4 ms·

The Everett school of thought is that every time you flip a coin, you create a pair of parallel universes. That everything that could ever have happened did in

by raldi 5y ago

The Everett school of thought is that every time you flip a coin, you create a pair of parallel universes. That everything that could ever have happened did in fact happen somewhere in the multiverse. The Bayesians believe you can look at past results to make good guesses about what will happen in the future. I must be missing something, because I don't understand how anyone could disagree, but apparently Karl Popper famously does. David Deutsch also apparently disagrees, and has written up an ostensibly simple explanation here: http://www.daviddeutsch.org.uk/2014/08/simple-refutation-of-the-bayesian-philosophy-of-science/ http://www.daviddeutsch.org.uk/2014/08/simple-refutation-of-...

klyrs 5y ago
I'm not one to gamble, but if any Popperians want to play some blackjack, hmu
GavinMcG 5y ago
It's more about deductive vs. inductive reasoning, rather than whether past results matter. Popper distinguished science from non-science by falsifiability: if a prediction could theoretically be disproved, it's a scientific one. New information is often useless, because new information can never prove a claim—the only useful information is disproof. On a larger scale, models that turn out to make false predictions are either altered or abandoned. It's all well and good to look at past results and make predictions, but Popper's point was that there can be plenty of models that are consistent with past results and there's no way of choosing between them until they're winnowed down through falsification. Bayesianism acknowledges that you can't necessarily pick one right explanation, but it tries to account for that by statistically weighting models, and with new information the distribution/weighting is updated. (This explanation is a gloss, of course, and so necessarily an incorrect model. But hopefully a helpful one.)
- mirceal 5y ago
  Popper was a cool guy and his ideas on falsifiability definitely moved the needle and gave us a robust scientific method. But… what do you do when you no longer can directly measure or even worse: the measurements impact the state of the system you are trying to observe? I think for things like quantum mechanics and other areas that are hard to observe directly (neuroscience comes to mind) there has to be a revision of how Popper’s principles are applied.
  - GavinMcG 5y ago
    That revision is probably expressed in Lakatos's view on "research programmes" that may have core non-falsifiable theories and that continue being useful even when scientists know something is off about the model as a whole: https://plato.stanford.edu/entries/lakatos/#ImprPoppScie https://plato.stanford.edu/entries/lakatos/#ImprPoppScie
- raldi 5y ago
  Popper: The statement, "There are no purple cows", can never be proven, but it could be disproven by presenting one. Fuchs: [Need some help with this one] Deutsch: [This one too] Yudkowsky: Based on the fact that we've studied cows all over the planet for millennia, and have never seen a verified report of a purple one, we can conclude with great certainty that there are no purple ones.
scythe 5y ago
>The Bayesians believe you can look at past results to make good guesses about what will happen in the future. I must be missing something, because I don't understand how anyone could disagree, but apparently Karl Popper famously does. Deutsch and Popper disagree with the notion of recording specific "probabilities" as (usually rational) numbers between 0 and 1 for the occurrence of specific future events. This is because the product law for coincidence and the law of the excluded middle lead to a contradiction when you try to do this. Similar arguments have been made elsewhere, e.g.: https://faculty.fuqua.duke.edu/~rnau/definettiwasright.pdf https://faculty.fuqua.duke.edu/~rnau/definettiwasright.pdf Looking at past results to make predictions about what will happen in the future is something everyone does, but that doesn't mean that the means of doing so produces a list of values in [0,1] updated in a very specific way.
civilized 5y ago
Bayes' Rule is a basic mathematical theorem in probability. Everyone agrees that the theorem is true and useful, but people disagree on its practical scope. The more fundamental and broadly applicable you think it is to practical real-world problems, the more Bayesian you are.
unholiness 5y ago
Both "every time you flip a coin, you create a pair of parallel universes" and "That everything that could ever have happened did in fact happen" are both fairly big misinterpretations of many worlds theory – both in fact (there's no real connection between quantum measurement process and a coin flip or other macro-scale outcomes), and in reasoning (the theory does not posit many worlds, they came out organically from the math). Everettians and non-Everettians agree that 1) The state of the universe is fully described by a set of quantum fields 2) The quantum fields will change over time according to the Schrödinger equation. So far, this is analogous to how classical physics is set up, but in classical physics the "state" is a bunch of particles with positions and momentums, and the "change over time" is F=ma (for all forces). Non-Everettians would say those two rules alone do not explain experimental results. We need extra rules, namely: 3) When a "measurement" occurs, the state of the quantum field collapses into a single definite value, and 4) The probability of the outcome of that measurement is given by the Born rule The main issues with non-Everettian interpretations are that 1) The definition of a measurement is vague, and 2) you need some theory of why the collapse occurs, and what exactly is happening as the measurement is being taken (there are many options here). Everettians instead ask, what would happen if we just... didn't have those extra rules? If so, the Schödinger equation on its own predicts the following evolution of quantum state during an experiment: To begin we have an "uncollapsed" quantum state. Say, an electron in a superposition of spin-up and spin-down, not entangled with anything else During the experiment, the electron becomes entangled with other particles in the measurement apparatus By the time the measurement is complete, we have full entanglement, the new quantum state is a superposition of "spin-up and a small current induced in the top plate" and "spin-down and a small current induced in the bottom plate" When the experimenter reads the result, we have even more entanglement into a superposition of "spin-up and the experimenter sees it's spin up" and "spin-down and the experimenter sees it's spin down" There are cascading interactions with every particle in the universe until the whole universe is entangled into a superposition of "spin-up and the universe is as it would be if the particle were spin-up" and "spin-down and the universe is as it would be if the particle were spin-down" At this point an Everettian would say: this is exactly consistent with what non-Everettians were trying to show with their extra rules! The quantum field has evolved into a superposition of two universes, and the Schrödinger equation predicts those two universes never interact with each other... so there's really no evidence that the Schrödinger equation ever needs to not hold true. Quantum collapse happens on its own when you just run the Schrödinger equation forward, and the extra universes are just an unobservable byproduct. The main issues with the Everettian interpretation are epistemic: I don't know what it should feel like to be in a superposition of two non-interacting universes with different states of knowledge, but to be an Everettian, you need to believe it feels like just being in one of them. Depending on your philosophy of consciousness, it's anywhere from trivial to impossible to prove that.

