= 27d567bc2d48d9f40e9ccc20ea370b15)
Eliezer_Yudkowsky comments on Welcome to Less Wrong! (5th thread, March 2013) - 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 (1750)
Bad advice for technical readers. Mihaly Barasz (IMO gold medalist) got here via HPMOR but only became seriously interested in working for MIRI after reading the QM sequence.
Given those particular circumstances, can I ask that you stop with that particular bit of helpful advice?
Do you have a solid idea of how many technical readers get here via HPMOR but become disinterested in working for MIRI after reading the QM sequence? If not, isn't this potentially just the selection effect?
EY can rationally prefer the certain evidence of some Mihaly-Barasz-caliber researchers joining when exposed to the QM sequence
over
speculations whether the loss of Mihaly Barasz (had he not read the QM sequence) would be outweighed by even more / better technical readers becoming interested in joining MIRI, taking into account the selection effect.
Personally, I'd go with what has been proven/demonstrated to work as a high-quality attractor.
Yep. I also tend to ignore nontechnical folks along the lines of RationalWiki getting offended by my thinking that I know something they don't about MWI. Carl often hears about, anonymizes, and warns me when technical folks outside the community are offended by something I do. I can't recall hearing any warnings from Carl about the QM sequence offending technical people.
Bluntly, if shminux can't grasp the technical argument for MWI then I wouldn't expect him to understand what really high-class technical people might think of the QM sequence. Mihaly said the rest of the Sequences seemed interesting but lacked sufficient visible I-wouldn't-have-thought-of-that nature. This is very plausible to me - after all, the Sequences do indeed seem to me like the sort of thing somebody might just think up. I'm just kind of surprised the QM part worked, and it's possible that might be due to Mihaly having already taken standard QM so that he could clearly see the contrast between the explanation he got in college and the explanation on LW. It's a pity I'll probably never have time to write up TDT.
I have a phd in physics (so I have at least some technical skill in this area) and find the QM sequence's argument for many worlds unconvincing. You lead the reader toward a false dichotomy (Copenhagen or many worlds) in order to suggest that the low probability of copenhagen implies many worlds. This ignores a vast array of other interpretations.
Its also the sort of argument that seems very likely to sway someone with an intro class in college (one or two semesters of a Copenhagen based shut-up-and-calculate approach), precisely because having seen Copenhagen and nothing else they 'know just enough to be dangerous', as it were.
For me personally, the quantum sequence threw me into some doubt about the previous sequences I had read. If I have issues with the area I know the most about, how much should I trust the rest? Other's mileage may vary.
Actually, attempting to steelman the QM Sequence made me realize that the objective collapse models are almost certainly wrong, due to the way they deal with the EPR correlations. So the sequence has been quite useful to me.
On the other hand, it also made me realize that the naive MWI is also almost certainly wrong, as it requires uncountable worlds created in any finite instance of time (unless I totally misunderstand the MWI version of radioactive decay, or any emission process for that matter). It has other issues, as well. Hence my current leanings toward some version of RQM, which EY seems to dislike almost as much as his straw Copenhagen, though for different reasons.
Right, I've had a similar experience, and I heard it voiced by others.
As a result of re-examining EY's take on epistemology of truth, I ended up drifting from the realist position (map vs territory) to an instrumentalist position (models vs inputs&outputs), but this is a topic for another thread. I am quite happy with the sequences related to cognitive science, where, admittedly, I have zero formal expertise. But they seem to match what the actual experts in the field say.
I am on the fence with the free-will "dissolution", precisely because I know that I am not qualified to spot an error and there is little else out there in terms of confirming evidence or testable predictions.
I am quite skeptical about the dangers of AGI x-risk, mainly because it seems to extrapolate too far beyond what is known into the fog of the unknown future, though do I appreciate quite a few points made in the relevant sequences. Again, I am not qualified to judge their validity.
How is that any more problematic than doing physics with real or complex numbers in the first place?
