By Paul Almond, 6 April 2006

Introduction

This is an article by me on a "day off" from my usual writing. Of all my articles so far, this is the one that will most appeal to people who have found my writing conservative and want me to get in touch with all kinds of things that people went on about a lot in the 1960s. It will annoy some people, mainly those who found my earlier articles far too speculative. We will go really far out here, much further than I normally go, and hopefully have fun playing with some rather extreme ideas and seeing where it takes us.

In 1987 Hans Moravec proposed the concept of quantum suicide [1]. It was also proposed by Bruno Marchal, the idea being contributed to further by Max Tegmark in 1988. The idea uses Hugh Everett's many-worlds interpretation of quantum mechanics [2,3] and imagines a physicist performing an experiment involving a series of quantum decisions - each having a 50% chance of triggering some device that kills him/her. If the many-worlds interpretation is true then, the thought experiment suggests, there should always be some worlds in which no quantum decision has caused the device to be triggered and the scientist is still alive. The thought experiment suggests that, as these are the only worlds in which the scientist is around to make observations after the experiment, by performing it the scientist will have proof, for him/her at least, that the many-worlds interpretation is correct.

An extension of this idea is quantum immortality [4]. This is the idea that, whenever a person's consciousness could end, there should always be possible sequences of quantum events that allow consciousness to continue and, no matter how improbable, these will always occur in some worlds. The person will continue to exist in these worlds and, when determining what a person will observe, it is only those worlds in which he/she continues to exist that are relevant.

Critics of quantum immortality suggest that it would fail because the many-worlds hypothesis does not say that everything has to happen and that it may be that there are situations in which no sequence of quantum decisions allows survival. I think those making this criticism tend to be proposing that we all will end up in this sort of inescapable situation eventually, and sooner rather than later, probably about the time that we would actually expect to be dying.

Both the quantum suicide experiment and the idea of quantum immortality are controversial.

A further issue relating to continuation of consciousness is that of mind uploading - making a digital description of a brain and then using it to run a computer simulation with the intention of providing continued conscious experience for the person whose mind is uploaded.

The purpose of this article is to propose and discuss a thought experiment relating to the many-worlds interpretation which uses the idea of mind uploading [5,6]. I will not be taking a position on quantum suicide, quantum immortality or the scenario proposed in this article: the intention of the article is simply to provide a scenario facilitating focusing on some of the issues surrounding these subjects.

Background

The thought experiment is based on an imaginary scenario as follows:

A company offers a mind uploading service. People can have a "scan" - a very detailed map - made of their brain. The data making up this map is known as a "scan file", a term used in Greg Egan's novel Permutation City [7]. The scan file is a sequence of "1"s and "0"s describing the brain.

The scan file is fed into a computer which uses it to simulate the continued thinking of the brain. The idea is this allows a person's personality to continue to exist after the destruction of the biological brain. Although controversial, this idea of mind uploading is well-established philosophically, if not technologically. What sort of existence is the simulated brain given? It could be allowed to interact with a virtual reality environment. It could be inserted into a robot and allowed to continue to exist in the real world - provided fast enough computers exist to make it worthwhile. If sufficiently advanced fabrication technology exists, maybe it could even be used as a template to build a replacement biological brain. What happens to the scan file is not very important in this thought experiment, provided that it is used to try to give the owner of the brain some continuation of life. For simplicity, we will assume that the scan file is used to make a model of the brain that "runs" in a virtual reality environment on a simulation computer.

Scenario

A technician is preparing a scan file to be "run". The owner of the biological brain on which the scan file is based is now deceased and his estate has requested that the scan file, taken recently, is now used to continue the personality of its owner from the point at which it was made.

A disaster happens. The technician accidentally activates a command that wipes the scan file. In a panic he tries to recover the data but it is all gone. He checks for any backups but none were made. It appears that all is lost.

The technician has an idea, however. What if he were to try to guess the scan file? The scan file is just "1"s and "0"s. In principle he could type out a random sequence of "1"s and "0"s, save it into a file and input it into the computer that simulates the brain as the scan file. He starts typing frantically, randomly hitting the "0" and "1" keys.

After a few seconds he realises the futility of his actions. Scan files are huge. An enormous number of "1"s and "0"s are needed to describe the state of a brain. It is not practical to type all this.

The technician quickly writes a computer program to generate a huge, random sequence of "1"s and "0"s of about the same length of the scan file. He starts it running.

After the program has been running for a while he realises that success is implausible. His program might reproduce the original scan file by chance, but the probability of this happening is absurdly low.

At this point he has another idea. He recalls reading about a quantum mechanical theory by Everett known as the many-worlds interpretation. According to the many world interpretation, each time a quantum decision is made the world "splits". Hoping that the many-worlds idea is correct he decides to try to use it to save the situation. This is what he does:

He writes a computer program which takes input from a physical system. The physical system, known as a quantum event generator, generates "1"s and "0"s randomly as a result of quantum events. The program will use the physical system to tell it what sequence of "1"s and "0"s will be used to try to recreate the lost scan file. The program starts with an empty scan file which will be filled with "1"s and "0"s and this is how it works:

1. The program takes a random input from the quantum event generator. If the input is "0" then the program appends a "0" to the scan file. If the input is "1" then the program appends a "1" to the scan file.
2. The program takes a further, random input from the quantum event generator. If the input is "0" then the program halts. If the input is "1" then the program jumps back to Step 1 to get the next binary digit in the scan file.

The above process continues until the program is halted by the quantum event generator at the appropriate time or the length of the scan file reaches some very large preset size beyond which there is no point in continuing.

The technician sets the computers up so that when the scan file is completed - after the program has terminated - it is automatically provided as input to the computer that will run the brain simulation in the virtual reality.

The technician reasons that, if the many-worlds interpretation is correct, this will guarantee continuation of his customer's consciousness.

Rationale

Why does the technician think that this will work? Without using a many-worlds interpretation point of view the outlook is bleak. Each time the program appends a "0" or "1" to the scan file it is effectively tossing a coin. The program could generate the exact scan file that was lost, but it is much more likely that the program will output one or two bits, or maybe a few more bits of random nonsense, before terminating. Even if the program generated a file with the same size as the original scan file it is absurdly unlikely that it would match it.

What, however, if the many-worlds interpretation is true?

In this case, the technician reasons, every time a "1" or "0" is generated by the quantum event generator, this does not correspond to just one of two alternatives happening: it corresponds to one of two outcomes being selected as the world splits. For each instance when the technician sees the computer writing a "0" to the scan file he will know that there are other worlds in which it wrote a "1". This means that every possible sequence of "1"s and "0"s that could be received by the computer is realised in different worlds, further meaning that every possible scan file is brought into existence in some worlds - at least subject to the condition that the length of the scan file does not exceed the maximum preset size. The whole purpose of making the scan file was to make sure that a computer exists which runs it - and, "somewhere" in the multiverse, reasons the technician, a computer will happen to be running a simulation based on the correct scan file.

Further Issues

If the technician's actions are successful then other issues become relevant:

• There is nothing about the program run that singles out his/her customer. His actions actually generate and "run" scan files taken at every moment of life for everybody who is currently living or who has ever lived, regardless of whether or not they actually had a real scan file made, provided that the upper limit on scan file size allows any human brain to be described.
• The program run also generates scan files for everyone who is going to live - subject to the limitation about scan file length. This raises the issue of whether a scan file made and run in the past of the brain that it is describing could be considered to be "successful".
• Lots of non-existent people would also be given free scan files and could then "resume" their lives from the point at which their scan files describe their brains, untroubled by the previous encumbrance of non-existence. We should consider, however, how non-existent these people are…
• We can expect some space aliens to get mind-uploaded as well.

A Consideration of the Thought Experiment

If this worked for the customer in the scenario it would be profound with regard to mind uploading and continuation of consciousness as it would suggest that a person can be uploaded accurately without any advanced brain scanning technology. This article is certainly not proposing this as something to do: it would be reckless to do that now in such a "far out" area. This is purely a thought experiment and something on which we can hang a discussion.

Scope for this to Happen

One of the criticisms of the quantum immortality idea is that it may be naïve to assume that there will always be some sequence of quantum events available to escape death and that certain situations may simply lack scope for the right quantum events to occur to continue consciousness. In this thought experiment we try to circumvent that criticism by actually trying to create adequate scope for the sequence of quantum decisions that is supposed to continue consciousness.