by raldi 5y ago

klyrs 5y ago

I'm not one to gamble, but if any Popperians want to play some blackjack, hmu

GavinMcG 5y ago

It's more about deductive vs. inductive reasoning, rather than whether past results matter. Popper distinguished science from non-science by falsifiability: if a prediction could theoretically be disproved, it's a scientific one. New information is often useless, because new information can never prove a claim—the only useful information is disproof. On a larger scale, models that turn out to make false predictions are either altered or abandoned. It's all well and good to look at past results and make predictions, but Popper's point was that there can be plenty of models that are consistent with past results and there's no way of choosing between them until they're winnowed down through falsification. Bayesianism acknowledges that you can't necessarily pick one right explanation, but it tries to account for that by statistically weighting models, and with new information the distribution/weighting is updated. (This explanation is a gloss, of course, and so necessarily an incorrect model. But hopefully a helpful one.)

mirceal 5y ago
Popper was a cool guy and his ideas on falsifiability definitely moved the needle and gave us a robust scientific method. But… what do you do when you no longer can directly measure or even worse: the measurements impact the state of the system you are trying to observe? I think for things like quantum mechanics and other areas that are hard to observe directly (neuroscience comes to mind) there has to be a revision of how Popper’s principles are applied.
- GavinMcG 5y ago
  That revision is probably expressed in Lakatos's view on "research programmes" that may have core non-falsifiable theories and that continue being useful even when scientists know something is off about the model as a whole: https://plato.stanford.edu/entries/lakatos/#ImprPoppScie https://plato.stanford.edu/entries/lakatos/#ImprPoppScie
raldi 5y ago
Popper: The statement, "There are no purple cows", can never be proven, but it could be disproven by presenting one. Fuchs: [Need some help with this one] Deutsch: [This one too] Yudkowsky: Based on the fact that we've studied cows all over the planet for millennia, and have never seen a verified report of a purple one, we can conclude with great certainty that there are no purple ones.