It means that EY's musings about the Eborians splitting into the world's of various thicknesses according to Born probabilities no longer make any sense. There is a continuum of worlds, all equally and infinitesimally thin, created every picosecond.
The way I understand it, it's not that “new” worlds are created that didn't previously exist (the total “thickness” (measure) stays constant). It's that two worlds that looked the same ten seconds ago look different now.
That's a common misconception. In the simplest case of the Schrodinger' cat, there are not just two worlds with cat is dead or cat is alive. When you open the box, you could find the cat in various stages of decomposition, which gives you uncountably many worlds right there. In a slightly more complicated version, where energy and the direction of the decay products are also measurable (and hence each possible value is measured in at least one world), your infinities keep piling up every which way, all equally probable or nearly so.
coughmeasurecough
Is there actually any physicists that find QM sequence to be making such a strongly compelling case for MWI as EY says it does?
I know Mitchell Porter is likewise a physicist and he's not convinced at all either.
Mitchell Porter also advocates Quantum Monadology and various things about fundamental qualia. The difference in assumptions about how physics (and rational thought) works between Eliezer (and most of Eliezer's target audience) and Mitchell Porter is probably insurmountable.
Yeah, and EY [any of the unmentionable things].
For other point, scott aaronson doesn't seem convinced either. Robin Hanson, while himself (it seems) a MWI believer but doesn't appear to think that its so conclusively settled.
The relevance of Porter's physics beliefs is that any reader who disagrees with Porter's premises but agrees with the premises used in an article can gain little additional information about the quality of the article by learning that Porter is not convinced by it. ie. Whatever degree of authority Mitchell Porter's status grants goes (approximately) in the direction of persuading the reader to adopt those different premises.
In this way mentioning Porter's beliefs is distinctly different from mentioning the people that you now bring up:
I defected from physics during my Master's, but this is basically the impression I had of the QM sequence as well.
That sounds like reasonable evidence against the selection effect.
I strongly recommend against both the "advises newcomers to skip the QM sequence -> can't grasp technical argument for MWI" and "disagrees with MWI argument -> poor technical skill" inferences.
That inference isn't made. Eliezer has other information from which to reach that conclusion. In particular, he has several years worth of ranting and sniping from Shminux about his particular pet peeve. Even if you disagree with Eliezer's conclusion it is not correct to claim that Eliezer is making this particular inference.
Again, Eliezer has a large body of comments from which to reach the conclusion that Shminux has poor technical skill in the areas necessary for reasoning on that subject. The specific nature of the disagreement would be relevant, for example.
That very well could be, in which case my recommendation about that inference does not apply to Eliezer.
I will note that this comment suggests that Eliezer's model of shminux may be underdeveloped, and that caution in ascribing motives or beliefs to others is often wise.
It really doesn't. At best it suggests Eliezer could have been more careful in word selection regarding Shminux's particular agenda. 'About' rather than 'with' would be sufficient.
Shminux's and Eliezer?
Don't do that. I think the rest of your post is fine, but this is not a debate-for-debate's-sake kind of place (and even if it were, that's not a winning move).
Please change your posting style or leave lesswrong. Not only is disingenuous rhetoric not welcome, your use thereof doesn't even seem particularly competent.
ie. What the heck? You think that the relevance of authority isn't obvious to everyone here and is a notion sufficiently clever to merit 'traps'? You think that forcing someone to repeat what is already clear and already something they plainly endorse even qualifies as entrapment? (It's like an undercover Vice cop having already been paid for a forthcoming sexual favor demanding "Say it again! Then I'll really have you!")
Did you not notice that even if you proved Eliezer's judgement were a blatant logical fallacy it still wouldn't invalidate the point in the comment you are directing your 'trap' games at? The comment even explained that explicitly.
If I ever have cause to send Shminux a letter I will be sure to play proper deference to his status by including "Dr." as the title. Alas, Shminux's arguments have screened off his authority, and then some.
"No rational grounds" means a different thing than "the particular evidence I mention points in the other direction". That difference matters rather a lot.
