We Seek Not To Become Machines, But To Keep Up With Them
Franco Cortese
2013-06-21 00:00:00

Substrate Independence 101:





Substrate Independence

The term substrate-independence denotes the philosophical thesis of Functionalism – that what is important about the mind and its constitutive sub-systems and processes is their relative function. If such a function could be recreated using an alternate series if component parts of procedural steps, or can be recreated on another substrate entirely, the philosophical thesis of Functionalism holds that it should be the same as the original, experientially speaking.

However, one rather common and ready-at-hand misinterpretation stemming from the term “Substrate Independence” is the notion that we as personal selves could arbitrarily jump from mental substrate to mental substrate, since mind is software and software can be run on various general purpose machines. The most common form of this notion is exemplified by scenarios laid out in various Greg Egan novels and stories, wherein a given person sends their mind encoded as a wireless signal to some distant receiver, to be reinstantiated upon arrival.

The term substrate independent minds should denote substrate independence for the minds in general, again, the philosophical thesis of functionalism, and not this second, illegitimate notion. In order to send oneself as such a signal, one would have to put all the processes constituting the mind “on pause” – that is, all causal interaction and thus causal continuity between the software components and processes instantiating our self would be halted while the software was encoded as a signal, transmitted and subsequently decoded. We could expect this to be equivalent to temporary brain death or to destructive uploading without any sort of gradual replacement, integration or transfer procedure. Each of these scenarios incurs the ceasing of all causal interaction and causal continuity between the constitutive components and processes instantiating the mind. Yes, we would be instantiated upon reaching our destination, but we can expect this to be as phenomenally discontinuous as brain death or destructive uploading.

There is much talk in the philosophical and futurist circles – where Substrate Independent Minds is a familiar topic and a common point of discussion – on how the mind is software. This sentiment ultimately derives from functionalism, and the notion that when it comes to mind it is not the material of the brain that matters, but the process(es) emerging therefrom. And due to the fact that almost all software is designed to as to be implemented on general purpose (i.e. standardized) hardware, that we should likewise be able to transfer the software of the mind into a new physical computational substrate with as much ease as we do software. While we would emerge from such a transfer functionally isomorphic with ourselves prior to the jump from computer to computer, we can expect this to be the phenomenal equivalent of brain death or destructive uploading, again, because all causal interaction and continuity between that software’s constitutive sub-processes has been discontinued. We would have been put on pause in the time between leaving one computer, whether as static signal or static solid-state storage, and arriving at the other.

This is not to say that we couldn’t transfer the physical substrate implementing the “software” of our mind to another body, provided they were equipped to receive such a physical substrate. But this doesn’t have quite the same advantage as beaming oneself to the other side of Earth, or Andromeda for that matter, at the speed of light.

But to transfer a given WBE to another mental substrate without incurring phenomenal discontinuity may very well involve a second gradual integration procedure, in addition to the one the WBE initially underwent (assuming it isn’t a product of destructive uploading). And indeed, this would be more properly thought of in the context of a new substrate being gradually integrated with the WBE’s existing substrate, rather than the other way around (i.e. portions of the WBE’s substrate being gradually integrated with an external substrate.) It is likely to be much easier to simply transfer a given physical mental substrate to another body, or to bypass this need altogether by actuating bodies via tele-operation instead.

In summary, it is substrate independence for mind in general, and not for a specific mind in particular (at least not without a gradual integration procedure, like the type underlying the notion of gradual uploading, so as to transfer such a mind to a new substrate without causing phenomenal discontinuity.)

The Term “Uploading” Misconstrues More Than It Clarifies

The term “Mind-Uploading” has some drawbacks and creates common initial misconceptions. It is based off terminology originating from the context of conventional, contemporary computers – which may lead to the initial impression that we are talking about uploading a given mind into a desktop PC, to be run in the manner that Microsoft Word is run. This makes the notion of WBE more fantastic and incredible – and thus improbable – than it actually is. I don’t think anyone seriously speculating about WBE would entertain such a notion.

