= 27d567bc2d48d9f40e9ccc20ea370b15)
Yvain comments on MWI, weird quantum experiments and future-directed continuity of conscious experience - 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 (89)
I think I've seen the following argument somewhere, but I can't remember where:
Consider the following villainous arrangement: you are locked in a watertight room, 6 feet high. At 8 o'clock, a computer flips a quantum coin. If it comes up heads, the computer opens a valve, causing water to flow into the room. The water level rises at a rate of 1 foot per minute, so by 8:06 the room is completely flooded with water and you drown and die before 8:15. In either case, the room unlocks automatically by 8:15, so if the coin landed tails you may walk out and continue with your life.
At 7:59, you assign a 50-50 probability that at 8:03, you will experience being up to your waist in water. If you believe in quantum immortality, you will also be assigning a 100 percent chance that at 8:15, you will experience being completely dry and walking out of the room in perfect safety. Since every world-line in which you're in the water at 8:03 is also a world-line in which you're dead at 8:15, these probabilities seem inconsistent.
This suggests that you should consider a 50-50 chance of having your subjective experience snuffed out, since you could find yourself in the water-rising world, and there's nowhere to go from there but death. The other possibility - that you'd be shunted for some reason into a valve-doesn't-open branch even before there's any threat of death - seems to require a little too much advance planning.
But no form of death is instantaneous. Even playing quantum Russian Roulette, there's still a split second between the bullet firing and your death, which is more than enough time to shunt you into a certain world-line.
This doesn't go so well with the adorable zombie-face graphics above. It would go a little better with a model in which, at any point, one copy of you was randomly selected to be the one you're experiencing, but that model has the slight disadvantage of making no sense. it would also go well with a model in which you can just plain die.
I think Jordan's method robustly seals this problem. It is the total absence of information-theoretic death, but no "continuity of consciousness" can be squeezed through. Whatever moral qualms can be given to the destruction of frozen people, are about clear-cut consequences, not the action itself.
This gets me too. Just how much advance planning is the universe allowed? Am I alive now, rather than in the year 1000, because we are sufficiently close to developing anti-aging treatments?
No, quantum immortality claims that you won't drown even if the room gets flooded with certainty. You'll be saved by a quantum-fluctuation air bubble or something.
By the way, quantum immortality must then run a consequentialist computation to distinguish between freezing people to be left frozen and freezing people to be later revived. In other words, magic.
Explain?
Assuming QI, if I get frozen to be unfrozen later, I don't expect QI to "save" me from being frozen -- I expect to experience whatever comes after unfreezing and not a magical malfunction of the freezing machine that prevents me from getting frozen. But if I'm being frozen for eternity, it's death, and so I expect QI to save me from it by a quantum fluctuation.
References: The Hidden Complexity of Wishes, Magical Categories. The concept of "death" is too complex to be captured by any phenomenon other than the process of computation of this concept in human minds, or something derived therefrom.
Sorry, I wish I had followed this earlier.
No, death can easily be explained in a reductionist way without positing ontologically-basic subjectivity.
Death simply refers to when a self-perpetuating process (usually labeled "life") stops maintaining itself far from equilibrium with its environment via expenditure of negentropy (free energy). Note that a common term for dying (in English) is "reaching room temperature". (Yes, yes, cold-blooded life forms are always staying close to room temperature, but they stay far from equilibrium in other ways -- chemically, structurally, etc..)
Being frozen in such a way that the process that is you can be recovered is not death, at least not completely. You are still far from equilibrium with your broader environment -- note that you still have a large KL divergence, so the information contained in you has not been irreversibly deleted.
Never heard that one. Is that an American idiom? "Passing away" seems to be the standard euphemism where I'm from, but I usually just say "dying".
For reference, I've never encountered that either, and I'm an American and a student of British English.
Well, it's a dysphemism rather than a euphemism, but forms of it are used, and it doesn't appear to be unique to America. Check this Googling and its alternate suggestion and you see a New Zealand blog mentioning that some "oxygen waster" has finally "reached room temperature".
A very insightful idiom indeed!
I find this reasoning opaque. "Equilibrium with your broader environment"? Replace the head of the frozen person with a watermelon, and you'll have as much distance from "equilibrium" as for the head, but the person will be dead.
Not quite right. If you remove the head, and (as I presume you mean) let it die, its information is gone, as is the infomation about its connection with the body, and the information recovered would not be capable of fully specifying the process constituting the original person. They would be "more dead".
As I defined life as the sustenance of a process far from equilibrium, you have destroyed more of the process that is that individual.
