Introduction

Our limited lifespan is inconvenient at best. Let’s look for a loophole! I want one that we can use now rather than one that someone else can use in a few hundred years. I also want a loophole that involves artificial intelligence (AI), mainly because this is an AI article.

I shall go through some possible loopholes and then discuss what I shall call indirect mind uploading, the particular idea that is the subject of this article.

Not dying in the first place

If being dead is undesirable then not dying in the first place seems to be a good way of trying to avoid ending up like this. Some people think that this is quite achievable in the future, at least if one wants to delay death for a very long time and not necessarily forever; others are sceptical that it will ever be possible.

Proponents of the idea of using scientific knowledge to delay death indefinitely can point to work in biochemistry that has lead to a greater understanding of the basic mechanisms that control life and they can also point to the idea of molecular nanotechnology [1, 2, 3, 4, 5] proposed by K. Eric Drexler. This idea involves acquiring the ability to build structures to an accuracy of 1 nanometre (one millionth of a millimetre), allowing machines to be built on the molecular scale. These machines would be able to manipulate individual atoms and could easily synthesise new molecules or make changes at the molecular level to biological systems. The hope is that they could perform medical intervention at a much greater level than is possible with our current ‘bulk matter handling’ technology.

There is one problem, though: I want to present a method that people could use now and, whether one accepts the idea of nanotechnology or not, all this would seem to require some science and technology that is not available today.

Using cryonics to avoid staying dead

When a person dies, he/she is considered dead because the body is severely damaged, but what is regarded as death depends on our ability, or lack of it, to reverse the process. We no longer consider a person whose lungs and heart have stopped to be necessarily dead because there is still a chance of intervening to restore functioning to them, but if cardiopulmonary resuscitation were not known to us we would probably declare death at this point.

There may be some other conditions, now considered to be death, which a future society will consider merely to be very serious forms of damage, from which recovery is possible. This is of little help to us though; soon after an individual dies, dissolution, or loss of the structure of the body, tends to follow, presumably putting hopes of a cure beyond the reach of any conceivable technology.

Ettinger [6] and Cooper [7] attempted to address this problem in the 1960s by introducing the idea of cryonic suspension [8]. (The word cryonics is a contraction of cryogenics, which means the science of low temperatures.) The idea is actually quite simple: when a person is considered dead, by our standards, the body is preserved by means of extreme cold to prevent dissolution occurring, the idea being that technology decades or centuries in the future will have advanced to the point where the individual is no longer considered dead, but is merely considered damaged and can be restored to health.

Reviving a person who has been preserved in this way would not be trivial and cryonics advocates do not tend to pretend otherwise. For a start, the individual would have whatever damage caused cessation of functioning in the first place. He/she would also have a lot of secondary damage that occurred as a result of cessation of functioning, for example, as a result of cells in the body being deprived of oxygen. In addition the use of low temperatures, typically provided by liquid helium or liquid nitrogen, to preserve the individual seems likely to cause damage at the cellular level. Proponents of cryonics suggest using chemicals that serve an antifreeze function, preventing damage due to freezing, although these introduce some complications. They point to the idea of molecular nanotechnology, mentioned previously, and suggest that this would provide the capability to repair the damage at a cellular level, allowing the individual to be thawed and revived.

I am not going to discuss cryonics in detail here; after all, this article is examining life extension from an AI perspective, but readers will be divided into two categories: those who take the cryonics idea seriously and those who do not. Even those who take it seriously would tend to admit that there are risks of failure associated with it. If another solution could be found that could be used in addition to cryonics it could be considered a plausible insurance policy. For anyone who does not take the cryonics idea seriously, yet still wants life extension, another solution is needed. In either case we could still do with another loophole.

Using ‘mind uploading’ to avoid staying dead

The cryonics idea suggests that life extension can be achieved by preserving your brain and sending it into the future. The mind uploading idea [9, 10] goes beyond this and suggests that, perhaps, you can stay alive even after your brain is destroyed. The idea has been used quite a lot in science fiction, a good example being Greg Egan’s novel Permutation City [11], in which wealthy people use mind uploading to live in virtual reality simulations after their biological lives have ended.

The process of mind uploading would be as follows:

• A digital model of your brain would be created, using some advanced form of scanning technology.
• The digital model would be stored and used by an extremely powerful computer as the basis of a computer simulation of the future behaviour of your brain.

The suggestion here is that it is not important that your brain itself survives and that your identity should be considered as associated with the information describing your personality rather than with any particular collection of matter. In this view, if something continues to exist with the same memories and personality that you had then you can be regarded as still being alive.

There are two main philosophical issues with mind uploading:

• the strong AI issue: can computers be conscious and have minds in the same way that we can?
• the identity issue [12]: would the software model of you actually be you, or would it merely be someone else who is acting like you?

For the purposes of this article, mind uploading has an obvious weakness: neither the scanning technology to capture a digital representation of a human brain with acceptable accuracy nor the computing power to actually run a computer model derived from it are available now. If you are planning on dying in the near future, unless there is a surprising technological breakthrough, you will not have this process at your disposal.