An objection could be that the software descriptions of the brain produced in the "successful" worlds do not continue the customer's consciousness because they are simply digital approximations rather than completely accurate representations of that consciousness. This kind of objection would also rule out more "conventional" mind uploading. It can be asked, however, what criteria a "completely accurate" description would need to satisfy and whether or not the original brain actually satisfies such criteria from moment to moment. In any instant the brain's next state is derived from its current state and the uncertainty principle would put limits on the brain's own capability to propagate its own description "with total accuracy" from moment to moment - something that people often seem to overlook. Some kind of causal link could be viewed as necessary for some other reason.

If an uploaded description is accepted as being valid then another interesting possibility is raised:

What if the technician decided that his program run would be successful, but chose not to bother running it? His reason for this could be that surely someone else will run a similar sort of program at some time in the future, and his customer could benefit from a program run by someone who does not even know about his customer.

The technician may think, however, that this is too dangerous. What if everyone else during the remainder of human history (whatever it is) comes to the same decision and - as in a situation in which nobody phones to report a fire because they think someone else has - nobody ever runs it? Well, maybe his customer's continuation could benefit from a program run by aliens?

What if nobody runs it in the future - maybe the world ends next week before anyone gets a chance - and no space aliens exist? Maybe all is not lost! What about all the other worlds that do not share the same past as ours? All that is needed is for someone in one of those worlds to run the same sort of program at any time in the future, or even, maybe, to have run it in the past.

What if nobody runs this sort of program? Maybe something which has the same sort of effect would occur as a result of intelligent entities doing something else? It may be that there are valid objections to this - that any such program run has to be in one of the futures of "our" world to be valid from the point of view of preserving consciousness, but what if we accepted this as valid?

If we accepted the idea that this would work as a result of an intelligent entity doing the right things then another possibility should also be considered: that a process equivalent to running of this sort of program may happen in other worlds naturally. Thinking that the program run would work may be equivalent to thinking that it is unnecessary. The process which would need to happen in another world would not necessarily need to duplicate the program run as described in the thought experiment. All it has to do is produce a physical system equivalent to a simulation of the customer's brain - and if we accept that mind uploading works it does not need to be exact - only as accurate as the kind of digital copy that we would accept. Would this work? I am not going to make any claim about the probability of this happening, but if this did work it would be providing extension of consciousness on a different basis than what is generally used in the idea of quantum immortality. In the quantum immortality idea, extension of consciousness is automatically achieved by some "lucky" sequence of coincidences in one of the futures of your own world. The criticism that such a fortunate sequence of coincidences may not be available loses most of its power with the sort of thing that we are considering here, because this type of consciousness extension would not rely on some lucky coincidences that save you in one of your own futures. It would simply use the occurrence of a pattern equivalent to your mind in your future, or even in worlds that do not share your past. There is no requirement for the events to be causally related to you anymore. Does this mean that it automatically works? No, there is plenty of room for debate on this.

Causal Connection with the Original Person

If we are allowed to play this trick of finding a pattern equivalent to you somewhere else, or bringing one about by quantum affected program runs, rejecting the need for causal connection with you may give more scope for extending your consciousness, but it is also a potential philosophical cost. Do we get away with removing causal connection with the original person?

The thought experiment is important to all this, even if we decide that its effects would occur "elsewhere" without us having to do it, because it exhibits the main characteristic of this type of quantum immortality or consciousness extension: the removal of any causal link between the original person and the physical system that is expected to continue his/her consciousness. For this to work it is required that the idea of mind uploading would work, requiring that a description of a person that is not 100% accurate could still be a viable continuation of that person. These issues are not a concern in the "conventional" idea of quantum immortality.

Any preservation of consciousness in the thought experiment relies, of course, on some kind of "many-worlds" description of reality as being correct and the many-worlds interpretation is controversial. The many-worlds view needed here does not have to be the same many-worlds model proposed by Everett. Various "ensemble" [8] views of nature have been suggested and some of these could be just as good as the many-worlds hypothesis for this sort of purpose. In fact, some could be better. One way of justifying Occam's razor is with an ensemble view of reality in which every world that can be algorithmically described exists. This would effectively ensure that other worlds produced the necessary configurations of matter to be equivalent to a mind uploading process.

Measure Across Worlds as a Problem

One objection to any claim of consciousness continuation in the thought experiment could be based on the idea of measure across worlds. If the many-worlds view is correct then at any instant you have a very large number of possible futures: your future has a large measure. When you are reliant on the sort of process discussed here for continuation, whether it is done deliberately or otherwise, you clearly end in almost all of these futures, continuing only in those in which the program run (or some other process) happens to bring about a suitable pattern. This objection could be countered by saying that the futures in which you do not continue do not matter, in just the same way that people do not worry unduly about not being everywhere in a non-many-worlds-interpretation of the world. I am not in China now, and I am not in many other places. I do not view my failure to go to China yesterday as suicide in terms of "ending" possible futures: I hardly mourn for "dead versions of myself" in all these places, but only regard the place in which I exist as being important to my own identity.

Strong Continuity of Self and Value

One position could be that the thought experiment is best considered from the point of view of value. Thinking about this in terms of whether your mind continues or not, and thinking that there is a hard answer to this question, implies acceptance of some idea of strong continuity of self. I discussed this sort of idea in a previous article [9], arguing that decisions about our survival are really decisions about what value we would give to future configurations of matter, or what allegiance we would have to them. This approach does not necessarily provide the answer to all these questions. The decision to base that philosophy on assigning value is little more than an admission that there is some arbitrary nature to this and that the answer is not hard-wired into physics or mathematics, but into our own psychology. We would still have to come up with a way of assigning value that answers this thought experiment.

It may seem that the two sorts of question "Does your mind exist next and if so where does it go?" and "What value do you place on various configurations of matter?" are two separate questions, but this is not necessarily the case. In everyday life we do think of things in terms of where our mind will be next and whether or not it will continue to exist. If we choose to work by assigning value it does not mean that we reject this - it would be very hard to do so in everyday life at least - but it could mean that talking about where our mind will be in the future is a way of describing how we assign value. Taking a rather silly example I think that in five minutes my consciousness will be involved in some way with a brain vaguely like the one that I am using now. I do not think that my consciousness will be involved with my mouse mat or my stapler. If my outlook is really based on the value that future configurations of matter hold for me then I do not need to regard such statements as wrong: I can simply regard them as a casual way of stating what I value. I regard my brain as important for satisfying my survival inclination. I do not regard my stapler or mouse mat as having similar value from a survival point of view.

In the course of writing this article, I should say that my confidence in my earlier assertions about the incoherence of strong continuity of self has been reduced somewhat and I am reconsidering my views currently. This is an issue to which I will be returning later in the article.

Dealing with Multiple Copies

One feature of the sort of logic in this thought experiment is the possible generation of multiple copies to continue a person. The technician could run his quantum-input program to continue his customer, but if the process were viable similar program runs or equivalent processes could occur "elsewhere" to cause multiple continuations of the customer from the same moment. How would we say which is/are valid?

We could declare that copies have to be in "our own future" - which would need some justification - and even then this would not deal with the issue of multiple copies being spawned in the customer's own future.

If we accept copies as being viable ways of continuing consciousness then, unless there is an obvious way of differentiating between them, in terms of "reality" or "value", we would have to presume that there is no way of choosing between them: any copy is as valid as any other copy made with the same accuracy. This suggests that if multiple copies were to exist, all modelling a person's brain from some moment in time, we should regard this as a splitting of the person's future in the same way that the probability of things happening to a person is generally treated when people consider the many-worlds interpretation. This means that if two copies were made of a brain - Copy A and Copy B - and the original brain were to cease exist immediately on the making of the copies, then it may be reasonable to say that, from the point of view of the owner of the brain before the copies were made, there is a 50% chance that he/she will find him/herself in the situation of Copy A and a 50% chance that he/she will find him/herself in the situation of Copy B. It should be noted, however, that this is based on the assumption that there is no way of differentiating between copies. There may well be!

This "probability of where you will find yourself" approach to splitting of your mind may appear to be a mystical kind of belief - almost implying that there is some kind of "soul" that moves through all this splitting of minds and decides which mind to enter, but that is not what I am saying at all. Leaving mind uploading aside, taking this sort of view is necessary if we want to reconcile the way that our futures would split in the many-worlds interpretation with our intuitive idea of probability.