On top of that, a frozen watermelon has a far smaller KL divergence from its environment than a human head. It is not the same distance from equilibrium -- it's closer.
You've just hidden the complexity in the choice of the system for which you define a simple metric (I doubt it's even right as you state, but assume it is). What you call the process is chosen by you to make the solution come out right (not deliberatively for that purpose, but by you anyway). Physics will be hard-pressed to even say what is the same rigid object over time (unless you trivially define that so in your formalism -- but then it'll be math), not to speak of the "process" of living person (where you can't define in math what that delineates -- the concept is too big for a mere human to see).
Get the print of a person in digital form and transmit it to the outer space by radio -- will the person's process involve the whole light cone now? How is that different from just exerting gravitational field?
I have not hidden any complexity nor made any arbitrary choice. The process that is the human body is mostly understood, in terms of what it does to maintain homeostasis (regulation of properties against environmental perturbations). Individual instances of a human body -- different people -- carry differences among each other -- what memories they have, what funcitonality their organs have, and so on.
Way up at the level of interpersonal relationships, we can recognize an individual, like "Bob", and his personality traits, etc. We can recognize when a re-instantiation of a person still acts like Bob. This is not an arbitrary choice -- it's based on a previous, non-arbitrary identification of a chunk of conceptspace called "the person Bob".
So we can know when Bob has irreversibly mixed with his environment.
The person will be in the same dormant state as when they are frozen, or as a seed is before it is planted, or the chemicals that mix to make a virus before they are mixed. The information to reconstitute the being is still there, but it is not yet restored to its self-sustaining, entropy-exporting process. When you transmit their information through space, you are giving structure to the EM waves propagating against background noise, so there's still a KL divergence from the environment: the waves you transmit are different from what you would expect if you expected normal background noise.
You still, of course, need someone capable of decoding that and reinstantiating the person. When all information about how to do so is lost, then the person is finally irreversibly mixed with their environment and permanently dead, in line with the definition I gave before.
I'm not sure of the purpose of this question. Could you state clearly what your position is, and which part you believe I'm disagreeing with, and why that disagreement is in error?
Ok, I see the point you are making. But When you say
You are thinking of QI as an agent who has to decide what to do at a given time. But suppose a proponent of QI thinks instead of QI as simply the brute fact that there are certain paths through the tree structure of MWI QM that continue your conscious experience forever, and the substantive fact that what I actually experience will be randomly chosen from that set of paths.
I disagree with QI because I think that the very language being used to frame the problem is severely defective; the semantics of the word "I" is the problem.
I think that perhaps the word "I" suffers from the same problem.
As a concept -- whether it's defined in the language of games is irrelevant.
Why wouldn't you expect to be "saved"? MWI simply means that anything that can happen - will happen (in some branch). So you'll be "saved" in both cases in some branches (if this is physically possible given the current situation).
So, at 7:59, what probability do you assign to experiencing, at 8:16, a memory of having been saved by the coin landing tails, versus a memory of having been saved by quantum fluctuation?
QI doesn't specify. A reasonable assumption would simply be to condition upon your survival, so at 7:59 you assign, say, a 1-10^-6 probability to the coin landing tails, and a 10^-6 probability to other ways you could be saved, for example rescue by an Idiran assault force, a quantum bubble, etc.
Well, the reason I ask is that if you're standing outside the multiverse at 8:16, and you count the number of universes with living mes that were saved by the coin landing tails, and those with living mes that were saved by quantum fluctuation, the ones with tails outnumber the ones with fluctuations several gazillion to one, since from an outsider's point of view there's a 1/2 chance I'll get saved by tails versus a one in a gazillion chance I'll be saved by fluctuations.
But from my perspective, if and only if we enforce continuity of experience, there's a 50-50 chance I'll find myself saved by tails vs. fluctuations. But this creates the odd situation of there being certain "mes" in the multiverse whom I am much more likely to end out as than others.
why? why not just condition your existing probability distributions on continued conscious experience?
Right, QI says that you will not die.
I agree. I also think this is why Christian stated in his problem setup for Quantum Russian Roulette that the participants are put into a deep sleep before they are potentially killed. If the method of death is quick enough, or if you aren't conscious when it occurs, then you shouldn't be shunted into any alternative world-lines.
No method is quick enough. At any time t some event can prevent someones subjective experience in some very large set of world-lines. But what you can't do is spin a quantum wheel at time t and then kill everyone else at t+1. From the subjective experience of the players either the causal connection between the roulette wheel and the killing mechanism would fail or the killing mechanism would fail. If you got lucky the failure you'd experience would happen early- but chances are you'd experience everything right up until the last possible plank-length of time (or wake up afterward).
