Mind uploading

Old - initially written on Transhumanity in 2004. The last part is dated.

Mind uploading

Copying a human mind with personality and memories to an appropriate intermediate support, and later loading the copy on a biological or electronic support different from the original brain, is still far from our reach. At the same time this ambitious proposal for a future technology, frequently called “uploading”, has the highest conceivable practical importance: uploading means effective immortality.

Why? Because then we can make multiple backup copies and, should anything fatal happen to a person, we can load the most recent backup. While the endless possibilities that follow are explored in some detail in the excellent science fiction references, the question asked here is what options may become available within our lifetimes.

The reconstruction of a human mind, from the data acquired by scanning the original brain, on some other computational support, requires a very detailed understanding of how the physics of the brain generates the conscious mind. Some hold the opinion that even this is not enough, and that some other phenomena not derived from the physics of the brain must play an important role in explaining the mind. If, of course, mind can be explained at all. In this note I will make the assumption that a “sufficiently complete” brain scan, followed by a “sufficiently accurate” reconstruction of the patterns encoded by the data in the scan, performed following appropriate physical models of the brain and target systems, permits re-instantiating the mind in its state immediately before the scan. I make this assumption because, in my opinion, it is the simplest assumption compatible with what is known today.

In other words, I am assuming that uploading is possible in principle. Is it feasible in practice? How long will uploading technology take to develop? Can we look forward to be uploaded, or is it rather for our grand grandchildrens?

In the recent novel “Down and Out in Magic Kingdom” by Cory Doctorow, situated at the end of this century, uploading and reconstruction are described matter-of-factly as routine medical procedures. While I strongly recommend reading this delightful novel, it may leave the reader with the impression that uploading is just behind the corner. The more sobering view held by most experts is that the technologies involved are so complex that uploading a human mind is more than 30 years away: 50 years are often given as a more realistic estimate even by “true believers”.

At the same time, the technology for the acquisition of the input brain scan, sufficiently complete to permit a future reconstruction as outlined above, might be available much sooner. Suppose for example that at some point, say 20 years from now, we have developed the capability to map a physical brain with the required accuracy, without having developed yet the capability to reconstrut the original person. This is a reasonable assumption, and indeed this is the current status of cryonics: we (believe that we) know how to properly freeze a recently dead person, but we do not know yet how to revive her. Cryonics is an expression of faith in future scientific advances. Then we may consider scanning a brain and storing the map as a viable self preservation procedure, equivalent to cryonics: scan now, store one or (better) multiple copies, reconstruct as soon as permitted by scientific and technical progress.

This sounds like a reasonable alternative to cryonics, at least to me. I am in my forties and I do not really hope that an operational end-to-end uploading technology is developed within my natural lifetime. At the same time I am sure that appropriate physical models of the brain and target systems, and consequently the technology to map one onto the other, will be developed sooner of later. So I will use the remaining part of this note to speculate on when the capability to acquire and store a “sufficiently complete” brain scan could be developed. Of course, the first key question is what “sufficiently complete” means, and nobody can answer that yet. Other key questions are, what technology can be used to acquire the scan, how long it takes, how much storage capacity is required, and, of course, how much money it costs.

I want to make an important assumption explicitely. I will assume that all properties of the brain/mind system can be completely understood in terms of macroscopic (non quantum) physics, or in other words that we can safely forget quantum effects. I believe this is the simplest assumption compatible with our objective knowledge at this time. Now there are theories that, on the contrary, quantum effects are absolutely vital to whatever the physical brain does to produce (or interact with) consciousness. If these quantum consciousness theories are validated by experiment, they are bound to have a profound impact on the engineering feasibility of mind uploading.

Back to the macroscopic viewpoint, let’s try to establish some required upper and lower resolution figures for a sufficiently complete brain map. Some experts believe that, since we have to know where all relevant molecular structures are and what they are doing, a resolution of 1 nanometer, with occasional sampling at a higher resolution, could very well be required. Remember that it takes one million nanometers to make a millimeter. Other experts believe that a resolution of 10 nanometers may be sufficient.