Weak Continuity and Probability

What if, however, we have abandoned any idea of strong continuity of self and are thinking in terms of value? If our way of valuing things makes copies as valuable as the original, then we would still need to assign them equal value - that is to say, give them equal allegiance - unless we have reason to think otherwise. This does not mean that we should think that all situations in which a copy is in are equally valuable to us, but that we would simply care about the fate of each of them to the same degree, and if we abandon strong continuity that is all that we really have. If we took such a position then our actions would be the same as if we took a probability view of splitting and we could use the language of probability to discuss it. In such a situation, whether or not we are talking about real probabilities or probabilities assumed for the purposes of determining advantageous actions to serve our allegiance to future versions of ourselves is an ontological issue that need not even concern us in this discussion, although it could be a meaningful thing to discuss at some other time.

This does not mean that abandonment of strong continuity automatically leads to a "splitting as probability" view. Such a position would rest on the assumption that copies - like those made in mind uploading or similar sorts of processes - are valuable in the sense that they are an arrangement of matter to which we should have allegiance.

This would not resolve the issue of what constitutes a valid copy. How accurate would it need to be for us to have allegiance to it? Any acceptance of mind uploading means that we must accept some inaccuracy in the copy - so that there will be a set of viable copies. If limitations in scanning technology did not ensure that then the uncertainty principle and the mere occurrence of digitization would. It may also be that the way we value things gives different value to copies based on how accurate they are or whether or not they are "corrupted" in some way - an issue that we will not get into deeply in this article.

If, however, we tend to value copies as having equivalent value with brains then we would probably be in a situation where we may as well use the language of probability to talk about possible futures, while also accepting debate about the ontology of what we are saying.

It should also been noted that if the many-worlds interpretation is correct then we are already using the language of probability to deal with splitting into different futures every time we use the idea of probability to discuss conventional events. If we adopted a value system that was inconsistent with the language of probability then, in a many-worlds cosmology, we would have to abandon our ideas of probability in issues relating to our own welfare - something we hardly seem likely to do.

The Conventional Continuity "Paradox"

Let us suppose that we consider the issue of splitting of futures - whether by many-worlds, multiple mind uploading copies, or both - to be describable in terms of "the probability of where you will be" - whether because we think that is the case, because we think it is a convenient way of describing how value is assigned, or because we think that it is ontologically irrelevant to differentiate between statements of probability and statements assigning value. If the many-worlds interpretation is true then, with regard to how we behave now, this is little more than saying that we continue to use the language of probability when discussing our futures. Let us also imagine that some of the things we have discussed in this article make sense and that there are many processes that could spawn copies of you - whether in your future or in other worlds that do not share your history and whether by using processes like the one discussed in this article's thought experiment or not. This suggests a possible paradox:

In a multiverse or some ensemble where quantum immortality works in the way discussed in this article then it would be because there would be all kinds of configurations of matter equivalent to your brain state at the time of death coming into existence and proceeding to be you from then onwards, in the past and future of all kinds of worlds, and this would lead to a splitting of your consciousness which could be viewed in a probabilistic way. There is nothing statistically special, however, about the moment of your death. If there is enough scope for generation of copies from that moment then there should be just as much scope for generation of copies describing your brain state - and continuing it - from any moment in your life.

If this is the case, then for any moment in our lives there could be many other versions of us branching off - not just due to the splitting of worlds in the many-worlds-interpretation, but merely because appropriate patterns for "copying" us from that moment onwards arise in other worlds. Some of these would be in worlds drastically different to ours. This is the problem: if the idea that copies are valid continuations of a person's future is accepted then each of these copies could be as valid a continuation of your future as your "conventional" future. At any moment the futures that are causally linked to your present and "split" off from it in many-worlds are just some of the possible ones. Some of your futures may be continuations from locally improbable configurations of matter in worlds that have no causal relationship to your present. If we are seeing things in probabilistic terms then each of these is equally likely to be your future.

If this is the case, could this not make a mockery of human actions? For example, why bother fastening a safety belt in a car? You can only assume that you will continue to be in a car if you think your future has to be one of those that is causally linked to your present. What if you fasten your safety belt and then find that your "next state" is as an uploaded copy, in a completely different situation, in the thought experiment program run that we discussed earlier, or your next state is in some totally incoherent situation in a totally different world, just because that world had some matter that hit on the right pattern? If this were the case we would not even have time even to react to the world. Our lives would be like a series of Quantum Leap, except written by overly enthusiastic writers who want a "leap" from one instant to the next. Observing the world would be an impossibility and even thinking in these terms - where you become some sort of detached observer in a meaningless blur of a world - may be assuming too much coherence.

We do not, however, observe this, hence my description of it as a "paradox". Is it really a paradox? Of course not: paradoxes cannot exist in nature and there must be an answer to it.

Any answer would probably need to be generally based on the idea that "conventional" futures have more measure than weird futures. This would mean that the number of patterns in the multiverse describing you in situations where things could be expected to proceed reasonably normally, in ways that are predictable - at least to some extent and in the short term - would be much greater than the number of situations where the pattern describing you manifests itself in weird situations. The sorts of brains we have, with personalities shaped by events and memories shaped by history, are easily explained by the sort of world we live in and there should be a large number of ways of getting brains into this sort of state which involve them having experiences similar to the ones that we have been having here. Worlds which split off from ours very recently provide obvious places for large numbers of versions of us with very similar histories. On the other hand, if we imagine a computer program which models your brain for no good reason than some binary digits randomly being thrown together to do it, or a version of you which has your personality and memories despite having lived on a planet full of penguins for no good reason than "it just randomly happened" there seems to be less scope for doing this. It seems then that the sorts of worlds in which your mind is like it is now because it has the sorts of experiences that you have had should be much more common than worlds in which your mind is like it is now for no good reason.

One strange aspect of this is that it seems to suggest that the reliability of our future expectation may not just depend on splitting of our current world but on the statistics of worlds "near" it which already split away and which would be practically the same, enabling them to provide a large supply of versions of us which are almost identical, making the existence of "out of context" versions of ourselves in "weird" worlds statistically irrelevant - allowing us in turn to ignore the "nearby worlds" that split recently and just obtain our future expectations from considering possible future splitting of our own world. It should be noted that this does not close discussion on the possibility of finding out that our situation is different than what it appears to be, nor does it rule out the possibility of something, possibly intelligence, upsetting the statistics.

More Expansive Ensembles and Histories as Patterns

We could consider this "conventional continuity" issue further by expanding our view to encompass ensemble descriptions that are more expansive than the many-worlds interpretation. Some ensemble description of reality may not incorporate time as a basic part of the "framework", as the many-worlds description does, and in such descriptions of reality, algorithms may extract various patterns. The "measure" of an object or situation could be considered to be related to the number of algorithms that extract a pattern representing it. In such an ensemble view it may be that the many-worlds interpretation is simply a consequence of this as suggested by Standish [8].

If time is not fundamental then any theory that contains time would only be - including its description of the events in time - part of a pattern extractable by an algorithm. This would imply that the entire history of a person within a physical system would be something that is extractable by an algorithm. This is a different situation to the one we have been considering. Instead of considering which possible future situations have the greatest measure, and then looking at which situations are the most likely for the observer to be in "next", we would be now looking at which kinds of possible histories - each of which is a pattern describing a sequence of events that contain the observer and are consistent with the observer's experiences - have the greatest measure.

This could have implications for ideas such as mind uploading. One thing that is likely to reduce the measure of a history is complexity. Complex histories are likely to need more information to express them, and therefore to have a smaller measure (to be fewer in number) than simpler histories.

One view considered in this article is that we can view any version of the observer as being equally valid as any other, but it may be that histories for an observer which contain discontinuities as strange things happen to the observer are more complex to express and have a smaller measure than more conventional, continuous histories.

As an example, let us consider a man over an interval of 10 years.

After the first five years the man makes a digital copy of his brain which is then used to launch a simulation. He does not die, so we now have two versions of him - the original brain and the uploaded version. Suppose that just before the uploading process he asks where he will find himself next. Whether the question is being asked on the basis of strong continuity of mind or on the basis of value with weak continuity need not concern him, provided that, if he has abandoned strong continuity, he is happy that the uploaded version is valid as something to which he can have allegiance as a survival motivation in his value system.

The sort of logic we have been using for most of this article would take a simple view of the situation and say that after the uploading there are two versions of him - the biological one and the uploaded one - and that each outcome represents a splitting of his future, just as in the many-worlds interpretation. As there are two outcomes and we do not have any prejudice in favour of the biological one then each is equally "likely". Therefore, the man should treat the process as if he has a 50% chance of finding himself in his current biological brain after it and a 50% chance of finding himself to have become an uploaded copy of himself. Whether or not these are "real" probabilities, or whether such a question is even ontologically meaningful, does not matter.