Another potential misinterpretation particularly likely to result from the term Mind-Uploading is that we seek to upload a mind into a computer – as though it were nothing more than a simple file transfer. This, again, connotes modern paradigms of computation and communications technology that are unlikely to be used for WBE. It also creates the connotation of putting the mind into a computer – whereas a more accurate connotation, at least as far as gradual uploading as opposed to destructive uploading is concerned, would be bringing the computer gradually into the biological mind.

It is easy to see why the term initially came into use. The notion of destructive uploading was the first embodiment of the concept – the notion of gradual uploading so as to mitigate the philosophical problems pertaining to how much a copy can be considered the same person as the original, especially in contexts where they are both simultaneously existent, came afterward. In the context of destructive uploading it makes more connotative sense to think of concepts like uploading and file transfer.

But in the notion of gradual uploading, portions of the biological brain – most commonly single neurons, as in Robert A. Freitas’s and Ray Kurzweil’s versions of gradual uploading – are replaced with in-vivo computational substrate, to be placed where the neuron it is replacing was located. Such a computational substrate would be operatively connected to electrical or electrochemical sensors (to translate the biochemical or more generally biophysical output of adjacent neurons into computational input that can be used by the computational emulation) and electrical or electrochemical actuators (to likewise translate computational output of the emulation into biophysical input that can be used by adjacent biological neurons). It is possible to have this computational emulation reside in a physical substrate existing outside of the biological brain, connected to in-vivo biophysical sensors and actuators via wireless communication (i.e. communicating via electromagnetic signal), but this simply introduces a potential lag-time that may then have to be overcome by faster sensors, faster actuators or a faster emulation. It is likely that the lag-time would be negligible (especially if it was located in a convenient module external to the body but “on it” at all times, to minimize transmission delays increasing as one gets farther away from such an external computational device – which would also likely necessitate additional computation to model the necessary changes to transmission speed in response to how far away the person is – otherwise signals that are meant to arrive at a given time could arrive too soon or too late thereby disrupting functionality) but placing the computational substrate in-vivo obviates these potential logistical obstacles.

This notion is I think not brought into the discussion enough. It is an intuitively-obvious notion if you’ve thought a great deal about Substrate-Independent-Minds and frequented discussions on Mind-Uploading. But to a newcomer who has heard the term Gradual Uploading for the first time, it is all too easy to think “yes, but then one emulated neuron would exist on a computer, and the original biological neuron would still be in the brain. So once you’ve gradually emulated all these neurons, you have an emulation on a computer, and the original biological brain, still as separate physical entities. Then you have an original and the copy – so where does the gradual in Gradual Uploading come in? How is this any different than destructive uploading? At the end of the day you still have a copy and an original as separate entities.”

This seeming impasse is I think enough to make the notion of Gradual Uploading seem at least intuitively or initially incredible and infeasible before people take the time to read the literature and discover how gradual uploading could actually be achieved (i.e. wherein each emulated neuron is connected to biophysical sensors and actuators to facilitate operational connection and causal interaction with existing in-vivo biological neurons) without fatally tripping upon such seeming logistical impasses, as in the example above. The connotations created by the term I think to some extent make it seem so fantastic (as in the overly-simplified misinterpretations considered above) that people write off the possibility before delving deep enough into the literature and discussion to actually ascertain the possibility with any rigor.

The Computability of the Mind

Another common misconception is that the feasibility of Mind-Uploading is based upon the notion that the brain is a computer or operates like a computer. The worst version of this misinterpretation that I’ve come across is that proponents and supporters of Mind-Uploading are claiming that the mind is similar in operation current and conventional paradigms of computer.