scythe 5y ago

>The Bayesians believe you can look at past results to make good guesses about what will happen in the future. I must be missing something, because I don't understand how anyone could disagree, but apparently Karl Popper famously does. Deutsch and Popper disagree with the notion of recording specific "probabilities" as (usually rational) numbers between 0 and 1 for the occurrence of specific future events. This is because the product law for coincidence and the law of the excluded middle lead to a contradiction when you try to do this. Similar arguments have been made elsewhere, e.g.: https://faculty.fuqua.duke.edu/~rnau/definettiwasright.pdf https://faculty.fuqua.duke.edu/~rnau/definettiwasright.pdf Looking at past results to make predictions about what will happen in the future is something everyone does, but that doesn't mean that the means of doing so produces a list of values in [0,1] updated in a very specific way.

civilized 5y ago

Bayes' Rule is a basic mathematical theorem in probability. Everyone agrees that the theorem is true and useful, but people disagree on its practical scope. The more fundamental and broadly applicable you think it is to practical real-world problems, the more Bayesian you are.

unholiness 5y ago

Both "every time you flip a coin, you create a pair of parallel universes" and "That everything that could ever have happened did in fact happen" are both fairly big misinterpretations of many worlds theory – both in fact (there's no real connection between quantum measurement process and a coin flip or other macro-scale outcomes), and in reasoning (the theory does not posit many worlds, they came out organically from the math). Everettians and non-Everettians agree that 1) The state of the universe is fully described by a set of quantum fields 2) The quantum fields will change over time according to the Schrödinger equation. So far, this is analogous to how classical physics is set up, but in classical physics the "state" is a bunch of particles with positions and momentums, and the "change over time" is F=ma (for all forces). Non-Everettians would say those two rules alone do not explain experimental results. We need extra rules, namely: 3) When a "measurement" occurs, the state of the quantum field collapses into a single definite value, and 4) The probability of the outcome of that measurement is given by the Born rule The main issues with non-Everettian interpretations are that 1) The definition of a measurement is vague, and 2) you need some theory of why the collapse occurs, and what exactly is happening as the measurement is being taken (there are many options here). Everettians instead ask, what would happen if we just... didn't have those extra rules? If so, the Schödinger equation on its own predicts the following evolution of quantum state during an experiment: To begin we have an "uncollapsed" quantum state. Say, an electron in a superposition of spin-up and spin-down, not entangled with anything else During the experiment, the electron becomes entangled with other particles in the measurement apparatus By the time the measurement is complete, we have full entanglement, the new quantum state is a superposition of "spin-up and a small current induced in the top plate" and "spin-down and a small current induced in the bottom plate" When the experimenter reads the result, we have even more entanglement into a superposition of "spin-up and the experimenter sees it's spin up" and "spin-down and the experimenter sees it's spin down" There are cascading interactions with every particle in the universe until the whole universe is entangled into a superposition of "spin-up and the universe is as it would be if the particle were spin-up" and "spin-down and the universe is as it would be if the particle were spin-down" At this point an Everettian would say: this is exactly consistent with what non-Everettians were trying to show with their extra rules! The quantum field has evolved into a superposition of two universes, and the Schrödinger equation predicts those two universes never interact with each other... so there's really no evidence that the Schrödinger equation ever needs to not hold true. Quantum collapse happens on its own when you just run the Schrödinger equation forward, and the extra universes are just an unobservable byproduct. The main issues with the Everettian interpretation are epistemic: I don't know what it should feel like to be in a superposition of two non-interacting universes with different states of knowledge, but to be an Everettian, you need to believe it feels like just being in one of them. Depending on your philosophy of consciousness, it's anywhere from trivial to impossible to prove that.