Is there a method of killing which, according to quantum probability, either kills someone outright or leaves them relatively undamaged?
I think the real question is whether the chance of being saved damaged is significantly higher than just being damaged without playing the game. For example if you get into a car, you have a relatively high probability to get out damaged. If you get below that threshold, then you don't take any extra risk
Lets say Smith is standing with a gun aimed at his head. The gun is aimed at his head is attached to a quantum coin with a 50% chance of flipping heads. If if flips heads the gun will go off. All this will happen at 8:00.
If we examine the universal wave function at 8:05 we'll find that in about 50% of worlds Smith will be dead(1). Similarly in about 50% of the worlds the gun will have gone off. But those sets of worlds won't overlap. There will be a few worlds where something else kills Smith and a few worlds where the gun doesn't. And if you look at tall the worlds in which Smith has conscious experience at 8:05 the vast majority would be worlds in which Smith is fine. But I don't think that those proportions accurately reflect the probability that Smith will experience being shot and brain damaged because once the gun is fired the worlds in which Smith has been shot and barely survived become the only worlds in which he is conscious. For the purposes of predicting future experience we don't want to be calculating over the entire set of possible worlds. Rather, distribution of outcomes in the worlds in which Smith survives should take on the probability space of of their sibling worlds in which Smith dies. The result is that Smith experiencing brain damage should be assigned almost a 50% probability. This is because once the gun is fired there is a nearly 100% chance that Smith will experience injury since all the worlds in which he doesn't are thrown out of the calculation of his future experiences. (2).
This means that unless the killing method is quantum binary (i.e. you either are fine, or you die) the players of quantum Russian roulette would actually most likely wake up short $50,000 and in serious pain (depending on the method). Even if the method is binary you will probably wake up down $50,000.
(1) I understand that quantum probabilities don't work out to just be the fraction of worlds... if the actual equation changes the thought experiment tell me, but I don't think it should.
(2) My confidence is admittedly low regarding all this.
I don't agree with your calculation.
The probability of winning the money over suffering injuries due to failed execution attempt is P=(1/16)/(1/16+15/16*epsilon) where epsilon is the chance that the excution attempt fails. If epsilon is small, it will get arbitrarily close to 1.
You should not be worried as long as P<Baseline where Baseline is the probability of you having a serious injury due to normal every day risks within let's say an hour.
That is the probability that an observer in any given world would observe someone (Contestant A) win and some other contestant (B) survive. But all of these outcomes are meaningless when calculating the subjective probability of experiencing an injury. If you don't win the only experience you can possibly have is that of being injured.
According to you calculations if a bullet is fired at my head there is only a small chance that I will experience being injured. And you calculations certainly do correctly predict that there is a small chance another observer will observe me being injured. But the entire conceit of quantum immortality is that since I can only experience the worlds in which I am not dead I am assured of living forever since there will always be a world in which I have not died. In other words, for the purposes of predicting future experiences the worlds in which I am not around are ignored. This means if there is a bullet flying toward my head the likelihood is that I will experience being alive and injured is very very high.
How is you calculation consistent with the fact that the probability for survival is always 1?
You experience all those world where you
You have the same in everyday life. You experience all those worlds where you
There is no difference. As long as the ratio of the above two probabilities is bigger than the ratio of the ones below, you don't go into any extra risk of being injured compared to normal every day life.
You agree that the probability of survival is 1 right? My estimation for the probability of experiencing injury given losing and surviving is very high (1-epsilon). There is a small chance the killing mechanism would not do any harm at all but most likely it would cause damage.
It follows from these two things that the probability of experiencing injury given losing the game is equally high (its the same set of worlds since one always experiences survival). The probability of experiencing injury is therefore approximately 1-epsilon (15/16). Actually, since there is also a very tiny possibility of injury for the winner the actual chances of injury are a bit higher. The diminishing possibilities of injury given winning and no injury given losing basically cancel each other out leaving the probability of experiencing injury at 15/16.
Wrong: it is epsilon 15/16.
You are confusing conditional probability with prior probability.
Quick death is fine. I just wanted to put up a realistic scenario which is very gentle and minimally scary.
It is more plausible to be able to struck death in a deep sleep without your noticing. If I say, you get struck by a lighting, then your first reaction would have been: "OUCH!".
But if you get sedated by some strong drug then it's sounds much more plausible to be a painless "experience".