The strong AI issue

Mind uploading relies on the idea that computers can be conscious (whatever that means) and experience things as human beings experience them. This is important: there would be little point in actually trying to continue your consciousness by using a machine that could not actually be conscious.

The term strong AI was given by John Searle, an American philosopher, to the proposition that machines can be conscious and he described it as follows:

The strong AI case would seem to imply that:

• consciousness can be viewed as existing if symbols are being manipulated in the right way by a physical device.
• consciousness can be associated with the behaviour of a system. If a system is acting in the right way, as determined by observations of its external behaviour, then it can be viewed as being conscious, or as one scientist casually put it, ‘If it walks like a duck and quacks like a duck, it is a duck.’
• if a system were conscious then a properly constructed model of that system would also be conscious, irrespective of how it was physically realised. Computers can be built from semiconductors or wooden rods and string. The strong artificial case would say that this choice of ‘matter substrate’ (the sort of physical technology that underpins the machine’s handling of information) is philosophically irrelevant.

Searle also gives the term weak AI to the position that computers are capable of modelling the behaviour of conscious entities. Weak AI differs from strong AI in that it does not regard such computers as necessarily conscious merely by virtue of them appearing to behave in the same way, to an external observer, as systems that are conscious.

There are objections to the strong AI idea, some of them made by Searle himself.

One objection, often associated with religion, is the idea that consciousness cannot be caused by arranging matter in various ways and that it involves something beyond the physical world or beyond human understanding in some sense. The most obvious form of this objection is in the claim that a machine ‘cannot have a soul’.

Roger Penrose, a British mathematician, objects to the strong AI idea by suggesting that there must exist laws of physics which have a non-computable nature and that no mere computer can simulate a system that works according to these laws [14, 15]. By taking this position Penrose also rejects the premise of weak AI. In itself, this does not automatically make mind uploading impossible: there is a lot of criticism of Penrose’s reasoning and, even if it is correct, we may still be able to copy a human mind by using a machine that exploits these non-computable laws, instead of using a computer.

Another objection to strong AI is the Chinese room argument [16, 17] proposed by Searle, who rejects strong AI. This seeks to show that consciousness and understanding cannot be regarded as existing in a system merely because it is executing an algorithm that makes it appear to behave in the right way. As with the work of Penrose, this attracts a lot of criticism. Searle accepts that human brains are machines that are capable of consciousness, but says that this does not mean that any machine that can behave in the same way as a human brain experiences consciousness. If we were to decide that Searle is correct, and I do not think that he is, there would still be a last resort available of ensuring that the copied human mind was uploaded into a suitable machine that would meet Searle’s standards, in terms of its internal workings, for being able to experience consciousness. For example, we might use molecular nanotechnology, or some technology with similar capabilities, to construct something physically similar to a human brain that is consistent with the information stored in the scanning of the original human brain.

The identity issue

Even if strong AI is a reasonable idea then it could still be suggested that a computer program that is running a copy of your mind is not really your mind, but is merely a copy that lacks the status of the original. If this pessimistic outlook were reasonable then mind uploading would be futile because you would still die and be replaced by a completely different individual who merely had a personality and memories that were similar to yours.

It is easy to see why such an objection could occur: if your brain is destroyed then anything else that is made to contain information from your brain is clearly not your brain; it is less obvious, however, that this means that it is not you. The whole issue here is one of what we should regard as constituting a person: the brain or the information contained in it.

The brain has long been regarded as the seat of consciousness by modern medicine and the brain itself is not regarded as something that can be sacrificed to help a person survive. Preserving the brain is the main function of a doctor, although other body parts may be removed or replaced to allow the brain to continue functioning. Should the brain itself really be that important to us or would it make more sense to preserve the description of a mind somehow? This really gives us two views of what is aimed for in ‘survival’: in one view trying to survive means trying to keep your brain intact and in the other view it means trying to ensure that the information in your brain survives.

Gradual uploading and the identity issue

If you are sceptical about the idea of replacing the brain, let us look at a slightly less severe process: gradual replacement of the brain. Imagine this situation: you have some sort of degenerative brain condition and your brain cells are gradually dying, starting with the ones that perform simple autonomous functions in the body and eventually involving the death of those that deal with your high level thoughts.

Each week you visit a hospital and your doctors cannot do anything to prevent the gradual destruction of your brain, but they can predict which brain cells are going to die next. They have found a way to grow replacement brain cells and they can select any one of your brain cells and grow a similar replacement which they can then painlessly implant in your brain, replacing the original cell before it dies. When each replacement cell is placed in your brain it is connected to the cells around it in the same way as the original cell and duplicates its behaviour, so that your memories and the organisation of your brain should remain intact. If your doctors wanted to install a small number of these machines in your brain to replace cells that they knew were going to die in the next few days, how would you respond to this? Would you consider it to be of any benefit? Would you regard it as having any ‘cost’ to your identity?

Let us imagine now that you continue to visit the hospital. Your doctors replace only a very small number of your brain cells in each visit, but your medical condition continues to deteriorate so that the doctors, in desperation, replace more and more of your brain. Is there any cut-off point at which you would say that the whole philosophy of the process is pointless and that so much of your brain has been replaced that you have effectively been killed?

