You know what, this is interesting. I would love to be proven wrong.

I searched for that paper but couldn't find it, nor any description of the experiment it proposes. I wrote an email to professor Deutsch asking him to send me a pdf copy. I will get back to you if/when he responds.

I still kind of suspect that the paper will make some assumptions that will effectively mean "if <magic> then we could test the MWI in the following way: ...".

He has a book. I haven't read it, but I'm sure he must discuss this issue in it.

In any case, I'm sure you can come up with a definition of "measurement" that might make the Copenhagen Interpretation experimentally indistinguishable from MWI, but why? MWI is and always will be a simpler theory, and thus preferred by Occam's razor.

The problem with the Copenhagen Interpretation is that it is NOT a scientific theory. It is not a scientific theory because it is not falsifiable. It also builds non-fundamental things like "measurement" right into the fundamental laws of physics, which is absurd. Copenhagen is not falsifiable, because if I were to attempt to falsify it by demonstrating that a "measurement" did not cause the wave function to collapse, you could always assert that I had used the wrong definition of "measurement".

GRW, on the other hand, can be seen as a sub-interpretation of the Copenhagen Interpretation because it rigorously defines the term "measurement". I.e., entanglement of the particles in question with a "large enough" collection of additional particles. "Large enough" here needs some experimental tuning, but some day we may be able to perform these experiments and attempt to falsify GRW. Because we can falsify GRW, it IS a scientific theory.

Most other collapse interpretations that I have heard of can likewise be seen as sub-interpretations of Copenhagen, in that they define "measurement".

One of those sub-interpretations is the Wigner Interpretation, or the "consciousness causes collapse" interpretation. Counter to your previous assertion, we could in theory experimentally determine whether this is true via trained rats: (1) Train a rat to perform measurements, (2) kill the rat before it has a chance to tell you the results, (3) check to see whether wave function has collapsed. We can do this because there are experiments that will tell you if two particles are entangled or not.

Animal consciousness not good enough for "measurement"; it needs to be human? Okay, Nazis could in theory perform this experiment, as could future evil alien overlords.

Now let's go back to MWI: I think that many people have a misconception about MWI, and perhaps this is due to its name. (If we had stuck to the name "Everett Interpretation" perhaps that would have been better.) MWI doesn't really imply multiple worlds. It implies one very complicated superimposed world. It is also the simplest theory, as it does not add the complication of wave function collapse. Furthermore, it is completely consistent with every bit of data that has ever been collected.

Another misconception is that MWI asserts that the "other worlds" that fall out of it are "real". This is not the case. MWI is agnostic on this issue. For instance, Stephen Hawking is in favor of MWI, but he doesn't like the name, because he thinks that asserting that the "other worlds" are "real", rather than just mathematical artifacts of the theory, is not something that we can scientifically know.

Executive summary: MWI is the simplest theory, and is consistent with all data. By Occam's razor, we are required to give this theory preference until we have evidence that contradicts it.

The counter argument to the above is that the ontological cost of all these many worlds (or maybe even the complicated superpositions of state) is too great, and that this somehow violates Occam's razor.

Well, first of all it doesn't, since Occam's razor these days is almost always taken to prefer the SIMPLEST THEORY, regardless of additional philosophical worries like, "It's just creepy to think that there might be so many other worlds."

Furthermore, this objection is based on a misinterpretation of MWI. MWI is completely agnostic about the ontological status of these "other worlds". It's just a mathematical formulation for making scientific predictions. There are many cases in the history of science where "creepy" things fall out of the math, if we were to grant them the status of being ontologically "real", and yet we don't reject the theories because of this. E.g., virtual particles and advanced waves. Sometimes scientists at some point decide that mathematical artifacts of theories are "real". E.g., virtual particles. And at other times, they remain just mathematical artifacts. E.g. (maybe), advanced waves.

Are the other worlds in MWI "real"? You tell me! Science cannot answer that question. This does not imply that MWI isn't the best theory.

OK I got hold of the paper, which is a very nice read, I can email it to you if you want. It's from a talk David Deutsch gave at some conference, I bet he is a very entertaining speaker. My summary won't do it justice, but anyway. The experiments are the following:

Experiment 1:

Measure the current time and call it t1. Note that you are conscious at the time you are observing the value t1. Wait. Check your watch again. It is showing a different time t2 now. You are still conscious and your consciousness is in a different state than at t1. Therefore you've detected experimentally a superposition of distinct states of human consciousness, since there exists a formulation of Quantum Mechanics in which time is a regular operator like any other observable and you've observed two different values of it.

Experiment 2:

Consider a computer which is so well isolated that interference can be observed between it's computational states. The computer is programmed to perform an algorithm which takes one bit of input, and produces one bit of output. The algorithm is very computationally expensive and takes a long time T to complete. We communicate with the computer via two observables, I (for input) and O (for output).