"Rational grounds" includes all Bayesian evidence... such things as costly affiliation signals (PhDs) and also other forms of evidence---including everything the PhD in question has said. Ignoring the other evidence would be crazy and lead to poor conclusions.
I'm no IMO gold medalist (which really just means I'm giving you explicit permission to ignore the rest of my comment) but it seems to me that a standard understanding of QM is necessary to get anything out of the QM sequence.
Revealed preferences are rarely attractive.
Adds to "Things I won't actually get put on a T-shirt but sort of feel I ought to" list.
As others noted, you seem to be falling prey to the selection bias. Do you have an estimate of how many "IMO gold medalists" gave up on MIRI because its founder, in defiance of everything he wrote before, confidently picks one untestable from a bunch and proclaims it to be the truth (with 100% certainty, no less, Bayes be damned), despite (or maybe due to) not even being an expert in the subject matter?
EDIT: My initial inclination was to simply comply with your request, probably because I grew up being taught deference to and respect for authority. Then it struck me as one of the most cultish things one could do.
Is this an April Fool's joke? He says nothing of the kind. The post which comes closest to this explicitly says that it could be wrong, but "the rational probability is pretty damned small." And counting the discovery of time-turners, he's named at least two conceivable pieces of evidence that could change that number.
What do you mean when you say you "just don't put nearly as much confidence in it as you do"?
Maybe it's a reference to the a priori nature of his arguments for MW? Or something? It's a strange claim to make, TBH.
The number of IMO gold medalists is sufficiently low, and the probability of any one of them having read the QM sequence is sufficiently small, that my own estimate would be less than one regardless of X.
(I don't have a good model of how much more likely an IMO gold medalist would be to have read the QM sequence than any other reference class, so I'm not massively confident.)
There's plenty of things roughly comparable to IMO in terms of selectivity (IMO gives what, ~35 golds a year?)... E.g. I'm #10th of all time on a popular programming contest site ( I'm dmytry ).
This discussion is really hilarious, especially the attempts to re-frame commoner orientated, qualitative and incomplete picture of QM - as something which technical people appreciate and non-technical people don't. (Don't you want to be one among the technies?) .
Selectivity, in the relevant sense, is more than just a question of how many people are granted something.
How many people are not on that site, but could rank highly if they chose to try? I'm guessing it's far more than the number of people who have never taken part in the IMO, but who could get a gold medal if they did.
(The IMO is more prestigious among mathematicians than topcoder is among programmers. And countries actively recruit their best mathematicians for the IMO. Nobody in the Finnish government thought it would be a good idea to convince and train Linus Torvalds to take part in an internet programming competition, so I doubt Linus Torvalds is on topcoder.)
There certainly are things as selective or more than the IMO (for example, the Fields medal), but I don't think topcoder is one of them, and I'm not convinced about "plenty". (Plenty for what purpose?)
I've tried to compare it more accurately.
It's very hard to evaluate selectivity. It's not just the raw number of people participating. It seems that large majority of serious ACM ICPC participants (both contestants and their coaches) are practising on Topcoder, and for the ICPC the best college CS students are recruited much the same as best highschool math students for IMO.
I don't know if Linus Torvalds would necessarily do great on this sort of thing - his talents are primarily within software design, and his persistence as the unifying force behind Linux. (And are you sure you'd recruit a 22 years old Linus Torvalds who just started writing a Unix clone?). It's also the case that 'programming contest' is a bit of misnomer - the winning is primarily about applied mathematics - just as 'computer science' is a misnomer.
In any case, its highly dubious that understanding of QM sequence is as selective as any contest. I get it fully that Copenhagen is clunky whereas MWI doesn't have the collapse, and that collapse fits in very badly. That's not at all the issue. However badly something fits, you can only throw it away when you figured out how to do without it. Also, commonly, the wavefunction, the collapse, and other internals, are seen as mechanisms of prediction which may, or may not, have anything to do with "how universe does it" (even if the question of "how universe does it" is meaningful, it may still be the case that internals of the theory have nothing to do with that, as the internals are massively based upon our convenience). And worse still, MWI is in many very important ways lacking.