If you think that gradual replacement of your brain’s matter would destroy you then you should already have grounds for feeling uncomfortable. Matter is routinely lost from the human body and replaced with new matter. Over a period of a few years almost all of the matter in your entire body is replaced. If you regard your identity as inextricably bound into the matter that makes you up then, quite literally, you are not the person you used to be: that person no longer exists and you are a reproduction, made from different matter into which the structure of the older version of you has been gradually copied. In a sense, this gradual replacement of matter in the human body could be considered as equivalent to mind uploading. The matter from which you are made now will not be part of your body in a few years, but there is matter in the world now that will later be part of your body. Your biochemistry relies, for preservation of you, on gradually uploading you into the food that you eat!

Let us take this brain replacement a stage further now and imagine that your doctors have developed an artificial brain cell that is actually some sort of machine that simulates a brain cell. Each artificial cell can be wired into the brain to replace a living cell and the way it interacts with other brain cells, once installed, is exactly like a ‘real’ brain cell; for example, it can synthesise chemicals that a real brain cell would release and connects up to other cells in the same way. Would you regard this treatment as much more drastic that the previous one, as far as preserving your identity is concerned? If so, why? Once more, is there any specific point at which you would say that you cease to exist during this process?

If you undergo such a process, and it is taken to its natural conclusion, then you will be left with a brain containing artificial neurons, each of which interacts, chemically and electrically, with its environment. By this stage it should be apparent that you are well on the way to actually becoming computer software. Emotionally, it may not seem like this: after all some of the ‘messy’ aspects of your brain, such as exchange of chemicals between various parts of it, have been preserved.

The process can be taken still further. Let us suppose now that the artificial neurons start to fail and that no similar replacements are available. Your doctors can, however, provide a device containing computer software that simulates a pair of neurons that are next to each other, modelling such things as exchanges of chemicals between them and transmission of electrical signals. Your brain could be replaced gradually by these units, so that the amount of real chemical activity in your brain, and the existence of any actual wiring that reflects its original wiring, slowly diminishes. Once this has occurred, the process could be repeated by gradually replacing each pair of these units by a device that uses computer software to mimic their behaviour, and so on…

Proceeding in gradual stages, you would have made the transition from being a flesh and blood human being to a computer program, but what if the computer on which you were running developed a fault? Would it be acceptable to replace various components of that machine? Would you regard being copied, suddenly or gradually, to another machine as acceptable?

I invite you to form your opinions on just how far you could go with such a process and still be you. I did not mention all this as proof that mind uploading is a viable survival strategy; I discussed it to try to show that the issue of transfer human minds into computer programs is not as philosophically clear-cut as some people might think when they casually remark that the replacement would not be you.

Using ‘indirect mind uploading’ to avoid staying dead

I use the term indirect mind uploading for a second method of attempting to preserve the mind as a computer model that does not actually rely on any sophisticated scanning technology, as opposed to the ‘conventional’ ways of mind uploading, previously discussed, that I shall refer to as direct mind uploading.

The concept of indirect mind uploading was used by a science fiction writer, Alastair Reynolds. In his novel Revelation Space [18], he referred to a computer model produced in this way as a beta-level simulation and this is how he introduced it:

In Reynolds’s story his character, living centuries in the future, has made a model of his mind, not by scanning the internal workings of his brain, but by having very powerful computers analyse recordings of his experiences and behaviour. The idea here is that the machines are given the task of producing a model of a brain based on external observations of its responses to inputs.

How indirect mind uploading could work

To explore how this would work, we will start by considering the problem of getting computers to make human-type minds. Let us imagine that at some time in the future we want a computer to provide a model of a human brain, complete with the memories, skills and personality that it contains.

Ideally, we would have a lot of information about how brains work and we would be able to provide this information to our computer. This information would not, of course, describe any individual brain and there are a lot of possible brain models that could be produced from it.

We are dealing with the set of every possible human brain, with every possible set of experiences that a brain could have. There are a lot of possible brains that we can imagine, each of which has had a different life and different experiences. Somewhere in this vast search space, in this set of possible minds, would be the brains of typical twenty-first century people. Elsewhere in this set are medieval minds and minds of humans who have memories of inhabiting all kinds of strange fantasy realms that do not actually exist, and somewhere in it is a description of your mind as it is now.

Let us first imagine that we want our computer to ‘find’ your mind in this set. We could start by instructing it to select one of these possible minds at random. This mind is unlikely to have much resemblance to yours; there are simply too many possible brains that can exist, but what if we could give the computer some information to limit its search?  Let us suppose that we decided to limit the computer to selecting only from those minds that could have been produced by living in the sort of environment that you inhabit. This would eliminate medieval minds and all those weird minds that contain memories of living on strange planets with flying dragons and so forth. What would emerge now is a model of the mind of a randomly selected fictional twenty-first century person. Of course, this would still not be very similar to yours: there are a lot of minds that we can imagine that are consistent with someone who has lived in the twenty-first century, but there would now be greater resemblance with your mind.

