Persons, Programs, and Uploading Consciousness

David Ross

Abstract:

The pursuit of indefinitely-extended life falls into two categories: repair of our current bodies and movement of our personalities into other bodies. This article deals with a subset of the latter: the uploading of the human personality into a computer. It is divided into four sections, each of them introduced by a question.

1. What would an intelligent program be like?
2. Is the human mind an intelligent program?
3. What does it mean to upload rather than copy a human?
4. How would you upload a human?

What Would an Intelligent Program Be Like?

The short form of the argument for human uploading goes like this. The human mind is a program running on a computer which is the brain. Silicon-based computers are getting more and more powerful all the time. Before long they will equal human brain power and then it should be possible, at least in principle, to transfer the mind-program from the brain to a silicon computer of equal or superior power. Ignoring for a moment the question of whether the human mind is a computer program, this line of reasoning conceals a serious misconception, which much be cleared up at the start.

The argument ignores the important distinction between the processing power of a computer and the IQ of a program running on that computer. This distinction is often lost in discussions of machine intelligence. Several authors, Moravec sometimes among them¹, have pointed to the continual exponential growth in desktop computing power, shown that it passes the human-level ten teraflop range about 2025, and from this argued that human-level intelligent programs will become possible around then.

The truth is that far more progress has been made increasing computer speeds in the past thirty years than in increasing the mind-power of computer programs. The two phenomena (increase in computer power and increase in program intelligence) have almost nothing to do with each other. A program with human-level intelligence does not require a

super-megacomputer to run. If any form of intelligent program, human or otherwise, is possible, it can run on the simplest form of computer, provided it has enough memory. It will run equally well, and with equal intelligence, on a computer built of tinker-toys as on one with multiple teraflops of capability. It will just run incomparably slower.

If this is true for intelligent programs, it is also true for human personalities uploaded onto computers. The shortcut argument that all it takes is faster machines to make possible intelligent programs and (by extension) human uploading is false. Establishing the possibility of uploading must go by a more circuitous route. Fortunately, following that route gives quite a bit of insight into how to actually perform the uploading.

What is an Intelligent System?

Intelligence is like pornography: We can’t define it, but we know it when we see it. We are probably willing to accept a system as intelligent if it “feels” intelligent in the same way as does another human. That is, if it acts intelligently in most of its dealings with us. The same thing is true for self-awareness. It too must be established from the outside. Until such time as we upload ourselves onto a computer and then observe ourselves, there is no way to prove whether any program is self-aware. For now, a self-aware or an intelligent program must be one that can pass whatever tests would convince us of another human’s self-awareness or intelligence.

This is one form of the so-called Turing test.

Self-awareness and intelligence are not the same thing. It seems quite possible that an intelligent program might not be self-aware, or that a self-aware program might not have human-level intelligence. Nevertheless, since we want an uploaded human to be both intelligent and self-aware, for the purposes of this article we will lump the two concepts together under the title “intelligent”.

Many people are uncomfortable with the subjective form of the Turing test, but all objective tests of intelligence in the form of a finite set of pre-established questions must fail. They must fail because a computer could always be programmed to give the same answers to those questions that a human had already given. In the same way, it is incorrect to say that “If only a computer could do ‘X’ I would consider it intelligent.” Time and again computers have accomplished tasks previously thought to require intelligence—playing chess, reading characters—and done so in clearly non-intelligent ways. Finite task testing, like finite question list testing must fail, and the only real way of establishing the intelligence of another entity is subjective.

One reason why there can never be a finite set of tests that establish the intelligence of a program is because whatever else intelligence may be, it is a function of complexity. A “simple” system is one where its basic components—such as the rules in an expert system—are still individually important. A “complex” system is one where the individual building blocks are submerged in levels of hierarchy and com-

EXTROPY #9 Summer 1992

12

plexity. An intelligent system must be complex. Perhaps unsurprisingly, it shares this property with life and with entropy.

Entropy cannot appear if the number of particles involved is too small or the particles are too simple. The equations of motion of a simple system of colliding particles are time-reversible. It is only when the system is sufficiently complex—when the basic building blocks of the system have all but vanished into the background—that entropy can arise. This phenomenon of a high-level property of a system appearing only when the low-level constituents have effectively disappeared is important. A system that demonstrates entropy contains too many particles to deal with individually; they must be tracked statistically. Note, however, that the arising of entropy is not just a statistical trick. Even a system of trillions of particles is in principle trackable using purely time-reversible Newtonian equations. And yet, despite this, such a system has irreversible, entropy-increasing modes of behavior. Entropy is not the same thing as chaos—the inability to track a system because of round-off error.

Life, too, appears only with a certain level of complexity. Very complex molecules, up to about the level of the simplest viruses, are not fully alive. But the more complex viruses, and the simplest single-cell objects, are alive. The system has passed a level of basic complexity separating life from non-life, and again done so when it is no longer sensible to talk about a group of single constituents, but rather of systems of systems or something even more complex.

