Somewhat exotic and recent example from my field: Don Rubin’s claim that adjusting for covariates always reduces confounding bias (it does not). There is a famous Pearl/Rubin argument about this. There is some discussion on Andrew Gelman’s blog about it, starting here: http://andrewgelman.com/2009/07/disputes_about/. It’s a pretty interesting read (perhaps less so if you aren’t familiar with the question or personalities involved), if for no other reason than the anatomy of modern academic disagreements.
Not sure this is a good example. Einstein didn’t deny the predictive utility of quantum mechanics, he denied that QM is a complete description of reality. He believed that there were hidden variables that would account for quantum statistics in a local and deterministic manner. Of course we now know that this is impossible, but its impossibility only became clear with Bell’s work in the 60s, after Einstein’s death. While Einstein was alive, I’m not aware of any conclusive reason to regard his hope for a local deterministic theory as futile.
Not to mention that Einstein was perfectly right about a correct physics containing no randomness and no mysteriously-non-communicating FTL influences, which at the time was part of the then-dominant Copenhagen interpretation of QM. Basically, everything that made Einstein throw up got thrown out. His intuitions were accurate.
Einstein was actually against what is now termed “objective collapse” models, not Bohr’s “shut up and calculate”. And yes, the hints are increasingly pointing toward something along the lines of RQM.
Einstein was perfectly right about a correct physics containing no randomness
Maybe you should try playing your own rationalist’s taboo with the term “randomness”. For example, if you define it instrumentally as inability to predict an outcome of a measurement, then it trivially goes through in the MWI model. What’s your definition?
no mysteriously-non-communicating FTL influences
Try similarly tabooing “influence”. It is likely that you will find that “non-communicating influence” is devoid of meaning (it’s a piece of logic disconnected from physics, using your favorite duality).
Ok, fair. Does the lesswrong dialect of Markdown support strikeouts?
edit: Hidden variable intuitions are fairly interesting. It must have seemed strange to Einstein, but these days it’s not so strange that you could have descriptions of objects that behave as a “latent variable model” of sorts without there being any hidden variables in its description. You don’t even need to go to quantum theory to find such objects. Here’s a simple graphical model (we call them “MAGs”):
A → B <-> C ← D
The way people usually interpret this model is that there is some DAG in the background like this:
A → B ← U → C ← D
and we then do not get to observe U. But you can think of another description of this model, which is that it is all probability distributions where the following independences hold:
(1) A _||_ C,D
(2) D _||_ A,B
Nothing in this description mentions U. We can even parameterize this model (say variables are binary) by a set of parameters that look like this:
q(a), q(d), q(b|a), q(c|d), q(b,c|a,d). Again, nothing in these parameters mentions U. And yet the model resembles a hidden variable DAG with a U in the pattern of constraints it imposes. And we know there is no U in the model, because if there was, there would be a Bell inequality, which there isn’t as the only constraints are (1) and (2).
Probably most people would add Jaynes position on QM to this. I believe he died still questioning Bell’s Theorem, and not accepting ontological randomness.
I actually think that’s correct, but I think I’m in the minority on that one.
It may be correct, but not for the reasons usually given: the non existence of ontological randomness is in now way entailed buy the existence of epistemic inderminism.
Einstein + quantum mechanics.
Somewhat exotic and recent example from my field: Don Rubin’s claim that adjusting for covariates always reduces confounding bias (it does not). There is a famous Pearl/Rubin argument about this. There is some discussion on Andrew Gelman’s blog about it, starting here: http://andrewgelman.com/2009/07/disputes_about/. It’s a pretty interesting read (perhaps less so if you aren’t familiar with the question or personalities involved), if for no other reason than the anatomy of modern academic disagreements.
Not sure this is a good example. Einstein didn’t deny the predictive utility of quantum mechanics, he denied that QM is a complete description of reality. He believed that there were hidden variables that would account for quantum statistics in a local and deterministic manner. Of course we now know that this is impossible, but its impossibility only became clear with Bell’s work in the 60s, after Einstein’s death. While Einstein was alive, I’m not aware of any conclusive reason to regard his hope for a local deterministic theory as futile.
Not to mention that Einstein was perfectly right about a correct physics containing no randomness and no mysteriously-non-communicating FTL influences, which at the time was part of the then-dominant Copenhagen interpretation of QM. Basically, everything that made Einstein throw up got thrown out. His intuitions were accurate.
Einstein backed local realism and the ensemble interpretation, both of which have been “thrown out”.
Einstein was actually against what is now termed “objective collapse” models, not Bohr’s “shut up and calculate”. And yes, the hints are increasingly pointing toward something along the lines of RQM.
Maybe you should try playing your own rationalist’s taboo with the term “randomness”. For example, if you define it instrumentally as inability to predict an outcome of a measurement, then it trivially goes through in the MWI model. What’s your definition?
Try similarly tabooing “influence”. It is likely that you will find that “non-communicating influence” is devoid of meaning (it’s a piece of logic disconnected from physics, using your favorite duality).
rolls eyes at RQM (due to physicists trying to play silly semantic games that don’t actually translate into any coherent epistemology)
Okay. “Without causal graphs that violate the Markov condition.”
See above.
If someone could draw this graph for me for the case of the purported EPR FTL influence without relying on objective collapse, I’d appreciate it.
Physical indeterminism is still an open question, so nothing got “thrown out”. You don’t get to pretend an answer you happen to like is accepted fact.
Ok, fair. Does the lesswrong dialect of Markdown support strikeouts?
edit: Hidden variable intuitions are fairly interesting. It must have seemed strange to Einstein, but these days it’s not so strange that you could have descriptions of objects that behave as a “latent variable model” of sorts without there being any hidden variables in its description. You don’t even need to go to quantum theory to find such objects. Here’s a simple graphical model (we call them “MAGs”):
A → B <-> C ← D
The way people usually interpret this model is that there is some DAG in the background like this:
A → B ← U → C ← D
and we then do not get to observe U. But you can think of another description of this model, which is that it is all probability distributions where the following independences hold:
(1) A _||_ C,D
(2) D _||_ A,B
Nothing in this description mentions U. We can even parameterize this model (say variables are binary) by a set of parameters that look like this:
q(a), q(d), q(b|a), q(c|d), q(b,c|a,d). Again, nothing in these parameters mentions U. And yet the model resembles a hidden variable DAG with a U in the pattern of constraints it imposes. And we know there is no U in the model, because if there was, there would be a Bell inequality, which there isn’t as the only constraints are (1) and (2).
No. (Unless you use the “no” symbol button to retract the comment, in which case the entire comment is struck through.)
It’s incredibly annoying, too. Someone should fix that.
test
Look up what Einstein’s Nobel was for. It wasn’t for relativity.
Probably most people would add Jaynes position on QM to this. I believe he died still questioning Bell’s Theorem, and not accepting ontological randomness.
I actually think that’s correct, but I think I’m in the minority on that one.
It may be correct, but not for the reasons usually given: the non existence of ontological randomness is in now way entailed buy the existence of epistemic inderminism.