Let us now suppose that we took this further and started to provide very specific information about you; for example, we could use your school report card from when you were eleven years old, any letters that you have written that are available, your bank statements and so forth. If you have kept a diary this would be excellent. We could now give our computer these instructions:

We have this information about a specific human and this information about how human brains work. Search the set of possible human brains and ‘find’ us a brain, subject to the constraint that it must be consistent with the experiences and behaviour for the specific human that we are considering.

At the end of this search the computer would provide a model of a fictional brain that could exist and we could use a computer to run a simulation based on this model. This brain would now have much more similarity with yours: it would be one of the possible brains that a human could have if he had lived in the twenty-first century, received the same school report when he was eleven years old, had the same spending habits and wrote exactly the same diary. I suggest that the mind described by this model, while it would not be exactly the same as yours, would still have much in common with it.

If you accept mind uploading as a means of life extension then you have already accepted that you can regard yourself as continuing to exist if something sufficiently like you continues to exist. If this is the case then there are good philosophical grounds for accepting the model obtained by indirect mind uploading as being you, at least to some degree, as it would have some resemblance to your mind and the model of your mind that would have been generated by scanning it in direct mind uploading.

In principle, if we observed the external behaviour of a mind for long enough, and used the observations to generate a computer model, then that model would be equivalent to the model that would have resulted from direct (that is to say,  ‘conventional’) mind uploading anyway, making any discussion about the validity of the model being dependent upon its origins pointless. Given this, it seems reasonable to think that a person who accepts mind uploading as being a valid way of extending his/her life should also think that making a model based on observation of a person is also desirable, the desirability increasing as the detail of the observations on which the model is based increases.

The implication of this is obvious: if we could make a large number of observations of a human and then use these to create a plausible computer model of that person’s mind, and if the detail with which the person was observed was high enough, the computer model created in this way may be close enough to the real person’s mind that we may actually consider his/her identity to have been copied into the computer model to a significant degree. This seems to suggest that indirect mind uploading, in which we do not need sophisticated means of scanning the brain’s structure, but instead base the modelling on large numbers of external observations, may actually be seen by a rational human as being a viable way of keeping his/her personal identity in existence after the destruction of his/her biological brain.

We can now consider how to get the massive number of observations needed to cause a very high similarity between the model that results from indirect mind uploading and the original biological brain on which it was based.

So far, I have suggested that any documents relating to your life would help in a ‘reconstruction’ and a personal diary would make a huge difference by eliminating large numbers of possible minds that could not plausibly have written that diary. We now need to add more observations. Each time we add any new data to our collection large numbers of ‘possible models of your mind’ are eliminated and we get closer to our goal of creating a computer model that resembles, as closely as possible, your mind at the time when this gathering of data ended. For this reason, we should try to use methods that create a large amount of data.

Audio recording of your activities and experiences, particularly conversations in which you take part, would make a significant difference, so carrying some sort of audio recording device around and using it as often as possible would seem to be useful. Video recording of as much of your life as possible would be even better, perhaps by wearing a small video camera which can record events in approximately your field of view and obtain a reasonable sort of record of the sorts of things that are happening to you.

How accurate would indirect mind uploading be?

Audio and video recording alone would seem to demand that any computer model built, based on that information, resemble you to a very significant degree. This may sound unreasonable to some people who might say, ‘That cannot be me. Even if it remembers saying everything that I ever said, it has not actually been made by looking inside my head. It will be nothing like me.’ To such people I would say this:

Imagine someone else existing who has memories and a mind consistent with having experienced situations similar to yours and having made the same sorts of decisions in response to them. How similar would that person be to you? Now imagine this person having a mind consistent with having said everything that you said over a number of years, not just with regard to the words spoken, but even with regard to their inflexion and the duration of pauses between them. Furthermore, imagine this other person’s mind being consistent with having made the same gestures that you have made at various times to a high degree of accuracy, possibly even moving the same way while sleeping. How similar would this person’s mind be to yours? Surely, this other person would have a very high degree of similarity with you, possibly enough so that any process that can create such a mind can be considered to be creating a continuation of your mind.

Using indirect mind uploading now

Creating a copy of a human mind in this way would require much more powerful computers than we actually have now; indeed, in Reynolds’s book, his character attempts his indirect mind uploading process centuries in the future.

There is, however, one way in which it may actually be possible for an individual who is alive now to use this technique. The idea is certainly not new: the idea of reconstructing a human mind from archived data about a human life appears to have occurred to a number of people, including Timothy Leary (1920–1996CE), a psychologist who is probably better known for advocating the use of hallucinogenic drugs.

The process would have two stages: recording of the data and using computers to generate the model from the data. It may be possible to record the data today, which, after all, merely involves making such things as audio and video recordings that could be captured by conventional equipment, possibly with some modifications to enhance portability and convenience, and then storing the data for a period of time, during which computing power will increase to such an extent that it will be able to use the data to construct a mind from it. The hope would be that the data recorded today could be used to make a mind in the future, effectively achieving what may be regarded, to some extent, as an uploading of the mind of a person living today, without any technology beyond our own being required until centuries after biological death.