Leaving aside for the time being the critical issue of how to acquire a volume map of a 3D object at this very high resolution, I wish to note that this is a huge amount of information. Assuming 10 cm as a typical linear dimension of the brain, and that 1 byte (8 bits) is sufficient to encode the information in each volume element, we have to acquire and store 10**24 bytes of information, or 60.000.000.000.000.000 CDROMS!!! This pile of CDROMS would extend much beyond Pluto’s orbit in interstellar space. While this is a good visual illustration of the sheer volume of information contained in a human brain, the CDROM is definitely not the last word in information storage density. Indeed, there are indications that a density of ten terabytes (10**13 bytes) per square cm may be attained soon. Then, instead of a pile of CDROMS reaching into interstellar space we have a surface of 10 square km that, after some more development in storage density and three-dimansional packaging, may well go further down to, say, a large building. In the rest of this note, I will assume that storing the amount of information contained in a human brain will be feasible in ten years and operational in twenty years.

Once the brain and self are better understood, data compression may permit dramatically reducing storage requirements. All digital video aficionados know than digital movies require less and less storage space: modern MPEG 4 video compression technology permits squeezing a full length movie with several audio tracks, with a quality comparable to that of a DVD, on a single CDROM. Data compression works by discarding information deemed not relevant for a given application. This is how MP3 audio compression works: it discards information that the human ear would not be able to hear anyway, such as frequencies above or below a treshold. Analogously, the bits corresponding to small variations in hue that the human eye is not able to appreciate are discarded in image compression. I think that perhaps a complete model of the interplay between brain and consciousness, when it is established, will show that much of the information encoded in the physical brain ("mind file") is not strongly coupled to consciousness and can be discarded while still preserving the essential information. In fact, estimates (e.g. Moravec and Tipler) for the information content of a human mind range between 10**13 and 10**17 bits, much lower than the 10**24 bits obtained by “brute force” mapping of the physical brain.

Also, it may be possible to replace specifc subsystems, like for example the ability to speak a foreign language, with standard “modules” available off-the-shelf. To make it clear: I am proud to speak a fair number of foreign anguages, but I do not consider it as something central to my identity. If I have no other choice, I would accept being revived from a brain scan without the bits associated a language, and later grafting this skill from a standard module. I think it would not be exactly the same thing, but perhaps it would be good enough, and would not strongly impact on “being me”, at least if it is not a language that I use frequently (the ability to speak a mother language or any language that one is very familiar with may well be too strongly coupled to personal identity). Something similar may probably be said for most mind subsystems associated with bodily functions and motor skills: an importat part of the information hardcoded in our brains may be dedicated to managing our body and, assuming that this information is not strongly coupled to the sense of self, it may be possible to replace it with standard subsystems. Please, install the very latest release of “Squash Master” before waking me up.

Moreover, we all know that an old and scratched picture can be turned into an image that might have been taken by a modern camera by using image processing techniques plus inference and rules: if we cannot see if a person portrayed in the old and scratched picture has a wedding ring on the finger covered by a scratch, we can decide to paste a wedding ring there if we know that the person was married, and with high probability the choice is right and the processed image will be more faithful to the original scene. In summary, when the brain is better understood it may be possible to use this knowledge to compress mind files. Our understanding of the brain is advancing rapidly, also as a result of Artificial Intelligence (AI) research.

After the outline of storage requirements, and in view of the rapid progress of AI and brain sciences, I am now satisfied that the technology needed to reliably store a mind file may well become available within my natural lifetime. Now I wish to evaluate the feasibility of generating a mind file by scanning a physical brain.

Most proposed methods require technologies that have not been developed yet. For example when operational nanotechnology is available, it will be possible to send swarms of nanomachines in a brain to copy the detailed status of relevant cellular and molecular structures such as neurons, neurotransmitters and synapses. The information copied can be later retrieved from the nanomachines with an appropriate method. The process may be performed on a recently dead brain (similarly to current cryonics, and this is the only feasible approach if the copy process is destructive) or on a living brain. In this case perhaps the nanomachines may take up permanent residence in the brain and copy/paste the information continuously. Alternatively, the “Introdus nanoware” in Greg Egan’s “Diaspora” destroys the subject’s brain in the nanomachine based copying process.

The problem is, we do not know when advanced nanotechnology will be developed and deployed operationally. We are now only approaching the capability to develop very crude nanotechnology demonstrators, and the time it may take to develop mature applications such as this is anyone’s guess. While most experts assume that it will take from thirty to fifty years, it may take less, or it may take much more. Having set my own natural lifetime as “useful” timeframe, I prefer to speculate on the development of today’s operational technologies.

