer to which it is alive”, and dead “relative to another observer to which it is dead”, how can we possibly make sense of that except in many-worlds fashion, by saying there are two cats and two observers?
If two observers measure a cat, they will get compatible results. However one observer can have less complete information (“the cat collapsed”) and another more complete (“the cat is uncollapsed”). Observers can disagree about “collapse” because that is just an issue of their information, not an objective property.
“Relational interpretation
The relational interpretation makes no fundamental distinction between the human experimenter, the cat, or the apparatus, or between animate and inanimate systems; all are quantum systems governed by the same rules of wavefunction evolution, and all may be considered “observers.” But the relational interpretation allows that different observers can give different accounts of the same series of events, depending on the information they have about the system.[11] The cat can be considered an observer of the apparatus; meanwhile, the experimenter can be considered another observer of the system in the box (the cat plus the apparatus). Before the box is opened, the cat, by nature of it being alive or dead, has information about the state of the apparatus (the atom has either decayed or not decayed); but the experimenter does not have information about the state of the box contents. In this way, the two observers simultaneously have different accounts of the situation: To the cat, the wavefunction of the apparatus has appeared to “collapse”; to the experimenter, the contents of the box appear to be in superposition. Not until the box is opened, and both observers have the same information about what happened, do both system states appear to “collapse” into the same definite result, a cat that is either alive or dead.”—WP
In the interpretation of QM, one of the divides is between ontic and epistemic interpretations of the wavefunction. Ontic interpretations of the wavefunction treat it as a thing, epistemic interpetations as an incomplete description or a tabulation of uncertainty, just like a probability distribution.
In the relational interpretation of QM, are the states understood as ontic or as epistemic? The passage you quote makes them sound epistemic: the cat knows but the observer outside the box doesn’t, so the observer outside the box uses a different wavefunction. That undoubtedly implies that the wavefunction of the observer outside the box is epistemic, not ontic; the cat knows something that the outside observer doesn’t, an aspect of reailty which is already definite even though it is not definite in the outside observer’s description.
Or at least, this ought to imply that quantum state in the relational interpretation are epistemic. However, this is never explicitly stated, and instead meaningless locutions are adopted which make it sound as if the quantum states are to be regarded as ontic, but “relative”.
There are certain very limited senses in which it makes sense to say that the state of something is relative. For example, we may be floating in space, and what is up to you may be down to me, so whether one object is above another object may be relative to an observer. But clearly such a dodge will not work for something like Schrodinger’s cat. Either the cat is alive, dead, both, or neither. It can’t be “alive for one observer and dead for another” and still be just one cat. But that is the ontological implication one gets, if “relational QM’ is interpreted as an ontic interpertation.
On the other hand, if it is an epistemic interpretation, then it still hasn’t answered the question, “what is the nature of reality? what is the physical ontology behind the formalism and the instrumental success?”
It can’t in rQM:
“However, the comparison does not lead to contradiction because the comparison is itself a physical process that must be understood in the context of quantum mechanics. Indeed, O′ can physically interact with the electron and then with the l.e.d. (or, equivalently, the other way around). If, for instance, he finds the spin of the electron up, quantum mechanics predicts that he will then consistently find the l.e.d. on (because in the first measurement the state of the composite system collapses on its [spin up/l.e.d. on] component). That is, the multiplicity of accounts leads to no contradiction precisely because the comparison between different accounts can only be a physical quantum interaction. This internal self-consistency of the quantum formalism is general, and it is perhaps its most remarkable aspect. This self consistency is taken in relational quantum mechanics as a strong indication of the relational nature of the world”—SEP
There were plenty of physicists reading those posts when they first came out on OB (the most famous name being Scott Aaronson). Some later readers have indeed asserted that there’s a problem involving a physically wrong factor of i in the first couple of posts (i.e. that’s allegedly not what a half-silvered mirror does to the phase in real life), which I haven’t yet corrected because I would need to verify with a trusted physicist that this was correct, and then possibly craft new illustrations instead of using the ones I found online, and this would take up too much time relative to the point that talking about a phase change of −1 instead of i so as to be faithful to real-world mirrors is an essentially trivial quibble which has no effect on any larger points. If anyone else wants to rejigger the illustration or the explanation so that it flows correctly, and get Scott Aaronson or another known trusted physicist to verify it, I’ll be happy to accept the correction.
Aside from that, real physicists haven’t objected to any of the math, which I’m actually pretty darned proud of considering that I am not a physicist.
As Scott keeps saying, he’s not a physicist! He’s a theoretical computer scientist with a focus on quantum computing. He clearly has very relevant expertise, but you should get his field right.
I still wonder why you haven’t written a update in 4 years regarding this topic. Especially in regards to the Born Rule probability not having a solution yet + the other problems.
You also have the issue of overlap vs non-overlapping of worlds, which again is a relevant issue in the Many Worlds interpretation. Overlap = the typical 1 world branching into 2 worlds. Non-overlap = 2 identical worlds diverging (Saunders 2010, Wilson 2005-present)
Also I feel like the QM sequence is a bit incomplete when you do not give any thought to things like Gerard ’t Hoofts proposal of a local deterministic reality giving rise to quantum mechanics from a cellular automaton at the planck scale? It’s misleading to say the MWI is “a slam dunk” winner when there are so many unanswered questions. Mitchell Porter is one of the few persons here who seem to have a deep understanding of the subject before reading your sequence, so he has raised some interesting points...
I agree that EY is probably overconfident in MWI, although I’m uniformed about QM so I can’t say much with confidence. I don’t think it’s accurate to damn all of Less Wrong because of this. For example, this post questioning the sequence was voted up highly.
I don’t think EY claims to have any original insights pointing to MWI. I think he’s just claiming that the state of the evidence in physics is such that MWI is obviously correct, and this is evidence as to the irrationality of physicists. I’m not too sure about this myself.
Well there have been responses to that point (here’s one). I wish you’d be a bit more self-skeptical and actually engage with that (ongoing) debate instead of summarizing your view on it and dismissing LW because it largely disagrees with your view.
It seems a bit bizarre to say I’ve dismissed LessWrong given how much time I’ve spent here lately.
Fair enough.
Your examples seem … how do I put this … unreliable. The first two are less examples and more insults, since you do not provide any actual examples of these tendencies; the last one would be more serious, if he hadn’t written extensively on why he believes this to be the safest way—the only way that isn’t suicidal—or if you had provided some evidence that his FAI proposals are “extremely dangerous”. And, of course, airily proclaiming that this is true of “pretty much every entry in the sequences” seems, in the context of these examples, like an overgeneralization at best and … well, I’m not going to bother outlining the worst possible interpretation for obvious reasons.