The sort of increase in computing power that would be required to do this is one that we can expect to arrive at some time in the future if progress in computing science continues in any way like it has done so far. In 1965 CE> Gordon Moore published what is now widely regarded as a classic paper [19] observing that the number of transistors on a microchip doubled every twelve months. This is known as Moore’s law. If Moore’s law continues to hold even approximately true then we can expect this exponential increase in computing power to deliver massive computing resources in the foreseeable future.

Pessimists could comment that there appear to be fundamental limits on semiconductor microchip fabrication. Microchips are made by projecting a photo-reduced image onto the surface of the chip and as smaller transistors are required the wavelength of the radiation required to do this decreases, making the radiation harder to focus. If progress with existing manufacturing methods continues for long enough we will eventually find ourselves with the problem of trying to focus X-rays. No viable method is known for focusing X-rays, nor is one likely to be available soon; this problem is known informally as the wall. Of course, progress might not even continue as far as this: the difficulty and costs could become prohibitive well before X-rays are needed. This, in itself, need not mean that improvement in computing hardware slows down for any appreciable period of time; such a problem will drive research into new ways of building computers and when the transfer to new forms of technology is made then progress could actually occur much more rapidly than is predicted by Moore’s law.

An obvious question here is: how much information would you need to preserve about yourself for indirect mind uploading to be viable? There is no easy answer; every item of data that is stored eliminates many hypothetical versions of you that would not match those criteria and ‘as much as possible’ would seem to be the rule. A diary would be highly advisable, as would copies of any emails that you send and audio recordings. Video recording, both from roughly your own point of view and, occasionally, from cameras mounted in various places, if the equipment could be afforded, would be even better.

A person who was ambitious about getting ‘properly’ uploaded could go even further by incorporating pulse readings and similar sorts of measurements. A DNA sample would probably help to eliminate minds that could not have resulted from your particular genetic code and preserving DNA for long periods of time is now quite cheap. Although current brain scans do not capture enough information to allow a sufficiently accurate computer model of the mind to be made, this does not necessarily mean that brain scans would be useless: if you really wanted to spend money, the data from occasional brain scans, stored with the other data about your life, might be significant in eliminating large numbers of possible minds that could not possibly have had those particular brain scan results in their pasts.

How the model could be generated

We will now look at how the archived data, documenting a person’s experiences and actions, might be used to construct a computer model. This discussion will be necessarily vague, because we have incomplete knowledge of how to make models. Generating models from observations is major problem in artificial intelligence, but our own brains are quite good at it, which suggests that solutions to modelling problems are at least possible. Nevertheless, it is possible to speculate now about how it might be done, and a possible method is as follows:

The process has to generate a model of the brain and, to a limited extent, the world in which the brain existed. This is because the data that is archived will only be partial. There will gaps in the recordings, especially before the recording of data actually started. The data is also of limited resolution. As examples: video recordings will not indicate exactly how the each individual receptor in your eye was stimulated by light and audio recordings will not indicate the nerve impulses that your brain transmitted to speak words.

The first step could be to create a physics model. This would allow another process to use it to perform general simulations of reality. The physics model would be constructed by using what is known about how things behave in reality at a general level. It would contain various assumptions about basic things such as the laws of nature.

The next step would be to create a general world model. This would use the physics model to simulate the sorts of things that are encountered by humans in the world and it would make use of the physics model to achieve this. It would also contain assumptions, as the physics model does, but these would be at a higher, more abstract level.

Following this, a time-specific world model would be made. This would use the general world model and would enable a higher-level process to use it to simulate the world at the specific period of history in which the person (whose recorded data we are analysing) actually lived. The time-specific world model would not contain specific details about the person’s experiences; it would simply be derived from what is known about the world during that period of history.

As well as simulating the environment in which the person’s brain existed, we also need to simulate the brain itself; in fact, this is the main objective. To do this there would be a general brain model that could be used by a higher-level process to simulate a human brain. This model would not contain information about any particular human brain; rather, it would be made it by using what is known about human brains in general.

The ultimate goal of the process is to simulate a human mind and its environment. To this end there would be a person-specific brain model that can simulate a specific human brain. The person-specific brain model would contain assumptions about the specific brain that it is simulating and it would also make use of the general brain model. There would also be a person-specific world model, which would simulate the world as experienced by a particular human being. The person-specific world model would contain some assumptions about the environment inhabited by a particular person and the sorts of things that can happen in it and it would make use of the time-specific world model.

The person-specific brain model and the person-specific world model would interact with each other. The person-specific brain model would produce, as outputs, the nerve impulses that the individual may use to control his body and these would be transmitted to the person-specific world model as an indication of the actions of the individual in his/her environment. The person-specific world model would, in turn, provide outputs in the form of the sensory nerve signals that would be received by the person’s brain in that environment and these would be sent to the person-specific brain model. In this way, the two models would work together to simulate the thought processes and behaviour of a human, as well as the events that happened in his/her environment.

Some of the simpler parts of this system can be made by using knowledge that is already available about the world, but the person-specific brain model and the person-specific world model are not as easy to make. For the process to have any point at all they need to be consistent with the data that was recorded about the individual’s life. This means that they need to be made by analysing this recorded data.