Is this correct? If we use the sort of ensemble approach which may have the many-worlds interpretation embedded in it as a kind of special case it may be too simplistic!

Let us imagine now that the above events occur in some sort of ensemble in which space and time are just parts of an extracted pattern describing reality as we perceive it. The history of the man over the ten years is considered as an object - a pattern - and we can consider it as being extracted by an algorithm. Two obvious sorts of patterns are available for extraction. The first is a pattern describing the mind of a man who lives for ten years using a biological brain. The second is a pattern describing the mind of a man who lives for five years using a biological brain and then abruptly turns into the uploaded person running on the computer. There could be many other possibilities that we do not know anything about, but we will ignore these and just assume that we have these two.

If we consider the situation from the point of view of the man just before the mind uploading process occurs then he is in the first five "biological" years and his future involves continuing to be biological or becoming the uploaded copy. As before, we can consider his future to be about to split into two different futures, one as a conventional, biological person and the other as an uploaded copy. If the man is in the pattern that describes him living all of the ten years biologically then after the upload things will go on for him as they did before, in his biological brain. If he happens to be in the pattern that describes him living the first five years biologically and the next five as an uploaded copy then after the upload has been performed he will abruptly find himself uploaded.

There is no real splitting in this view. The man is in one of two sorts of pattern, right from the start, which diverge from each other, but he does not know in which pattern he is embedded because, for the first five years, until the mind uploading occurs, the experience of being embedded in each pattern is identical. After the mind uploading process he will find out which sort of pattern he was in.

Before the uploading process, can the man estimate the probability that he is "in" one type of pattern or the other? He could say that there is a 50% chance that he is in each pattern, but that would be assuming that the measure of each type of pattern is the same. For each type of pattern there will be many different ways in which it could be extracted from reality and the number of such ways is the measure of the pattern. The probability that he is embedded in one pattern or the other is dependent on the relative measure of the two patterns and this means that the probability that he finds himself in one situation or other after the mind uploading process depends on the relative measure of the corresponding pattern for all of the ten years that we are considering.

There is no guarantee that the probability would end up as 50% in each case. In fact, it may very well not be anything like a 50/50 situation. If the number of ways of extracting the pattern of an uploaded person were very much less than the number of ways of extracting the pattern for a conventional brain then this would in turn have an effect on the relative number of ways of describing the entire ten year span for the man who finds himself uploaded. It could, of course, work the other way. This raises what is, to me, a rather intriguing possibility: that if a copy is made of a person and used to launch a simulation then, according to this view, the probability that the person will find himself to be the copy will depend on how the copy is physically put together, as this will influence the measure of that outcome.

This also suggests that having multiple versions of a copy, even if they are identical and in identical situations, is not trivial as it will tend to increase measure. In fact, a situation in which we have multiple identical copies in multiple identical computers could really be considered a special case of "how a copy is put together".

Something else that could affect measure is the incorporation of multiple substrates. With the "conventional" biological existence for the whole ten years any pattern needs to be described by an algorithm long enough to deal with biological brains in general and the specific details of what is happening with this particular brain. With the existence that involves five years as a biological brain and the next five years uploaded any pattern needs to be described by an algorithm long enough to deal with biological brains in general and computers in general as well as what is happening with this person. Situations needing longer algorithms to describe them will tend to have lower measure in ensemble views. This means that, all else being equal, having multiple substrates (such as brains or computers) in someone's existence should tend to reduce measure, and therefore the probability that anyone about to have a copy made that relies on a different substrate will later find himself in the position of being that copy. This does not mean that computers are automatically "poor" substrates for mind uploading, merely that the change of substrate seems likely to incur some "cost" in terms of measure and probability. This "probability" cost could be made up for in other ways.

Measure would also be affected by discontinuity. Any algorithm describing a history involving a transfer from one situation to another would need to specify the time and, potentially, other specifics of that transfer - again increasing the length of any describing algorithm and decreasing measure.

None of this should be taken as an argument that strong artificial intelligence is invalid or that mind uploading would not work.

Even if the measure for someone finding himself in a particular situation is low, this does not mean that it must always be unlikely. In the above situation, we have been considering what would happen if a copy were made of someone's brain and the original brain continued to live. In this situation, the history of a person who continues to be biological is effectively competing with the histories in which the person makes the transition for relative measure and probability. It may be that the chance that the person finds himself uploaded into a different substrate is small for the reasons given above, although this may not be the case: the substrate may be one that has high enough measure, to make a history involving transfer to it a high measure history even when the costs to measure of change of substrate and discontinuity are allowed for.

What, however, if the brain were to cease existing after the mind uploading process? Things now become more interesting. A pattern describing the history of the personality without any uploading onto the computer substrate simply ends when the brain ceases to exist. Any such pattern, however, will be contained in some of the patterns describing histories in which the personality gets uploaded into a computer. If we were to take the view of people advocating quantum immortality, translated into this context, we would say that any observer in one of the patterns that truncates when the brain ceases to exist should merely be viewed as being part of one the patterns involving mind uploading that extends this pattern. If this is correct, we have a strange result: a person trying to use mind uploading just before death to escape can expect it to work, even though a person having a mind uploading type process performed after which his brain continues to exist may well have a reasonable expectation, depending on what the relative measures are like, before the process, that it is unlikely that he will suddenly find himself uploaded.

I know that my discussion about this ensemble situation and continuation of consciousness in it has been a bit rough and could do to be deeper. One reason for this is that I am still formulating my own opinions on ensemble views in my series of articles about Occam's razor and these are just some preliminary thoughts. When that series has got further I may return to this issue within that context for a more detailed discussion.

Of course, if the ideas of the quantum immortality advocates are right it could be argued that we may not need to bother with mind uploading at all, nor would we need to bother with the kind of many-worlds assisted mind uploading process discussed at the start of the article, as some sequence of quantum decisions (or some pattern allowing continuation) would always present itself. We will look at this issue shortly.

Substrate Independent Patterns and Measure

What I have just said may seem somewhat messy. One way to make this easier to think about may be by introducing the idea of substrate independence.

We have previously discussed the idea of ensemble view of reality in which patterns are considered real by virtue of being extractable by algorithms. The idea of substrate independent patterns will rely on the idea of a ensemble view that is hierarchical; that is to say one in which algorithms extract patterns and these patterns themselves can have patterns extracted from them by other algorithms, and so on. A hierarchical view of the ontology of reality seems quite reasonable to me: it fits in well with our intuitive view of the world as hierarchical. In such a hierarchical view, something would exist by virtue of its pattern being extractable by an algorithm, or by virtue of its pattern being extractable from other patterns by an algorithm and so on. The measure of a thing would relate to the proportion of all possible algorithms that will actually extract its pattern. This is a concept that I start to discuss (though I have not discussed it very far at this stage) in my previous article Occam's Razor Part 6: Partial Models as "Envelopes" [10].

There are two basic ways in which we could look at patterns describing the history of a person.

One way would be to consider a substrate dependent pattern describing the history of a specific physical system, such as a brain or computer, that is supposed to do a person's thinking. This is what we did in the previous discussion.

Another way would be to consider a substrate independent, more abstract sort of pattern describing the history of a person's mind independently of a substrate. This is not incompatible with the previous view: any instance of a substrate independent pattern is likely to exist by virtue of being algorithmically extractable from one or more substrate dependent patterns.

This deals with Searle's criticism of strong artificial intelligence as being based on an incoherent idea of algorithms. In this view we would associate intelligence with a brain because there is the capability to extract patterns from the history of a brain that describe the history of a mind in a substrate independent way; that is to say in a way independently of the physical system that is causing the mind. Similarly, we may extract patterns from the histories of certain computers that describe the histories of minds in a substrate independent way.

In a given situation there will be many ways of extracting the same type of substrate independent pattern and the chance of finding yourself being "caused" by some physical system, such as a brain or a computer, could be considered to be related to the measure of substrate independent patterns of your mind's history being extracted by algorithms from the pattern of that physical system's history. The measure of the pattern for the history of the physical system itself would also, of course, need to be taken into account, as greater occurrence of such patterns would give more scope for extraction of substrate independent patterns from them. Some of the measure for the substrate independent pattern would, however, be related to the nature of the physical system itself.