Magnetic Resonance Imaging (MRI) technology is used in hospitals to acquire three-dimansional images of living brains, reaching a resolution of 1 mm that permits, in some cases, detecting early-stage brain tumors or other neural disordes. At this time, MRI is many orders of magnitude away from the submolecular resolution that may be required. Might MRI technology progress to the point where it is able to achieve this resolution (of the order of 1 nanometer, one million times smaller than a mm)?

MRI is based on the principles of nuclear magnetic resonance (NMR) and works by submitting a brain, or whatever living tissue, to a strong magnetic field. The magnetic resonance signal emitted by hydrogen nuclei in the magnetic field produces the MRI signal, that can be processed to generate two or three dimensional images. The image resolution that can be obtained from the MRI signal depends on the gradient of the applied magnetic field, that is, on how much it changes from place to place.

MRI can be performed on a living brain and, indeed, has well known clinical applications in neurology and brain oncology. It can, however, be performed at a much higher resolution on an appropriately prepared dead brain. This is due to the fact that generating magnetic fields with the required gradient is much easier if the field is confined to short distances (for example, a magnetic field confined between two plates separated by a very short distance). Therefore the highest resolution applications of MRI are performed on very small samples. For brain imaging, the sample can be reduced to the required size by cutting the brain in small cubes or very thin slices (taking care of not destroying too much tissue in the cutting or slicing process). Of course, this can only be done to a dead frozen brain.

Assuming that this continues to be true as MRI technology evolves, the first viable applications of MRI to mind uploading shall be performed on recently dead corpses (as is the case for today’s cryonic procedures), and destroy the corpse’s brain in the process (in other words, if they screw up you are irreversibly dead). This is the technique used in the novel “Eater” by Gregory Benford, where a sentient black hole visiting our galactic neighborhood tries to bully the world into sending it uploaded humans, and recommends this procedure as the only one compatible with the current status of our technology.

So how fast is MRI technology developing? Resoutions of the order of 1 micrometer (one thousandth of a millimeter) in all three directions have been recently demonstrated by experiments done on vary small tissue samples, cells, and single cell organisms (ref. 4). 1 micrometer is “only” 1000 times bigger than the required 1 nanometer, so it seems reasonable to think that soon, say within the next 30 years, uploading level resolutions might have been demonstrated in the laboratory on small tissue samples. This would already permit reliably copying the information stored in a dead frozen brain. Establishing the required extended magnetic fields and engineering a suitable data acquisition system to permit copying the information stored in a living brain is a challenge of the same magnitude, that will also require important scientific and technological breakthroughs.

The conclusions? Well on the basis of the available hard facts (few), expert opinions (some), less expert opinions (plenty), and a lot of personal guesswork, I believe that if I manage to stay alive and mentally fit until well into my eighties or nineties (personal discipline and medical advances required) I may have a reasonable chance of having my mind scanned, stored, loaded on a better support system after a few decades or centuries, and then living forever. Perhaps multiple copies of me will explore the stars and merge their memories every few thousands of years. I prefer not to start speculating on the legal, societal and political adaptations that a world where uploading is a routine procedure would require, since things would become really complex then.

Perhaps I am too pessimist on the development rate of nanometric resolution MRI and fully operational nanotechnology, which may materialize sooner than expected. As Kurzweil points out we have already entered an era of exponential growth and while we usually overestimate developments in the short term (a few years), we tend to underestimate developments in the medium term (a few decades). Perhaps by the time when the first crude assemblers are developed we may already have operational AI, smarter than humans, to rapidly push to advanced nanotechnology.

Or perhaps I am too optimist, and things are going to take much longer than expected. Back in the sixties we all thought that we would have cities on Mars in the 21st century, and they just did not materialize. So while I hope that advanced technologies for uploading may develop sufficiently fast, I would not bet too much on it. For the time being “conventional” cryonics looks like the safest way to maximize our chances of indefinite survival, by transporting revivable physical bodies a few decades or centuries in the future. 

Posted by G.P. on 11/30 at 10:08 AM
transceng • (0) Comments • Permalink • Print
Page 1 of 1 pages