It is possible to state a criterion that the person-specific world and brain models must satisfy. Just as recorded data was obtained about the original person’s life, it would be possible, by running these models, and also using the other modelling systems mentioned above, to generate the recorded data that would have resulted if these models had accurately described the person. The criteria that needs to be satisfied is simple: both the original archives from which we are working (made by recording the original person’s biological life) and these archives obtained from running the model must be the same for a given combination of person-specific brain and world models to be valid.

To put the above idea more simply: if we have a possible model describing someone’s life and experiences, and we run that model so that it relives the experiences of that person, then what that model describes the person doing and experiencing must match with the information we already have about what the original person did and experienced.

This criterion would allow a lot of possible models to be eliminated, but there are two problems:

• An exhaustive search of all possible models until we happened to find one that meets this criterion would be impractical. It would take too long.
• There will actually be an infinite number of possible models that meet this criterion: for any model that we can find that meets it we will always be able to produce a number of more complicated variations of that model that also meet it.

How do we deal with these problems? The first one must be dealt with by using a search method that does not involve generating and checking every possible model. The second problem needs a method of selecting the most plausible model from all of the possible models.

One way that this could be achieved is to start by expressing the model in very vague terms, only assigning very general qualities to it. If the degree of generalisation is adequate then there would be a limited number of such models available and all of them could be checked. There is no point checking each such model to see if it satisfies the criterion of consistency with the recorded data: it is an incomplete model and cannot be expected to do this. For each partial model, we should, however, make an estimate of the number of complete models that could be made, which have the same general characteristics of the partial model that we are testing and that would be consistent with the recorded data. When we have tested all the possible ways of setting up general characteristics of the model we would select the one that we estimate to have the largest number of complete models with the same characteristics that are consistent with the recorded data. We would then continue the process, by adding more characteristics to this partial model, testing at each stage for the number of complete, consistent models that are available with the characteristics that we are trying.

Some readers will recognise such a computation as equivalent to Occam’s razor, the philosophical idea that scientific models should contain as little as possible. It would be not be a trivial computation to perform: it would require more information than we currently have about model generation to do it really well and would merit an article in itself. However, human brains can construct sophisticated models of reality so we have reason to believe that such sophisticated modelling processes can be performed.

This process would continue, with the person-specific brain and world models being selectively refined until they reach an acceptable standard.  The brain model would finally be run through the simulation process (as would have happened many times during the testing process) so that it reaches the state that it has when it is modelling the original brain at the time when the recording ended (which, if the person whose archives these are has been very thorough, will actually be the time at which biological life ended). The interface between the brain model and the world model would now be removed and the brain model would have its inputs and outputs connected to either the real world or a virtual reality simulation. It would then continue modelling the brain’s response to this environment past the point at which the recording ended. The brain model is now being used to perform a simulation of the original person’s behaviour after the time at which he/she actually died. If it is reasonable to view the computer model as being conscious and to view the similarity between the computer model and the original person as giving it some of the original person’s identity then it would make sense to say that the original biological person has had his mind preserved and is still experiencing things after his/her biological death. The goal of using AI to avoid staying dead would therefore have been satisfied.

As an aside here, it will also occur to some readers that a process of this nature is likely to involve running a very large number of models of various versions of you in various simulated worlds for test purposes; if you are considering actually making archives of your life at some point in the future then an extremely paranoid speculation, and one that I find amusing, could occur to you now about your current status.

Confidentiality of information

Having all this information recorded about you could cause problems. There would be a lot of personal and confidential details and having large periods of your life stored so that people could access that data today could have unpleasant consequences. On the other hand, we would need to make sure that whoever wants to reconstruct your mind in the future could get at the data. Fortunately, there is a simple solution.

Encryption techniques are well developed. These rely on having a decryption key to decode the data. The decryption key is made long enough so that repeated attempts to guess it are not likely to succeed in any reasonable period of time.

Any attempt to use this method now would rely on much faster computers existing in the future and they would be able to solve some encryption problems that are practically unsolvable for current computers. All that is necessary is to encrypt the data in such a way that the decryption key is too long to make decryption feasible in the near future, but not so long that it would be difficult to break for someone in the future who has computers powerful enough to model human minds. The decryption key would not be stored anywhere, making the encryption process effectively one way in the absence of computers of adequate power; in fact, decryption of the data need not even be possible for the person who set this up, as he would not need to know a decryption key.

Encryption would be used in this way as a sort of time lock, to protect secrets until computing capability has reached a certain level in the future. Security of the information could therefore be assured, at least in the short term, quite easily.

The idea of using encryption as a time lock has already been proposed by Ronald L. Rivest, Adi Shamir and David Wagner [20], who also state that this issue has been previously discussed by Timothy C. May. This work has a slightly different focus than the one being used here. Rivest, Shamir and Wagner proposed encrypting information so that a long period of computation would be needed before it could be decrypted, so that approximately the required time would have elapsed when the computation was complete; their method proposes compensating for expected improvements in processing rates. They also proposed encrypting in such a way as to make the required decryption process highly sequential so that little, if any, reduction in the time for a solution could be achieved by using parallel processing.