This suggests that to have a good chance of "finding yourself" being caused by a physical system then:

• The pattern describing the history of the physical system should have high measure.
• The substrate independent pattern describing the history of your mind should be derivable with high measure from the pattern describing the history of the physical system.

It may seem strange to suggest that the nature of a physical system could determine the measure with which substrate independent patterns of a mind can be derived from it. To explore this, let us imagine making an uploaded copy of a person using two identical computers, with identical software. "Conventional" thought by advocates of mind uploading about this situation would suggest that this sort of duplication is irrelevant. Considering this from the point of view of measures of algorithmically extracted patterns in an ensemble view gives a different result. Having two instances of the computer and its software means higher measure for the sort of pattern describing the computer: whatever measure the computer's pattern had is effectively doubled. This gives more scope for extracting substrate independent patterns describing the history of a mind from the pattern describing the history of such a computer. If we consider a situation like this then an obvious thing to ask is what happens if we start to somehow move the computers closer together so that they eventually overlap and become effectively just one computer. Where has the extra measure provided by the duplication of the computers gone when this happens? If we are to be consistent we must say that it has gone nowhere. It is still there and we have effectively produced a single computer which provides more measure for the substrate independent pattern for the history of a person's mind than either of the two computers with which we started. This makes sense if we accept that some substrates are described by patterns that provide greater measure of algorithms that extract substrate independent mind descriptions than others.

When a substrate independent pattern of a history is derived from a situation involving a change of substrate (for example, a history in which uploading to a computer occurs) why should the algorithm generating it to have greater complexity. This is because some of the information needed to generate the substrate independent pattern is specific to the substrate from which the information is extracted: the pattern extraction algorithm will need to contain information to "read" the substrate dependent pattern and convert it to a substrate independent pattern. If there are multiple substrates then information will probably be needed to deal with each substrate in this way. Multiple substrates will therefore tend to require a more complex, longer algorithm to make a substrate independent pattern and will tend to result in a loss of measure.

A further issue affecting the measure of a substrate independent pattern of a mind is that of discontinuity. Let us suppose that we want to consider the case of a person with a biological brain who later finds himself uploaded into a computer simulation. The way to consider this would be to consider the measure of appropriate substrate independent patterns extracted from the pattern of the biological brain and the pattern of the computer. The discontinuity in the history when the "transfer" is made from the brain to the computer will have an effect here.

Discontinuity relates to transition between situations. Whether or not the "situation" has any place in a substrate independent pattern depends on whether it relates to the substrate or not. Some sorts of discontinuity - for example, a person finding himself in a different place, or experiencing different things, as a result of mind uploading should still have this change in experience reflected in any substrate independent pattern. On the other hand, the change in substrate itself clearly should not be referenced by a substrate independent pattern which should not make any reference to specific situations. Whether various aspects of discontinuity are incorporated into a substrate independent pattern of a mind or not, any algorithm that generates a substrate independent pattern, where discontinuity is involved, will need to extract information from different patterns. This will make it more complicated - though not necessarily much more complicated - and increase its length, thereby decreasing the measure of this sort of pattern. This means that substrate independent patterns for the history of a mind could have a further reduction of measure when extracted from situations where the complexities of change of substrate are involved, such as a person finding him/herself in a biological brain at one point and then uploaded.

Some readers may be confused by the difference between the complexity caused by multiple substrates and discontinuity. The complexity caused by multiple substrates is because different patterns have to be interpreted to produce a substrate independent pattern. The discontinuity issue is more specific and relates not just to the existence of different substrates, but the extra complexity caused by the transition itself. Both of these will tend to reduce measure.

Is this strong continuity of self?

Given what I have just said about representing histories and patterns, it could be asked if this shows strong continuity of self to exist and if this refutes my earlier article arguing against strong continuity of self.

This is debatable and I may form a more definite opinion on it later. An obvious answer could be that the issue of what an acceptable state would be that amounts to continuation is simply replaced by the issue of what an acceptable pattern would be that would describe the self, which in turn would come back to what sort of continuity would be acceptable. If such a view is correct, in the end it could all come back to evolution and the survival motivation.

As an example, and I know it is a rather silly one, it should be possible, in an ensemble type view, to make a pattern describing how I exist by virtue of my brain for the next ten minutes and then "find myself" in a nearby potted plant - a state which would probably not involve too much awareness of what has happened to me. Why does it sound silly? We automatically reject that sort of future as meaningful, yet many such absurd patterns could be constructed. It is an entirely reasonable question to ask why we expect our consciousness to "know" how to stay in systems that have complex processing capabilities.

Why do we reject most of these absurd histories for ourselves? If strong continuity does not exist the reason is probably that thinking in these terms does not really help organisms make copies of themselves. We have this idea of what sorts of histories really can be ours, possibly, because the sort of an organism that really thought it could view something like a watering can as an acceptable continuation is not going to do a very good job of surviving and replicating.

Another possibility is that we may consider ourselves to exist in an ensemble as a pattern, extracted by an algorithm, describing a history of our mind in some way. It could be that any pattern describing a mind that has an established history has higher measure if it continues to describe a mind that does the same sort of things. This would be rather like applying an ensemble-type version of Occam's razor to a mind itself to suggest that any mind can reasonably expect its next state to be the same sort of thing - that is to say something which meets some specification of being "a mind" - as any algorithm that describes the pattern for a mind has to contain the information to do this anyway and extra complexity, and therefore an increase in meaning extraction algorithm size and a loss of measure, may result if a mind does something not expected of a mind. This approach would allow brains and uploading, but would deal, almost, with really stupid ideas like minds becoming incoherent patterns, shrubbery or furniture.

If it did turn out that we could expect to be in certain patterns describing the history of our mind and not others, solely on the grounds of measure, without any arbitrariness being required by "survival motivation" would it imply strong continuity?

I am not entirely sure. It still seems to me there is an element of weak continuity here, although I have to admit that this issue troubles me. We would still have to make the decision to accept this as explaining our subjective experience. It may seem to fit our subjective experience, but how would we answer someone who said that they don't care about all this: as far as they are concerned they have continuity as a result, specifically, of brain activity? It would still be hard to disprove such a position. The very decision to adopt such a methodology as representing our continuity is arguably a weak continuity position.

A further issue could be that such a measure based approach to deciding what acceptable patterns of a mind's history are does not eliminate silly possibilities: it merely relegates them to low probabilities. It is debatable whether or not a position that still allows any future configuration of matter, to be a future state of the mind can be considered a strong continuity position.

If this sort of approach is weak continuity this does not mean that it is not useful. In fact, if such a thing can be said to be make sense at all, it does not seem to be very far from a strong continuity approach - potentially even close enough that semantics start to become an issue. Why should this be the case?

In an ensemble approach all that can be meaningfully said to exist are patterns. Patterns can be extracted from reality by algorithms. Let us imagine that an ensemble approach is hierarchical, and becomes infinite so that there is nothing more than an infinite extent of possible algorithms with the capability of extracting patterns from other patterns, which in turn can be extracted by algorithms, and so on. Some positions on weak continuity may reject the idea of preserving patterns to try to continue consciousness, for example by using computers, arguing instead for the physical specificity of human brains, but this view would be untenable in this sort of ensemble view in which everything is a pattern. A human mind would be a pattern just as much as the software in a computer is. In an infinite, hierarchical ensemble, it would not even make sense to say that a particular pattern like a brain was somehow more "fundamental" than others in any really important sense, as there would be no bottom to the hierarchy and, in any case, patterns could reoccur at different levels in it.

Suppose that we took a substrate dependent view? Someone could argue that a system is not viable for continuing a mind as it is "not the right substrate". If a computer were able to generate a certain pattern such people would say that without the right substrate underpinning it this would be irrelevant. For example, a computer may produce the correct behaviour but it could be argued that it is merely producing a simulation of a substrate independent pattern while any consciousness and continuity has to come from the correct underlying substrate. In this sort of hierarchical, infinite ensemble view, however, anything, including a particular substrate, only has the status of a pattern anyway: there is nothing else. There is no ontological difference between a simulation and a real thing. The "pattern" produced by a computer's software could be said to be a simulation, if we liked, but if we take that approach we could equally well regard a brain as a simulation, a pattern, extracted from the underlying matter, which is itself patterns. This means that differentiating between real and simulated things would fail in an ensemble view. This point, by the way - that once we adopt a particular way of looking at the world everything simply becomes patterns extracted by algorithms - has catastrophic consequences for John Searle's arguments against strong artificial intelligence [11] - an issue to which I may return in a future article.

