Rationality in Parallel Universes

As I understand it, one of the points of Derek Parfit's Reasons and Persons is that questions of rationality and personality identity are intimately connected.

People will frequently give up immediate benefits for their present self in favour of a greater benefit for some future self, whereas they might not give up the same benefit for their present self in favour of a greater benefit for a present or future other. They treat their future selves with some special consideration which they do not extend to the present selves of others. This special consideration is considered rational, on the grounds that their future self is the same person as their present self. So, this concept of the rationality of self-interest, assumes a concept of personal identity across time.

The problem is — if I have multiple future selves, not temporally subsequent from each other, but rather temporally simultaneous with each other, how do I make sense of the concept of "giving up immediate benefits for my present self in favour of a future benefit for some future self"? What if, a certain course of action, will produce a better outcome for future self A, and a worse outcome for future self B? If I knew that one was such self was uniquely I in a way in which other such selves was not, I would be rationally justified in giving that future self special consideration above the others. But, in the absence of such knowledge, it seems that I am rationally obliged to extend to them all equal consideration.

Let us assume that MWI is true. Then, it seems, that I have an innumerable number of future selves, and that, whatever course of action I might take in the present, there will be future worlds, and hence future selves, in which that course of action has resulted in a positive outcome, and other future worlds in which that course of action has resulted in a negative outcome. So, there does not seem to be any course of action in the present, which can be said to benefit myself in the future; since, whatever I do, there are some future selves who are better, and other future selves that are worse. So, it seems, that rational behaviour is impossible.

The obvious objection is as follows: Act such that the future self whose world has the greatest probability amplitude experiences the most benefit. The problem I see here is, as I have earlier discussed, can we really claim to know the probability amplitudes of other universes? The only way we could know them, is by observation of this world, combined with an assumption that our world is an “average” world — and yet, there are no grounds to justify the assumption that our world is average. (We could say that, average worlds have the highest probability amplitudes; and yet, if we were in a non-average world, by this argument we would still think our world was average — thus, it seems equally epistemically likely that our world is average as it is not.) We should also ask, even if the concept of probability amplitudes can be justified, whether

Now, many minds might not have this problem, since many minds would hold that there are many present selves, which are exactly identical in every way, other than having as their successors differing future selves, those future selves themselves differing from each other in ways other than succession. (Which is not to say that many minds may not have other, different, problems.)

Does modal realism have this problem? Suppose there exists two universes, which are right now exactly identical, and one of which is our universe, but which begin to diverge in a thousand years time. Let us assume both universes are equally real; let us also assume that I only live for a normal human lifespan (thus I will be dead before the universes begin to differ), and let us assume there is no pre-mortal nor post-mortal existence. Therefore, for the entire duration of my life then there is an I in this world, and there is another person in that other world which is exactly like me in every possible way. Now, if we adopt Lewis’ approach to trans-world identity, which states that if two entities belong to different universes they are non-identical, even if their universe membership is the sole property by which they differ, then it follows that there actually exist two people, one of which who is me, and another one who is not me, but is exactly like me, in every possible way, and has always been exactly like me and always will be exactly like me. But, there seems to be no rational reason to claim two entities who are absolutely identical in content are nonetheless distinct, simply because each is distantly related to two distinct future events. It seems more intuitive to claim that there is one single entity, which is doubly related, once each to each of the distinct future different events. Therefore, I would say, we should reject Lewis’ approach to trans-world identity.

But, if we do so, then modal realism does fall victim to this very same issue? Any theory

I think, that for rational behaviour to be possible, the set of our actual futures must be sufficiently smaller than the set of all possible futures. It may be that there is a single actual future from the present; it may be there are several. But so long as there are only a few, it seems we do not have this problem.

Hypothetical scenario: I am homeless, and have been living on the streets for the last twenty years. An extraordinarily rich man appears, who has is famous for giving immense gifts to randomly selected homeless people. He offers me the following choice: $1 million today, or $1 billion in 2 weeks time. Which is it rational for me to accept? Almost everyone would say, all things being equal, that the $1 billion in 2 weeks is better than the $1 million today, and that I would be irrational to prefer the $1 million today. Suppose the world from today has a single unique future. Then, although because I don’t know the future, it could perchance turn out that the $1 million today is the best option (e.g. maybe (an epistemic maybe), whatever I do, the famous rich man will go mad and murder me in one weeks time — better then 1 week of living the good life on $1 million followed by death, then 1 week of continued poverty and homelessness followed by death, all the while waiting patiently to receive $1 billion which in the end will never come), I still feel justified in saying that the $100 million is the better choice.