Prepare the initial state as a superposition of both input values 1/sqrt(2) * (|I=0> + |I=1>). After the time T, the computer will be in the state 1/sqrt(2) (|I=0,O=f(0)> + |I=1,O=f(1)>), where f(n) is the output the algorithm produces for the input n. So by measuring I and O at this point we will either learn the value f(0) or f(1), but not both. But say we are really interested in f(0) XOR f(1). Classically, it's impossible to calculate it without computing both f(0) and f(1) so it has to take the time 2T. But with our computer, which is in a quantum superposition of states, and with the help of some clever algebra, we can construct another observable R. When we measure R one of the two things happen with equal probabilities: either we get the correct value of f(0) XOR f(1), or we lose any hope of learning it from our system. We know which one happened, i.e., with probability 0.5 we will have the correct value for f(0) XOR f(1) and we will know for sure it is correct.

Since we obtained f(0) XOR f(1) in half the time, clearly there existed two parallel worlds in which two versions of the computer calculated f(0) and f(1).

(It is noted that some members of the audience objected that Experiment 2 is conceptually no different from the two-slit interference experiment. The author allows that this may be so in a sense, since indeed, the two-slit experiment alone should be enough to make it obvious that the Everett's interpretation is right, however Experiment 2 makes it even more obvious.)

Experiment 3 (simplified version):

Consider a system consisting of a spin 1/2 particle and a quantum computer running a simulation of human consciousness. Prepare the spin in the state |→> = 1/sqrt(2) (|↑>+|↓>), i.e., measuring the spin along the x axis will always show it's pointing to the right, which means that measuring the spin along the z axis may give up or down with equal probabilities. Have the conscious being in the computer measure the spin along the z axis and communicate to the outside world the fact that he/she observed one of the values 'up' or 'down' (without saying which one). Then undo all the transitions the combined system underwent, i.e., revert it to the original state (which is in principle possible for a system consisting of a quantum computer and a microscopic system). Then measure the spin of the particle along the x axis. If it shows 'right' every time (we need to repeat the whole procedure many times), then the Everett interpretation must be true, since otherwise the fact that a conscious being observed 'up' or 'down' would have caused the particle to collapse to a state in which 'left' and 'right' are equally likely.

(The original formulation of Experiment 3 was more complicated, with three spins not one and some more clever algebra, the purpose of which if I understand correctly is to prove that it's possible to communicate to the outside world the fact that a measurement along z axis was taken, without losing the ability to revert the system to the original state.)

So, there. I'll let you form your own judgement.

Counter to your previous assertion, we could in theory experimentally determine whether this is true via trained rats: (1) Train a rat to perform measurements, (2) kill the rat before it has a chance to tell you the results, (3) check to see whether wave function has collapsed.

It doesn't work that way, because even if the rat doesn't cause the wave function to collapse, it interacts with the system and causes a transition, from the original state to a state which is a superposition of states corresponding to various values the rat might have gotten from the measurement, each of these states individually looking exactly as if the rat collapsed the wave function, with the coefficients such that the probabilities for each value come out right. And killing the rat afterwards does not undo it. So you will observe that the wave function has collapsed. Same if you use a mechanical detector in place of a rat.

Thanks muchly for that excellent summary!

Yes, I would like copy of the paper. Please send it to doug at alum dot mit dot edu. Thanks!

> And killing the rat afterwards does not undo it.

Yes, sorry; it's been a long long time since I've thought about this sort of stuff in any detail. This is what I should have written:

Train a rat to perform measurements on an observable with two possible outcomes and have it press lever A for outcome A and lever B for outcome B. Put the rat into a sealed box to perform the measurement. A dial on the outside of box will read either A or B once the rat has performed the experiment and recorded the result.

You can now come up with a complex observable on the whole system, i.e., the original observable being measured by the rat, plus the rat, and the box, that will give two different results on different occasions if the rat did not collapse the wave, but will always give you the same result if the rat did collapse the wave.

The problem with this complex observable is that for it to work, you must consider every molecule in the rat, every molecule of air it interacts with, etc., etc., etc. Miss a single molecule and the results are randomized.

Are we ever going to be up to this task? Not any time soon! But it could be child's play for the aforementioned evil alien overlords.

One complication, I can imagine, is that for this to work, you'd need to have a perfect model of the rat's biology and cognition in order for you to come up with the right observable. In the face of not yet being sure how collapse works, this might be very difficult. But then again, I'm sure that evil alien overlords are up the task.

David Albert talks about doing these sorts of experiments on p. 88 of Quantum Mechanics and Experience, and it was this that I was thinking of. Only Albert's examples don't have a trained rat, but rather other, simpler measuring equipment for which we are trying to determine whether or not it causes collapse.

As for Deutsch three experiments, it looks like Experiment 3 has two huge advantages over my trained rat system: (1) Since it's all contained inside a quantum computer, it seems a lot more feasible without the help of aliens. (2) If I understand correctly, the reversal stages means that you end up with a very simple observable, rather than the unfeasibly complex observable that you would need for my trained rat system.

As for Experiment 1 and Experiment 2, I don't understand #1 at the moment. And #2 seems to me so obvious as to go without saying. But it doesn't seem to prove anything that we didn't already know. Of course an uncollapsed wave can compute more than a collapsed wave!