Someone could attempt to save such an argument by saying that the substrate does matter, while recognising the substrate merely as a pattern. Such a view would be proposing that a particular substrate dependent type of pattern is necessary to describe the history of a mind. Such an argument could be used to propose some type of weak continuity based on an arbitrary choice of pattern for the substrate. The most likely way for this to happen is for someone to declare specific patterns describing brain physiology to be necessary for continuity. In this sort of hierarchical ensemble view however, a pattern could imply the existence of patterns above it in the hierarchy, which is another way of saying that a pattern could have other patterns extracted from it by algorithms. This means, for example, that if we decide that a particular biochemistry specific substrate (probably a brain description) is required then we have the problem that some other system, such as a computer, could be programmed to arrange itself in such a way as to allow the biochemical pattern to be easily extracted from it by an algorithm - that is to say, to simulate the biochemistry of a brain, and to do this with sufficient detail to satisfy the requirements of the biochemistry specific substrate pattern - or worse still, and maybe ontologically problematically to some people, an infinite number of algorithms could be available to extract any pattern from any other, the only effect of the original pattern being on measure in the layer of the hierarchy above it. This would imply that an infinite and hierarchical ensemble has patterns that spawn patterns anyway without humans even using computers.

Nor would the argument of "Yes, but the computer's simulation of the brain chemistry is not real" be of much use. If a person had accepted that the substrate is itself a pattern he/she would be accepting this for brains as well. In fact, the word "real" would be getting less useful at this point.

All of this means that demanding a particular sort of substrate in this sort of ontological view does not seem to achieve what a person using such an argument would want. It may limit the concept of continuity to specific patterns, but the idea of using such an argument probably is not to allow computers to cause such patterns, or for the patterns to reoccur all over the place in the hierarchy merely due to them being extractable by algorithms.

Someone may attempt to get out of this by saying that, even if everything is patterns, the pattern describing the history of a brain is different to other patterns that could be claimed to achieve the same sort of continuity because the underlying patterns from which it is extracted are different. This argument would be saying that it is not just the pattern itself that matters, but also the underlying hierarchy of patterns from which it is derived. This argument is weak. Firstly, we would be justified in asking why the underlying pattern is important. Is there anything to justify this requirement other than an arbitrary decision? Secondly, even if we accepted this idea, if something like a computer can be arranged so that it is possible to extract a particular substrate dependent pattern from it then it could be similarly arranged to have as many levels of hierarchy as we wanted extracted from it. This could also be expected to happen "naturally". Just as a particular pattern could occur anywhere within a hierarchical ensemble, so could any pattern together with any specified levels of the hierarchy underneath it. The objection would then have to be that continuity is dependent on the absolute position in the hierarchy of a particular substrate dependent pattern.

I think that, while such a substrate specific view could be forwarded, it starts to appear arbitrary in this sort of ensemble view, and it becomes difficult to take such arbitrary ideas of continuity seriously.

So, what if we adopted a position on continuity based on the histories of minds being described by substrate independent patterns? It could be argued that there is some arbitrariness with this, because we would have to decide whether a pattern describing a transfer from one situation to another was acceptable or not, but in any situation in which we had some sort of transfer we would be able to derive a higher level pattern that seamlessly fitted the two parts of the history together and did not explicitly mention a transfer between situations, although this would be done at some cost in terms of measure: the extra complexity needed in the algorithm that derives the pattern would reduce the frequency with which that kind of pattern is derived.

All of this would seem to make it difficult to maintain various arbitrary views about continuity. Any reasonable idea of continuity would seem likely to be based on having some very general way of describing a pattern that represents the history of a mind over time and if it can be shown that an observer is more likely to find him/herself in some patterns than others, where a pattern describes his/her history, does strong continuity of self follow from this?

There are some issues to be resolved here. Even if some patterns for the history of an observer's mind have higher measure than others, it does not mean that one of these must be the right one: it still leaves room for an observer to be part of an absurd description. This could be used as an argument that it does not provide some non-negotiable, strong continuity because we could imagine any future for an observer and find some pattern, with a non-zero probability of being the one which describes the observer's history, containing that future. The idea would be that the observer may not be sure in which pattern, out of all patterns, he/she is contained, so it could be argued that it is in fact incoherent to say that the observer, is even "in" a single pattern. Against this, it could be said that it does work if we accept that continuity can be meaningful in an absolute and statistical way - that, if different patterns have different measure, not all patterns are equal, that some patterns for an observer's history can be said to be "the right ones" by being more likely than others, and that they are more likely in a very real way, rather than due to some preference.

Another issue is that of discontinuity. An objection could be made that even if a substrate independent pattern is used to describe a mind, we would still have the issue of transferral to different substrates or situations and we would need to decide what sorts of transfers were valid. This objection is not very strong. If part of the history of a mind is in one situation, and part of it is in another situation then it is not necessary for a pattern describing the mind to describe any transfer. A substrate independent pattern of the mind's history could be derived from the patterns describing it in each situation and this pattern need not feature any "discontinuity" at all. The discontinuity would be removed in the extraction of the substrate independent pattern, leaving a "seamless" history, though not without some expense in terms of the length of any extraction algorithm and, therefore, measure.

We would also still have the issue of deciding what would be needed for a substrate independent pattern containing a mind's history. What would a pattern need to contain to capture our feeling of existing as an observer? Even if we cannot be sure about this, we could still do a lot of serious philosophy with a vague idea of what such a pattern has to achieve. Not having a precise idea of the sort of pattern that would be needed does not mean that we will never have it in future. Some readers may note that this is, essentially, what is known to many people as the reference class problem in so much as it relates to the "what is an observer?" issue in such areas as the Doomsday argument.

So, where does this leave strong continuity?

Where does this leave the issue of strong continuity of self? It seems to me that, with this sort of ensemble view, a persuasive argument can be made that an observer can be regarded as being contained in a pattern that describes a history. The adoption of this sort of "ensemble pattern of a history" approach does suggest to me that we could define a way of extracting patterns that would be much less arbitrary than other ways of defining weak continuity, because it could be done in a way that would not need to deal explicitly with any particular situations, but would simply declare some sort of minimum standards that a pattern has to meet. Whether or not this amounts to strong continuity would depend on a careful consideration of the semantics involved in a claim for strong continuity and how well the issues described above can be resolved.

At this stage I accept that my earlier argument against continuity of self may be wrong. I do not want to go further than that now, simply because I do not want to have to keep changing my mind in public. If I am wrong, and I do need to change my position on strong continuity, I would like to get it right next time.

If strong continuity cannot be justified, it at least seems likely to me that some ways of dealing with weak continuity that focus on very specific substrates or processes start to seem very arbitrary indeed and others will not have the same problems, simply by making less specific demands. If we do have to continue with weak continuity, in such a hierarchical ensemble, we should at least be able to have much less arbitrariness.

Personal Singularity

In this article I am not really taking a definite position on the issue of quantum immortality, but let us assume that it is true. If you happened to be "continued" by some non-conventional means what would it be like?

Our previous discussion relates to this. If we take the concept of quantum immortality seriously at all then measure has to be regarded as the indicator of how likely we think various experiences are.

Let us start with you having a normal life, with a normal brain, and experiencing quite normal events. There may be some futures which split off from yours in the many-worlds interpretation in which quantum events very quickly make "Alice in Wonderland" type events start to happen. This is unlikely to happen, from your point of view, because these sorts of futures are low measure. If we now extend our ontological scope and consider some kind of ensemble, in which space and time are merely derived things and the many-worlds interpretation may just be a special case, then we have scope for many "copies" existing in various places that would be as reasonable as candidates for your "next state" as what you expect to be your "next state" in one of these mundane futures. Why should you not expect one of these to be your "next state"? The answer, once more, is measure: a pattern describing your history as containing mundane events can be expected to have higher measure, making it more likely you are in such a pattern.

Let us imagine that you are going to copy your brain's physical structure into some sort of computer simulation and your brain is going to continue existing after you do this. Looking at things from a probabilistic view, there is a chance that you will find yourself actually in the situation of the uploaded copy. This is not necessarily 50%: it will depend on the measure of a pattern describing a history where you continue "in your existing brain", the measure of a pattern describing a history where you find yourself in the computer and the (much smaller) measures of lots of other, weirder, histories that we have not considered here and that are too unlikely to merit serious consideration.