Now, suppose it is true that every logically conceivable future is an actual future, and there is no reason to consider one future more “real” than another, and I know these facts. Then, it seems, that if I choose A, there exist futures in which C or D follows; and if choose B, there also exist universes in which C or D follows. So, it seems, that based on the universes which actually exist, I have no reason to believe that A causes C over D (or vis-à-vis), or that B causes C over D (or vis-à-vis). Therefore, it seems, preferring D over C, but being neutral in and of themselves as to A and B, that I have no reason therefore to choose one of A or B over the other. Thus, it seems, whatever I desire, that any course of action is as rational as any other, and thus, rational behaviour is impossible.

Suppose I know there are a limited number of actual futures (less than every possible, but more than only one). Would rationality be possible in these circumstances? It would depend, on what assumptions I could make about the choice function which chooses which worlds are actual and which are merely possible.

Let us suppose that Bostrom’s simulation hypothesis is true — suppose that, at a certain point in human history, technology develops to the point that human beings can create highly accurate computer simulations, and that we, rather than being “real” humans, are actually part of one of those computer simulations. Suppose we can assume that the simulators have reasonably similar goals, motivations, desires, etc., to us, and that they have created us to fulfil some desires on their part (e.g. desires for drama, for excitement, for sex, for role-playing, to explore historical what-if scenarios, etc.; the kind of reasons for which we today would create highly accurate computer simulations, if we had the power.) So, we know there are likely multiple future histories, and we don’t know which possible future histories are actual, but we do have at least some idea of the kind of factors which determine which possible future histories are actual and which are not. Our judgement of a certain course of action being rational is based on our holding certain causal theories, and our past observations support those theories. (e.g. (1) this rich man’s behaviour is well known, he has been on TV and in the newspaper, everyone talks about him, etc.; (2) a person, having established a pattern of behaviour, is likely to continue in it.) Our question is, are simulators having these motivations likely to create simulations in which these judgements are valid, or invalid? Or in other words, are simulators having these motivations likely to create simulations in which basic principles of inductive reasoning and reasoning from authority hold, or are so frequently violated as to be impossible to rationally adhere to? It seems, if the simulators are motivated by the factors aforementioned, they are likely to leave the causal structure of the world largely as we would expect, and only violate it on rare occasions, as necessary to achieve their goals; and given that these principles are only expressed as generalities, an occasional violation of a general principle is no violation at all, and thus, it seems the simulators will uphold them. Suppose, by contrast, that modal realism is true instead. Then, we have no reason to believe that the choice of universes which exist, would tend to support belief in inductive principles, since it seems every possible violation of those principles would occur in some future history, and every possible combination of those violations. So it seems, if we know there are multiple actual futures, then rationality is only possible if we can rationally assume that the selection of actual futures out of possible futures occurs in such a way as to encourage rationality.

Now, I have argued that under MWI, it is impossible to derive the probabilities of universes in the multiverse from the probabilities observed within any particular universe, since every possible particular set of probabilities will be observed by someone, and if it justified for any of them to derive multiverse-level probabilities from universe-level probabilities, then it is justified for everyone to make the very same derivation. Thus, we have no reason, based on observation of the probabilities in this universe, to believe that any particular assignment of probabilities in the multiverse is true.

Is the simulation scenario self-defeating in the same way? No. This is because, we have no reason to believe every possible particular set of probabilities will be observed by someone. Since only some possible universes exist, and the choice of which is based on certain motivations, we thus determine the multiverse-level structure. Since the universe-level structure of every universe would be driven by the same motivations, we have good reason to believe that one is a good cue to the other. In other words — if the actual universe determining principle is based on humanoid motivations, we have good reason to believe “as above so below” — i.e. intra-universe distributions are similar to inter-universe distributions. If, on the other hand, the determining principle is based on everything logically possible is, then we have no reason to believe there is any parallel between inter-universe and intra-universe distributions.

[Also relevant here is Bostrom’s general work on observer-dependence, as seen in both the simulation and doomsday arguments. I suppose, with MWI — if X observers actually exist, do we consider it equally likely for us to be any of them, or can we somehow claim that we are more likely to be some of those observers than others (e.g. more likely to be those observers having higher probability amplitudes?)? We might say, well, we observe p. But, all we can conclude from that, is 100% we are in the group of observers who observe p, and 0% that we are in the group of observers who do not. It still doesn’t justify the belief that we are an ‘average’ observer, i.e. one whose local probability measurements give an accurate measure of global probability measurements.]