Before I elaborate why this is wrong, I’d like to point out a particularly harmful sentiment that can result from this notion. It makes the concept of Mind-Uploading seem dehumanizing, because conventional computers don’t display anything like intelligence or emotion. This makes people conflate the possible behaviors of future computers with the behaviors of current computers. “Obviously computers don’t feel happiness or love, and so to say that the brain is like a computer is a farcical claim.

Machines don’t have to be as simple or as inadaptable and invariant as the are today. The universe itself is a machine – in other words either everything is a machine or nothing is.

It also makes people think that advocates and supporters of Mind-Uploading are claiming that the mind is reducible to basic or simple autonomous operations, like cogs in a machine, which constitutes for many people a seeming affront to our privileged place in the universe as humans, in general, and to our culturally-engrained notions of human dignity being inextricably tied to physical irreducibility, in particular. The intuitive notions of human dignity and the ontologically-privileged nature of humanity have yet to catch up with physicalism and scientific materialism (a.k.a. metaphysical naturalism). It is not the proponents of Mind-Uploading that are raising these claims, but science itself – and for hundreds of years I might add. Man’s privileged and physically-irreducible ontological status has become more and more undermined throughout history since at least as far back as the Darwin’s theory of Evolution, which brought the notion of the past and future phenotypic evolution of humanity into scientific plausibility for the first time.

It is also seemingly disenfranchising to many people, in that notions of human free-will and autonomy seem to be challenged by physical reductionism and determinism – perhaps because many people’s notion of free-will are still associated with a non-physical, untouchably-metaphysical human soul (i.e. mind-body dualism) which lies outside the purview of physical causality. To compare the brain to a “mindless machine” is still for many people disenfranchising to the extent that it questions the legitimacy of their metaphysically-tied notions of free-will.

Just because the sheer audacity of experience and the raucous beauty of feeling is ultimately reducible to physical and procedural operations (I hesitate to use the word mechanisms for its likewise-misconnotative conceptual associations) does not take away from it. If it were the result of some untouchable metaphysical property, a sentiment that mind-body-dualism promulgated for quite some time, then there would be no way for us to understand it, to really appreciate it, and to change it (e.g. improve upon it) in any way. Physicalism and scientific materialism were needed if we are to ever see how it is done and to ever hope to change it for the better. Figuring out how things work is one of Man’s highest merits – and there is no reason Man’s urge to discover and determine the underlying causes of the world should not apply to his own self as well.

Moreover, the fact that experience, feeling, being and mind result from the convergence of singly-simple systems and processes makes the minds emergence from such simple convergence all the more astounding, amazing and rare, not less! If the complexity and unpredictability of mind were the result of complex and unpredictable underlying causes (like the metaphysical notions of mind-body dualism connote) then the fact that mind turned out to be complex and unpredictable wouldn’t be much of a surprise. The simplicity of mind’s underlying mechanisms makes mind’s emergence all the more amazing, and should not take away from our human dignity but should instead raise it up to heights yet-unheralded.

Now that we have addressed such potentially-harmful second-order misinterpretations, we will address their root: the common misinterpretations likely to result from the phrase “the computability of the mind”. Not only does this phrase not say that the mind is similar in basic operation to conventional paradigms of computation – as though a neuron were comparable to a logic gate or transistor – but neither does it necessarily make the more credible claim that the mind is like a computer in general. This makes the notion of Mind-Uploading seem dubious because it conflates two different types of physical system – computers and the brain.

The kidney is just as computable as the brain. That is to say that the computability of mind denotes the ability to make predictively-accurate computational models (i.e. simulations and emulations) of biological systems like the brain, and is not dependent on anything like a fundamental operational similarity between biological brains and digital computers. We can make computational models of a given physical system, feed it some typical inputs and get a resulting output that approximately matches the real-world (i.e. physical) output of such a system.

The computability of the mind has very little to do with the mind acting as or operating like a computer, and much, much more to do with the fact that we can build predictively accurate computational models of physical systems in general. This also, advantageously, negates and obviates many of the seemingly dehumanizing and indignifying connotations identified above that often result from the claim that the brain is like a machine or like a computer. It is not that the brain is like a computer – it is just that computers are capable of predictively modeling the physical systems of the universe itself.