Let us imagine now that you are still the version of you that is associated with your brain. You undergo another mind uploading process, but this one is different: at the end of it, your brain is destroyed. In this situation, the status of a pattern involving the uploaded version of you has changed totally: it no longer has competition from the biological pattern. While its measure may have been small before, in comparison to the biological pattern, without this pattern it is now very significant, probably to the point of meaning that it describes your most likely future, unless you resorted to any other extreme measures that we are not discussing.

We cannot have certainty that the uploaded version of you is your sole possible future. If we take quantum immortality ideas seriously, and the many-worlds interpretation is correct, we could argue that it is rather presumptive to even say that your brain has ceased to exist. There may be some futures, after what we presumed to be your death, in which various sequences of quantum events contrived to keep your consciousness going. What would these sorts of situations be like? We cannot really say. They would rely on what we would classically regard as outrageous coincidences to preserve your consciousness somehow.

We also have the possibility that your future could lie in patterns that do not involve simple continuation into one of the many-worlds interpretation possible futures, but simply rely on the existence of some non-causally connected copy of you in another world in a many-worlds model, or anywhere in some more general ensemble. We can expect such futures to have low measure. Furthermore, any substrate independent pattern that encompasses your biological existence and one of these futures is likely to incur some cost in terms of discontinuity and multiple substrates.

Both of these kinds of future continuations - the Everett kind and the non-causally connected kind - are dependent on very improbable coincidences. We would find it very difficult to make predications about what will be going on in these futures. Fortunately, their very reliance on improbable sequences of events makes them very low measure. From the point of view of someone facing all this in the future, before the mind uploading, you should regard the mind uploading future as almost certainly "yours".

What if you are not trying to use mind uploading, but instead use some method that has a low probability of success - some method of extending your life that could conceivably work and which is not totally insane, but is maybe just a bit unlikely? Strangely, it may be that if quantum immortality makes sense you should expect to find yourself where this method happens to put you. This is because, even if it is unlikely to succeed, the measure of worlds in which it is successful may still be much higher than the measures of all the worlds in which you continue in unknown, arbitrary ways.

Let us suppose now that you do not attempt any radical method of continuation, simply relying on the idea of quantum immortality to sort things out for you, and that the quantum immortality hypothesis is true. Any continuation will now be reliant on either unlikely sequences of events happening, due either to fortunate sequences of quantum events occurring, or to the appropriate pattern happening to exist, in a non-causally connected way, somewhere else in an ensemble. Such continuation of your consciousness - if it occurs - may seem attractive, but is that really an appropriate view? Relying on the automatic provision of continuation by reality means finding yourself in the sorts of situations that would be considered wildly implausible if you had anywhere else to be. What would these ultra-low measure situations be like? We probably cannot have any idea at all. We normally make predictions by working out which situations have high measure. Of course, if we could examine possible situations carefully we may work out which of all these ultra-low measure situations have higher measure than others, but that may be easier said than done: it would probably be very difficult to distinguish between all the ultra-low measure situations like this.

I have decided to call this personal singularity. Personal singularity is a hypothetical situation in which your future continuation becomes dependent upon what are, from your point of view, ultra-low measure situations such that it is difficult to make predictions.

Personal singularity means that there is no guarantee that such situations would be desirable. In fact, as it is probably a lot easier to make things undesirable than desirable, it is quite likely that these low measure situations would not be very desirable.

Personal Deterioration

One thing that could be an obstacle with quantum immortality is the idea that your mind could somehow become compromised. Suppose this happened in most of your futures? It could get continually worse, with every version of your mind that exists facing a future of further deterioration. Quantum immortality in itself would not save you from such a fate as it would merely guarantee the extension of a diminishing consciousness.

The problem with personal deterioration is that, while it is easy to philosophize about what could happen to prevent death - we have been doing this some time in this article - the idea that our mental faculties can increase and decrease is quite obvious to us. As an example, even if I could make a viable argument for continuation of human consciousness, it would be difficult for me to persuade you that if you drink a lot of alcohol there will be no detraction from your thinking ability. If we can imagine some sort of possible deterioration in any situation then there seems to be nothing requiring human consciousness to stay intact.

This would provide a less reassuring answer to the issue of quantum immortality. It would suggest that your "death" could involve a deterioration in your mental functioning, which occurs with very high measure and, after this has gone on some time, any discussion of whether you continue or not is worthless to you.

How can we approach this issue? One thing that we should consider is the conventional continuity "paradox" that we discussed earlier. If it is possible for a lot of alternative futures to exist for any instant that are utterly weird then we could raise the same issue about personal deterioration: if we are going to worry about personal deterioration getting in the way of immortality we should also worry about it getting in the way of minute to minute existence by providing many ways for you to deteriorate at any moment in time.

We do not seem to experience such personal deterioration, just as we do not seem to experience a failure of conventional continuity. A reasonable answer is the same as that used to answer the "paradox" of conventional continuity - that patterns describing "normal" continuity have a much higher measure than others. In this context such an answer would mean that algorithms describing patterns in which people continue to exist "conventionally", without deterioration, have higher measure than those which describe patterns in which people suffer abrupt deterioration for no good reason. This does not mean, however that this continues indefinitely, but merely while patterns in which deterioration is not occurring have higher measure as part of the "conventional continuation" that was discussed earlier.

Consideration of this issue may be affected by any decisions we make about what sorts of configurations of matter, or patterns, are plausible (if not necessarily desirable) continuations of people. Some people would call this a reference class issue. It may be that we need to rule out various futures just to ensure conventional continuity which is free of deterioration - just to ensure that your moment to moment existence now - continues conventionally, rather than from various patterns that could exist that continue you in a deteriorated form. On the other hand, it may be that no reference class assumptions are required to explain conventional continuity and lack of deterioration and that the explanation suggested earlier - that, while we enjoy conventional continuity, patterns describing conventional continuation simply have higher measure on statistical grounds is adequate.

This matter is somewhat involved to consider, but as with personal singularity it is more likely to become an issue when you run out of viable ways to continue your consciousness, so that any situations in which you are continued in some compromised way assume a higher relative measure. At this time, deterioration could become an obstacle to quantum immortality. This issue needs some consideration, but we should also consider whether or not processes like many-worlds assisted mind uploading, or other processes intended to exploit "loopholes", could have any effect here.

There is also some overlap between personal deterioration and personal singularity. Some of the ways in which personal singularity could occur could include deterioration.

"Survival" as Measure Control

If quantum immortality is a reasonable hypothesis it could be asked why we should bother trying to survive at all. Why would anyone bother trying a real version of the thought experiment considered in the article? Why should anyone try anything like "conventional" mind uploading? Why even bother fastening a car safety belt or looking before crossing the road?

If the quantum immortality hypothesis is correct the answer lies in measure. Right now we can consider some of our possible futures to have high measure - meaning that we can consider it likely that we will find ourselves living in one of them. It is for this reason that we expect the world to be reasonably predictable and for any of our plans and actions to matter.

Even if the quantum immortality hypothesis were correct, not engaging in survival would invite personal singularity and possibly personal deterioration. This would mean accepting unpredictability about circumstances and, if the statistics of patterns suggest it or the reference class issue is resolved in certain ways, could even involve you being compromised to such a degree that you may not even regard it as consciousness. It will always be in your interests to try to maintain a high measure for a reasonably predictable and coherent set of futures, then you can act so as to select those futures which are preferable. Any "survival" behaviour can therefore be viewed, in a quantum immortality context, as attempting to avoid low measure situations.

If the thought experiment did work what would be the point?

It could be argued that the thought experiment will only work if the universe functions in such a way as to make the quantum immortality hypothesis correct and that, if the thought experiment would actually work and continue a person's consciousness, reality itself would do this anyway by throwing up obliging sequences of quantum events or patterns that provide suitable continuation. Given this, what purpose does the thought experiment serve, and could there ever be a case for anyone performing the procedure that it describes?

One use of the thought experiment is to make one objection to many-worlds continuation of consciousness redundant. This is the objection that the many-worlds interpretation does not imply that a sequence of quantum events would always be possible to guarantee continuation of a person in some future and that you may die in all futures. Although this could be argued against by suggesting that patterns equivalent to mind uploading you could exist for non-causally related continuation, the same sort of objection could be made against this, or the lack of causal relationship in such continuation could be made an issue. The objection could be also be opposed by suggesting that the many-worlds interpretation is simply a special case of some infinite ensemble in which all patterns occur, but such a view would be controversial and the same sort of argument about such patterns not being causally connected to "our world" could be made. Some readers may point out that causality does not really have the same status in some ensemble views, and I happen to hold the view that causality is something that appears in algorithmically extracted patterns rather than anything fundamental, but it remains that this is controversial.