In the way that it is being used here, an encryption time lock would not be designed with compensation for future increases in processing ability: it would be designed to explicitly require those future processing rates to perform a decryption in any reasonable time, so the emphasis would be slightly different. It may not be necessary to force the decryption process to be sequential, because, whereas in the scheme of Rivest, Shamir and Wagner, someone who has got his hands on a massive number of parallel processors may be about to try to cheat and decrypt the data too soon, in this scheme such a person may have just the sort of processing needed to recreate a human mind; we are more concerned with how much processing power people in a future society have than when they do the decryption; older cryptographic systems, such as those devised by Ralph Merkle [21], may be more suitable for this purpose.

The privacy of people you meet is another issue entirely. If you are recording your own life in minute detail then you will incidentally be recording a lot of details about other people who interact with you. Some of these people may not like their personal conversations and habits being sent into the future. It may be difficult to negotiate some sort of agreement with all these people; it would be a lot easier not to tell them of course.

Motivation

There is some element of chance in a process of this nature. Even if the philosophy and technological ideas are sound, you still have to rely on someone in the future wanting to go to the trouble of recreating you. It would seem reasonable to assume, however, that, if this sort of process is possible at all, the computing capability needed to perform it will be inexpensive one day and, of course, being conventionally cremated or buried without making any attempt to preserve your brain or mind is a fairly dangerous process if you plan on coming back.

Another concern may be relevant: what about the possibility of your mind being reconstructed by a future society for motives that are not in your best interests? A future society could potentially reconstruct you as an altruistic gesture or may do it merely to demonstrate technical capability or to use the software model for entertainment, education or historical research. On the other hand, these possibilities may not be too bad; being reconstructed by a society of trans-human ultra-sadists with an interest in twenty-first century history would be bad.

As a word of warning, if any readers are engaged in war crimes or other such deeds and are thinking of trying this method, it could be a serious error of judgement: a software version of you, created in the distant future, could conceivably find that it has been resurrected merely to answer extremely serious charges…

Indirect mind uploading versus other methods

Indirect mind uploading may appear to be a competitor to cryonic suspension. Both have the same feature: the preservation of something, whether it is a human brain or information, with the hope that it can be used by a future society to continue your life.

There is not really any such competition. If an individual decides that the chance of revival from cryonic suspension is too low to justify using any resources on it, then he/she can make this decision independently of any consideration of mind uploading; however if an individual decides that cryonic suspension is worthwhile he can pursue this, also independently of any decision he makes about indirect mind uploading.

An individual who has signed up for cryonic suspension may also create the archives needed for indirect mind uploading for another reason. Such archives may allow indirect uploading to be attempted if the cryonic method fails, but there is a possible extra use for them: if the cryonic method is only partially successful and allows a revival with substantial loss of information about the brain’s structure then it is possible that the computer model generated by indirect mind uploading could be used as a reference for making adjustments to such things as the strengths of connections between neurons in the brain. In this way, indirect mind uploading may complement cryonic suspension by increasing the accuracy and reliability of cryonic preservation of a human personality.

There is also no conflict between indirect uploading and direct uploading (in which the brain is actually scanned). If direct mind uploading were available, with a sufficiently safe and reliable method, then it would be used in preference to indirect mind uploading. Data is only likely to be recorded for indirect mind uploading while direct mind uploading is not available.

Editing your own archives

There is one idea that would occur to almost anyone who was attempting indirect uploading of him/herself: would it be possible, by editing your archives, or selectively withholding information from them, to modify the reconstructed version of yourself in a desirable way?

Let us say, for example, that you would like to be a nicer person. Maybe you could alter the information in your archives so that you appear to help friends more, make a lot of donations to charity and so forth? It would be easy to edit diaries in this way. If you are making audio-visual recordings of your life it may be more difficult, but not impossible. The intention would be that the reconstructed version of you would be one that has led your life, except just a little more generously. As another example, let us imagine that you are committing a small crime on a regular basis, for example stealing small items of stationery from a place where you work, and that you would rather not be, after reconstruction, the person who had done this. It would be a relatively simple matter to ensure that any recording devices were turned off while performing such acts. You would also have to make sure that you did not mention these things when they were turned back on, of course! This could actually be a bad idea. If you are reconstructed then it is very likely going to be by making the ‘most plausible’ model. Getting this to be as close as possible to you is the main goal and interfering may not give the results that you want.

For a start, there is a clear problem of interpretation: if you use recording devices and turn them off at 10:30pm on every Monday evening then the information that you turn off the recording devices every Monday is in itself data and could be considered as providing some of the constraints that the history of your ‘new’ mind must satisfy, even without the data being there itself. If your recording is really good then your hand may even be video recorded pressing the ‘off’ switch every Monday night. Even if this information is not recorded, the gaps created in your archives could be used to infer that you acted to prevent recordings at various times, unless the computer which was reconstructing you was specifically instructed to disregard this.