We Want Not To Become Machines, But To Keep Up With Them!

Too often is uploading portrayed as the means to superhuman speed of thought or to transcending our humanity. It is not that we want to become less human, or to become like a machine. For most Transhumanists and indeed most proponents of Mind-Uploading and Substrate-Independent Minds, meat is machinery anyways – in other words there is no real (i.e. legitimate) ontological distinction between them to begin with. Too often is uploading seen as the desire for superhuman abilities. Too often is it seen as a bonus, nice but ultimately unnecessary.

I vehemently disagree. Uploading has been from the start for me and I think for many other proponents and supporters of Mind-Uploading a means of life-extension, of deferring and ultimately defeating untimely, involuntary death, as opposed to and ultimately unnecessary means to better powers, a more privileged position relative to the rest of humanity or to eschewing our humanity in a fit of contempt-of-the-flesh. We do not want to turn ourselves into Artificial Intelligence, which is a somewhat perverse and burlesque caricature that is associated with Mind-Uploading far too often.

The notion of gradual uploading is implicitly a means of life-extension. Gradual uploading will be significantly harder to accomplish than destructive uploading. It requires a host of technologies and methodologies – brain-scanning, in-vivo locomotive systems such as but not limited to nanotechnology or else extremely robust biotechnology – and a host of precautions to prevent causing phenomenal discontinuity, such as letting each non-biological functional replacement time to causally interact with adjacent biological components before the next biological component that it causally interacts with is likewise replaced. Gradual uploading is a much harder feat than destructive uploading, and the only advantage it has over destructive uploading is preserving the phenomenal continuity of a single specific person. In this way it is implicitly a means of life-extension, rather than a means to the creation of AGI, because its only benefit is the preservation and continuation of a single, specific human life, and that benefit entails a host of added precautions and additional necessitated technological and methodological infrastructures.

If we didn’t have to fear the creation of recursively-self-improving AI, biased towards being likely to recursively-self-modify at a rate faster than humans are likely to (or indeed, are able to safely – that is, gradually enough to prevent phenomenal discontinuity), then I would favor biotechnological methods of achieving indefinite lifespans over gradual uploading. But with the way things are, I am an advocate of gradual Mind-Uploading first and foremost because I think it may prove necessary to prevent humanity from being left behind by recursively self-modifying superintelligences. I hope that it ultimately will not prove necessary – but at the current time I feel that it is somewhat likely.

Most people who wish to implement or accelerate an intelligence explosion al a I.J. Good, and more recently Vernor Vinge and Ray Kurzweil, wish to do so because they feel that such a recursively self-modifying superintelligence (RSMSI) could essentially solve all of humanity’s problems – disease, death, scarcity, existential insecurity. I think that the potential benefits of creating a RSMSI are superseded by the drastic increase in existential risk it would entail in making any one entity superintelligent relative to humanity. The old God of yore is finally going out of fashion, one and a quarter centuries late to his own eulogy. Let’s please not make another one, now a little reality under his belt this time around.

Intelligence is a far greater source of existential and global catastrophic risk than any technology that could be wielded by such an intelligence (except, of course, for technologies that would allow an intelligence to increase its own intelligence). Intelligence can invent new technologies and conceive of ways to counteract any defense systems we put in place to protect against the destructive potentials of any given technology. A superintelligence is far more dangerous than rogue nanotech (i.e. grey-goo) or bioweapons. When intelligence comes into play then all bets are off. I think culture exemplifies this prominently enough. Moreover, for the first time in history the technological solutions to these problems – death, disease, scarcity – are on the conceptual horizon. We can fix these problems ourselves, without creating an effective God relative to Man and incurring the extreme potential for complete human extinction that such a relative superintelligence would entail.