Note that the original text was "gold," not "good".
I assume IMO is the International Mathematical Olympiad(1). Not that this in any way addresses or mitigates your point; just figured I'd point it out.
(1) If I've understood the wiki article, ~35 IMO gold medals are awarded every year.
Thanks, I fixed the typo.
Huh. I'd assumed it was short for "In My Opinion".
Yeah, that confused me on initial reading, though some googling clarified matters, and I inferred from the way shminux (mis)quoted that something similar might be going on there, which is why I mentioned it.
Again, you seem to be generalizing from a single example, unless you have more data points than just Mihaly.
QM Sequence is two parts:
(1) QM for beginners
(2) Philosophy-of-science on believing things when evidence is equipoise (or absent) - pick the simpler hypothesis.
I got part (1) from reading Dancing Wu-Li Masters, but I can clearly see the value to readers without that background. But teaching foundational science is separate from teaching Bayesian rationalism.
The philosophy of the second part is incredibly controversial. Much more than you acknowledge in the essays, or acknowledge now. Treating the other side of any unresolved philosophical controversy as if it is stupid, not merely wrong, is excessive and unjustified.
In short, the QM sequence would seriously benefit from the sort of philosophical background stuff that is included in your more recent essays. Including some more technical discussion of the opposing position.
If you learned quantum mechanics from that book, you may have seriously mislearned it. It's actually pretty decent describing everything up to but excluding quantum physics. When it comes to QM, however, the author sacrifices useful understanding in favor of mysticism.
Hrm? On a conceptual level, is there more to QM than the Uncertainty Principle and Wave-Particle Duality? DWLM mentions the competing interpretations, but choosing an interpretation is not strictly necessary to understand QM predictions.
For clarity, I consider the double-slit experimental results to be an expression of wave-particle duality.
I will admit that DWLM does a poor job of preventing billiard-ball QM theory ("Of course you can't tell momentum and velocity at the same time. The only way to check is to hit the particle with a proton, and that's going to change the results.").
That's a wrong understanding, but a less wrong understanding than "It's classical physics all the way down."
Yes. Very yes. There are several different ways to get at that next conceptual level (matrix mechanics, the behavior of the Schrödinger equation, configuration spaces, Hamiltonian and Lagrangian mechanics, to name ones that I know at least a little about), but qualitative descriptions of the Uncertainty Principle, Schrödinger's Cat, Wave-Particle Duality, and the Measurement Problem do not get you to that level.
Rejoice—the reality of quantum mechanics is way more awesome than you think it is, and you can find out about it!
Let me rephrase: I'm sure there is more to cutting edge QM than that which I understand (or even have heard of). Is any of that necessary to engage with the philosophy-of-science questions raised by the end of the Sequence, such as Science Doesn't Trust Your Rationality?
From a writing point of view, some scientific controversy needed to be introduced to motivate the later discussion - and Eliezer choose QM. As examples go, it has advantages:
(1) QM is cutting edge - you can't just go to Wikipedia to figure out who won. EY could have written a Lamarckian / Darwinian evolution sequence with similar concluding essays, but indisputably knowing who was right would slant how the philosophy-of-science point would be interpreted.
(2) A non-expert should recognize that their intuitions are hopelessly misleading when dealing with QM, opening them to serious consideration of the new-to-them philosophy-of-science position EY articulates.
But let's not confuse the benefits of the motivating example with arguing that there is philosophy-of-science benefit in writing an understandable description of QM.
In other words, if the essays in the sequence after and including The Failures of Eld Science were omitted from the Sequence, it wouldn't belong on LessWrong.
A deeper, more natural way to express both is "wavefunction reality," which also incorporates some of the more exotic effects that come from using complex numbers. (The Uncertainty Principle also should be called the "uncertainty consequence," since it's a simple derivation from how the position and momentum operators work on wavefunctions.)
(I haven't read DWLM, so I can't comment on its quality.)