Let us suppose that the computer that reconstructs you is programmed to disregard any gaps in the recording. It would have to take account of them, but it could fill them in with the most plausible memories that could be constructed from the other data that it has, just as it would have to do so for any memories of what you experienced before you actually started archiving your life. This may still not be an advisable action. The degree to which the loss of that information affects the accuracy of the reconstruction would depend on the importance of it in terms of evidence that it provides about your mind and you may omit something trivial that turns out to be very important. This could be worse than the unwanted gaps that will occur due to starting the archiving partway through life, batteries running out and so forth, because it introduces a consistency to the way in which information is withheld from the archives. Normally, if a particular experience were omitted from the archives there would probably be extra evidence of it somewhere else, or similar sorts of experiences that could affect the model in similar ways. Deliberately withholding certain types of information, however, could have a much more dramatic effect on the accuracy of the reconstruction.

Courses of action

What you should decide to do, when thinking about the possibility of indirect mind uploading depends on the philosophical opinions that you have:

• If you have no aversion to being dead, then none of this matters.
• If you think that a computer can never be conscious in the same way that you are and you also reject the idea that the information from your archives could be placed in any conscious machine, then you should view mind uploading of any type (direct or indirect) as futile, at least with regard to preserving your mind.
• If you think that computer consciousness is possible, or that some sort of conscious machine could be produced, but that a model of your mind, complete with your memories, would not be a continuation of your identity, then you should again view any sort of mind uploading as futile.
• You may find the concept of mind uploading as a way of preserving a human identity to be reasonable, but you may still find indirect uploading futile if you think that the differences between the software model and your own mind should preclude viewing it as a continuation of your own identity.
• If you find the concept of indirect mind uploading to be workable, then this does not mean that you should automatically invest resources in actually archiving your life: there are still the issues of motivation in the future to be considered. Furthermore, if you decide that archiving for indirect mind uploading is reasonable and desirable this does not mean that you should rely on it as the only way of continuing your identity.

Conclusion

The idea has been discussed that recording a lot of information about your life and storing this information so that it is available in the future, after your biological death, could allow a software model of your mind to be constructed by using very powerful computers available in the future. Although such a computer model would not resemble your mind in every detail it would still be very similar to your mind and there may be good grounds, if this happens, for regarding yourself as still being alive.

This method has a distinct advantage over other suggestions for copying minds: it requires no technological advances to capture the data and this could be done now, so that the data could be used in the future.

Philosophical issues do arise. The two most important issues are the ones that are relevant in discussions of direct uploading, that is to say the more ‘conventional’ method proposed for copying human minds:

• the strong AI issue: can computers be conscious and have minds in the same way that we can?
• the identity issue: would the software model of you actually be you, or would it merely be someone else who is acting like you?

References

[1] Drexler, K.E. (1986). Engines of Creation. Published by Anchor Books.

[2] Drexler, K.E., Peterson, C., Pergamit, G. (2000). Unbounding the Future: the Nanotechnology Revolution. New York: William Morrow.

[3] Drexler, K.E. (1992). Nanosystems. New York: John Wiley and Sons Inc.

[4] The Foresight Institute. (1986-2002). Retrieved June 22, 2003 from http://www.foresight.org/

[5] Zyvex. (n.d.) Retrieved June 22, 2003 from http://www.zyvex.com

[6] Ettinger, R.C.W. (1964). The Prospect of Immortality. New York: Doubleday.

[7] Cooper, E. (writing as During, N.). (1962). Immortality: Physically, Scientifically, Now. Privately published.

[8] Alcor Life Extension Foundation. Retrieved June 22, 2003 from www.alcor.org/

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

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

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

[12] Chislenko, A. (n.d.). Drifting Identities. Retrieved June 22, 2003 from http://www.aleph.se/Trans/Cultural/Philosophy/identity.html

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

[14] Penrose, R. (1989). The Emperor’s New Mind: Concerning Computers, Minds and the Laws of Physics. Oxford: Oxford University Press.

[15] Penrose, R. (1994). Shadows of the Mind: A Search for the Missing Science of Consciousness. Oxford: Oxford University Press.

[16] Searle, J. R. (1997). The Mystery of Consciousness. New York: The New York Review of Books.

[17] Hauser, L. (2001). The Chinese Room Argument. The Internet Encyclopedia of Philosophy. Retrieved June 22, 2003 from http://www.utm.edu/research/iep/c/chineser.htm

[18] Reynolds, A. (2000). Revelation Space. London: Victor Gollancz. Chapter 7. (Fiction).

[19] Moore, G. E. (1965). Cramming more components onto integrated circuits.
Electronics, Volume 38, Number 8. Retrieved June 22, 2003 from ftp://download.intel.com/research/silicon/moorespaper.pdf

[20] Rivest, R. L., Shamir, A., Wagner, D.A. (1996). Time-lock puzzles and timed release crypto. Retrieved June 22, 2003 from http://secinf.net/cryptography/Timelock_puzzles_and_timedrelease_Crypto.html

[21] Merkle, R. C. (1978). Secure communications over insecure channels. Communications of the ACM 291:294-299, April 1978. Retrieved June 22, 2003 from http://www.itas.fzk.de/mahp/weber/merkle.htm