Thus uploading constitutes one of the means by which humanity can choose, volitionally, to stay on the leading edge of change, discovery, invention and novelty, if the creation of a RSMSI is indeed imminent. It is not that we wish to become machines and eschew our humanity – rather the loss of autonomy and freedom inherent in the creation of a relative Superintelligence is antithetical to the defining features of humanity, and in order to preserve the uniquely human thrust toward greater self-determination in the face of such a RSMSI, or at least be given the choice of doing so, may necessitate the ability to gradually upload so as to stay on equal footing in terms of speed of thought and general level of intelligence (which is roughly correlative with the capacity to affect change in the world and thus to determine its determining circumstances and conditions as well).

In a perfect world we wouldn’t need to take the chance of phenomenal discontinuity inherent in gradual uploading. In gradual uploading there is always a chance, no matter how small, that we will come out the other side of the procedure as a different (i.e. phenomenally distinct) person. We can seek to minimize the chances of that outcome by extending the degree of graduality with which we gradually replace the material constituents of the mind, and by minimizing the scale at which we gradually replace those material constituents (i.e. gradual substrate replacement one ion-channel at a time would be likelier to ensure the preservation of phenomenal continuity than gradual substrate replacement neuron by neuron would be). But there is always a chance.

This is why biotechnological means of indefinite lifespans have an immediate advantage over uploading, and why if non-human RSMSI were not a worry, I would favor biotechnological methods of indefinite lifespans over Mind-Uploading. But this isn’t the case, rogue RSMSI are a potential problem, and so the ability to secure our own autonomy in the face of a rising RSMSI may necessitate advocating Mind-Uploading over biotechnological methods of indefinite lifespans.

Mind-Uploading also has some ancillary benefits over biotechnological means of indefinite lifespans as well, however. If functional equivalence is validated (i.e. if it is validated that the basic approach works), mitigating existing sources of damage becomes categorically easier. In physical embodiment, repairing structural, connectional or procedural sub-systems in the body requires (1) a means of determining the source of damage and (2) a host of technologies and corresponding methodologies to enter the body and make physical changes to negate or otherwise obviate the structural, connectional or procedural source of such damages, and then exit the body without damaging or causing dysfunction to other systems in the process. Both of these requirements become much easier in the virtual embodiment of whole-brain-emulation.

First, looking toward requirement (2), we do not need to actually design any technologies and methodologies for entering and leaving the system without damage or dysfunction or for actually implementing physical changes leading to the remediation of the sources of damage. In virtual embodiment this requires nothing more than rewriting information. Since in the case of WBE we have the capacity to rewrite information as easily as it was written in the first place, while we would still need to know what changes to make (which is really the hard part in this case), actually implementing those changes is as easy as rewriting a word file. There is no categorical difference, since it is information and we would already have a means of rewriting information.

Looking toward requirement (1), actually elucidating the structural, connectional or procedural sources of damage and/or dysfunction, we see that virtual embodiment makes this much easier as well. In physical embodiment we would need to make changes to the system in order to determine the source of the damage. In virtual embodiment we could run a section of emulation for a given amount of time, change or eliminate a given informational variable (i.e. structure, component, etc.) and see how this affects the emergent system-state of the emulation instance.

Iteratively doing this to different components and different sequences of components, in trial-and-error fashion, should lead to the elucidation of the structural, connectional or procedural sources of damage and dysfunction. The fact that an emulation can be run faster (thus accelerating this iterative change-and-check procedure) and that we can “rewind” or “play-back” an instance of emulation time exactly as it occurred initially means that noise (i.e. sources of error) from natural systemic state-changes would not affect the results of this procedure, whereas in physicality systems and structures are always changing, which constitutes a source of experimental noise. The conditions of the experiment would be exactly the same in every iteration of this change-and-check procedure. Moreover, the ability to arbitrarily speed up and slow down the emulation will aid in our detecting and locating the emergent changes caused by changing or eliminating a given microscale component, structure or process.