The point of the thought experiment is that it avoids all of this issue of "would consciousness be able to continue by itself without humans doing anything?" by forcing the issue. It does this by setting up a scenario where the sequence of quantum events that should continue a person's consciousness is inevitable. This does not mean that the thought experiment easily proves that consciousness can be continued in this way: there are other issues relating to the validity of the many-worlds interpretation, the validity of the mind uploading idea and measure. What the thought experiment does achieve is the disposal of this issue of whether or not the appropriate coincidences to continue consciousness would occur, to allow consideration to focus on these other issues. In this respect the thought experiment is facilitating philosophical discussion about what we are and what continuity means without a real version of it needing to be performed.

The thought experiment is, however, in such an emotive area of human interest - life and death - that it would be hard not to consider really doing it. If someone thought that the quantum immortality hypothesis were true would he/she have any reason to really use many-worlds assisted mind uploading in an attempt to continue his/her consciousness?

We could start by asking why someone would prefer such an approach to "conventional" mind uploading. One answer is that the technology to make a sufficiently accurate "scan file" for mind uploading may not exist during a person's lifetime and a method like this would not require such technology. Furthermore, the computing capabilities needed to run such a program need not exist during the person's lifetime as the running of the simulation does not even have to be set up specifically for a particular person and can be done afterwards. This, of course, is bound to have enough of a "free lunch" appearance to it to cause suspicion from readers and it is suspicion that I share.

Even if the approach in the thought experiment does have some advantages over "conventional" mind uploading, we could ask why someone who accepts the idea of quantum immortality would bother doing it in reality. Why not just wait for a lucky sequence of quantum events in the future or the lucky existence of the appropriate pattern "somewhere else" to continue him/her? Why bother doing something if reality is expected to do it anyway? There are two answers to such a question:

Firstly, someone may think that the broad ideas behind quantum immortality are valid while not being certain that reality is guaranteed to provide the necessary things to give continuity. This would make a process like that described in the thought experiment an insurance policy.

Secondly, there are the issues of personal singularity, personal deterioration and measure control that we previously discussed. Even if the quantum immortality is correct, relying on it could invite personal singularity or personal deterioration, whereas a process like that in the thought experiment may be used to try to generate viable and desirable continuation possibilities. In this respect, trying the thought experiment in reality would be an extension of other processes such as using a car safety belt.

An objection could be made to this measure control justification, however. The process in the thought experiment is spectacularly wasteful in terms of what it needs to do to upload someone. It uses many-worlds splitting of the world to generate a huge amount of worlds, with different computer simulations in them, simply to create a small number of worlds with the right kind of simulation. This could be considered to be a dramatic loss of measure, so much so that if any "conventional" means of continuation were available we would probably not regard such a process as providing a serious candidate for a person's future. With all the conventional means of continuation, along with their competition for relative measure, gone, however, we might take the scenario in the thought experiment more seriously, but there is a problem: what if the measure of such continuation is not much more than the measure of various other ways we have not thought of in which continuation may occur anyway? As an example, what if we set up a situation like that in the thought experiment, but the measure of a pattern in which such continuation occurs is not significantly better than the measures of various patterns which involve random quantum events continuing consciousness in unforeseen ways? It may be that, from the point of view of a person about to undergo such a process, the probability expectation of being continued by many-worlds assisted mind uploading is not significantly greater than the probability of being continued, unintentionally, in any one of many other ultra-low measure ways. If this is the case then the process in the thought experiment would not contribute much to avoidance of personal singularity. This would be an unusual situation in which quantum immortality could be viewed as actually getting in the way, rather than helping.

One way of answering this would be to try to show that the measure provided by continuation in the thought experiment is significantly higher than that which may be provided by other means. I do not intend to do this in this article. Another answer is to suggest increasing the measure of continuation provided by the thought experiment's process.

One of way of increasing the measure in the thought experiment would simply be to keep doing it. If we do it a hundred billion times then, surely, the scope for algorithmically extracting patterns from it will be a hundred billion times better? An objection to this could be that the measure in the thought experiment is so ultra-low that it may not be practical to perform the vast number of repetitions needed to obtain any decent amount of measure.

Another way of increasing the measure provided by the thought experiment could be to adapt it more specifically to a particular individual. One obvious way of doing this would be simply to use "conventional" mind uploading and actually obtain a "scan file" for the individual's brain, but this does not particularly exploit the many-worlds interpretation and is not what this article is about. Another way would be to include some known information about the individual that falls short of being the complete brain description obtained by some futuristic scanning process. In a previous article Indirect Mind Uploading: Using AI to Avoid Staying Dead [12] I discussed how such information could be obtained for the purpose of what I called indirect mind uploading. Such data could be personal diaries, video recording, the sorts of crude brain scans that we can do and many other sorts of information that we could obtain about a person with available, early 21^st century technology. Suppose we provided this information about a person to the computer that is going to generate the "scan file" and required the computer to generate a brain description that is consistent with it, in much the same way that is proposed in the indirect uploading idea? The computer could use the information that we have provided to generate part of the information in a brain description and then use the sort of quantum event process in the thought experiment to generate the rest of the description in a strange hybrid of indirect mind uploading and many-worlds assisted mind uploading. The advantage of this is that the amount of information needing to be generated by quantum decisions will now be reduced, increasing the measure of worlds where the right information is generated and possibly helping with the issue of ultra-low measure.

Conclusion

This article has proposed a thought experiment in which the many-worlds interpretation is used to continue human consciousness after destruction of a person's brain in a way similar to mind uploading, but with the rather extreme idea of using a sequence of quantum decisions to generate the information that is needed to do it in some world where the quantum decisions happen to have given the correct results.

It would be wildly presumptuous of me to say that this would work. This is a highly speculative article. The thought experiment does involve an extreme suggestion and it should be noted that there are a lot of conditions associated with it working in reality. One obvious such condition is that the many-worlds interpretation needs to be true, or some kind of ensemble view of reality needs to be correct.

Rather than being taken as an attempt to seriously propose a method for a real process I hope that this document is regarded more as part of the ongoing debate about what human minds are, what continuity means and how such ideas may need to be reassessed if the many-worlds interpretation or some sort of ensemble view happens to be correct.

References

[1] Web Reference: Wikipedia Contributors. (2002-2006). Quantum Suicide. Wikipedia, The Free Encycopedia. Retrieved 1 March 2006 from http://en.wikipedia.org/w/index.php?title=Quantum_suicide&oldid=39372228.

[2] Everett, H. (1957). Relative State Formulation of Quantum Mechanics. Reviews of Modern Physics 29, pp454-462.

[3] Web Reference: Vaidman, L. (2002). Many-Worlds Interpretation of Quantum Mechanics. Stanford Encylopedia of Philosophy. Retrieved 19 August 2005 from http://plato.stanford.edu/entries/qm-manyworlds/#1.

[4] Web Reference: Wikipedia Contributors. (2002-2006). Quantum Immortality. Wikipedia, The Free Encycopedia. Retrieved 28 February 2006 from http://en.wikipedia.org/w/index.php?title=Quantum_immortality&oldid=41200689.

[5] Web Reference: Strout, J. Mind Uploading Home Page. (2002). Retrieved June 22, 2003 from http://www.ibiblio.org/jstrout/uploading/MUHomePage.html.

[6] Web Reference: Mind Uploading Research Group. (2002). Retrieved June 22, 2003 from http://minduploading.org/.

[7] Egan, G. (1994). Permutation City. London: Millennium. (Fiction).

[8] Web Reference: Standish, R. K. (2002). Why Occam's Razor. Retrieved 24 December 2005 from http://parallel.hpc.unsw.edu.au/rks/docs/occam/occam.html.

[9] Web Reference: Almond, P. (2004). Did I Write This? Retrieved 8 October 2004 from http://www.paul-almond.com/DidIWriteThis.htm.

[10] Web Reference: Almond, P. (2006). Occam's Razor Part 6: Partial Models as Envelopes. Retrieved 26 February 2006 from http://www.paul-almond.com/OccamsRazorPart06.htm.

[11] Searle, J. R. (1980). Minds, brains and computers. The Behavioral and Brain Sciences 3:417-457

[12] Web Reference: Almond, P. (2003). Indirect Mind Uploading: Using AI to Avoid Staying Dead. Retrieved 9 August 2003 from http://www.paul-almond.com/IndirectMindUploading.htm.