Thus the process of finding the sources of damage correlative with disease and aging (especially insofar as the brain is concerned) could be greatly improved through the process of uploading. Moreover, WBE should accelerate the technological and methodological development of the computational emulation of biological systems in general, meaning that using such procedures to detect the structural, connectional and procedural sources of age-related damage and systemic dysfunction in the body itself, as opposed to just the brain, as well.

Note that this iterative change-and-check procedure would be just as possible via destructive uploading as it would with gradual uploading. Moreover, in terms of people actually instantiated as whole-brain-emulations, actually remediating those structural, connectional and/or procedural sources of damage as it pertains to WBEs is much easier than physically-embodied humans. Anecdotally, if being able to distinguish between the homeostatic, regulatory and metabolic structures and processes in the brain from the computational or signal-processing structures and processes in the brain is a requirement for uploading (which I don’t think it necessarily is, although I do think that such a distinction would decrease the ultimate computational intensity and thus computational requirements of uploading, thereby allowing it to be implemented sooner and have wider availability), then this iterative change-and-check procedure could also be used to accelerate the elucidation of such a distinction as well, for the same reasons that it could accelerate the elucidation of structural, connectional and procedural sources of age-related systemic damage and dysfunction.

Lastly, while uploading (particularly instances in which a single entity or small group of entities is uploaded prior to the rest of humanity, i.e. not a maximally distributed intelligence explosion) itself constitutes a source of existential risk, it also constitutes a means of mitigating existential risk as well. Currently we stand on the surface of the earth, naked to whatever might lurk in the deep night of space. We have not been watching the sky for long enough to know with any certainty that some unforeseen cosmic process could not come along to wipe us out at any time. Uploading would allow at least a small portion of humanity to live virtually on a computational substrate located deep underground, away from the surface of the earth and its inherent dangers, thus preserving the future human heritage should an extinction event befall humanity. Uploading would also prevent the danger of being physically killed by some accident of physicality, light being hit by a bus or struck by lightning.

Uploading is also the most resource-efficient means of life-extension on the table, because virtual embodiment not only essentially negates the need for many physical resources (instead necessitating one, namely energy – and increasing computational price performance means that just how much a given amount of energy can do is continually increasing).

It also mitigates the most pressing ethical problem of indefinite lifespans – overpopulation. In virtual embodiment, overpopulation ceases to be an issue almost ipso facto. I agree with John Smart’s STEM compression hypothesis – that in the long run the advantages proffered by virtual embodiment will make choosing it over physical embodiment, in the long run at least, an obvious choice for most civilizations, and I think it will be the volitional choice for most future persons. It is safer, more resource efficient (and thus more ethical, if one thinks that forestalling future births in order to maintain existing life is unethical) and the more advantageous choice. We will not need say: migrate into virtuality if you want another physically-embodied child. Most people will make the choice to go VR themselves simply due to the numerous advantages and the lack of any experiential-incomparabilities (i.e. modalities of experience possible in physicality but not possible in VR).

So in summary, yes, Mind-Uploading (especially gradual uploading) is more a means of life-extension than a means to arbitrarily greater speed of though, intelligence or power (i.e. capacity to affect change in the world). We do not seek to become machines, only to retain the capability of choosing to remain on equal footing with them if the creation of RSMSI is indeed imminent. There is no other reason to increase our collective speed of thought, and to do so would be arbitrary – unless we expected to be unable to prevent the physical end of the universe, in which case it would increase the ultimate amount of time and number of lives that could be instantiated in the time we have left.




The fallibility of many of these misconceptions may be glaringly obvious, especially to those readers familiar with Mind-Uploading as notion and Substrate-Independent-Minds and/or Whole Brain Emulation as disciplines. I may be to some extent preaching to the choir in these cases. But I find many of these misinterpretations far too predominant and recurrent to be left alone.