So it seems to be with intelligence. Only a ‘sufficiently complex’ system can be intelligent. Some heuristic arguments for why this is so are possible. An intelligent system is capable of generating new information from old information. This is not just a recombining of the old information, but a recombining of it in ways that demonstrate non-obvious links and consequences—in short, in ways which generate new information. But Shannon has shown that information can be expressed mathematically in the same form as entropy (with a change of sign). In effect, information is the opposite of entropy.

To some degree, the generation of previously unavailable information can be seen as reducing the entropy of the system. For the entropy of a system to be reduced, it must be sufficiently complex to have entropy in the first place. Note that it is not the processing of information that requires either complexity or intelligence. It is the generation of new information that does so.

Probably the single most egregious

failure of the Expert System school of Artificial Intelligence, led by Marvin Minsky, has been the failure to understand that an intelligent system must be complex. Thus, a simple rule-following system, no matter how well it may mimic a human in some limited regime, is not intelligent, and adding more discrete rules can never lead to intelligent behavior. This, perhaps, explains the failure of the Expert System school of AI to produce intelligent systems.

This failure is, in a way, unfortunate. If intelligence could be the product of a simple rule-based system, then those rules could, in principle, be programmed into a computer. The resulting system would then presumably be intelligent. Uploading a human (or at least copying a human onto a computer—we’ll look at the difference later) would then be a question of determining the complete set of rules that specify the human and programming a suitable computer with those rules. This point of view has been one of the driving concepts behind the expert systems view of artificial intelligence from its beginning. Unfortunately, if intelligence requires complexity great enough that the system building blocks effectively vanish, the only way a rule-based system could be intelligent would require it to have so many rules arranged so complexly that the individual rules are effectively lost in the noise. It is difficult to see how anyone could write such a system. Perhaps one could eventually be grown by an automatic rule-writing program—or by several generations of them.

It should be kept in mind that complexity alone is not sufficient to generate intelligence, any more than it is sufficient to generate entropy or life. The Internet, the interlocking system of computer networks that connects most military and educational institutions and an increasing number of commercial ones, has on it computing power collectively approaching that of a single human brain, and that computing power is arranged in multitudes of very complex systems, but it is shown no sign of either life or intelligence. In conclusion, for a program to be intelligent it must be “complex” and it must be able to convince a human of its intelligence. We do not yet know how to write such programs, but they certainly do not seem impossible in principle. Surprisingly, if the human mind turns out to be such a program, we may be able to copy (or transfer) ourselves without completely understanding how the program works, modify the result, and produce truly intelligent programs that way — a kind of reverse engineering of ourselves.

Is the Human Mind an Intelligent Program?

The Computers Can’t Think Fallacy

While most students of spontaneous order, and indeed most computer scientists, would claim that intelligent programs are possible, there are several other schools of philosophy that say intelligent or self-aware algorithms are not possible, and that therefore, by extension, a human cannot be uploaded onto a machine without losing his intelligence or his self-awareness. Because this question is central to the problem of uploading, we’ll take a slight detour to attempt to answer some of their objections. More detailed refutations of the “computers can’t think” school of philosophy can be found in endnotes.

Simulating intelligence versus real intelligence

One false path the “computers can’t think” school goes down is when they discuss the difference between a simulation of something and the thing itself. This argument shows up in several forms, many of them made by John Searle. One form of it is that all computers can do is simulate physical processes. They cannot duplicate them.

Searle has correctly pointed out that no matter how well a computer simulates the chemical processes in a flame, it will not produce the flame’s heat. This is because a flame is an example of a process that is “essentially material”. The simulation of an “essentially material” process will always be different in kind from the process itself.

There are other kinds of processes, however, which are not “essentially material”. For these processes there is little or no difference between the process itself and the simulation of the process. Consider addition. I can add two numbers in my head, and everyone would say I was performing addition. Would anyone say that if I programmed a computer to simulate what my brain is doing that the resulting system would not be doing addition? Addition is a process for which the simulation of the process is the same as the process itself.

A form of simulation that is directly relevant to human uploading is the construction of an emulator. If a programmer has a program which he must run on a different kind of computer than it was written for, but for which he does not have the source code, he will build an emulator.

13

EXTROPY #9 Summer 1992

An emulator is a special computer program that translates the system calls of one kind of computer into system calls of the other. Thus, if the pointer to the “ADD” instructions is stored in register 18, say, on the computer for which the program was originally written, and in register 32 on the computer on which it must run, the emulator will intercept all calls to register 18 and redirect them to register 32. Except for some degradation in speed, the executing program will perform the same on the new system as it did on the one for which it was developed.

Although the program-plus-emulator is “simulating” the program running on its original computer, there is no doubt that the program is actually running on the new system. In essence, it is the hardware, not the software that is being simulated. The software is just being run. If we can show that a computer program can be a brain emulator — that is, emulate the hardware at a sufficiently low level — then such a computer-plus-emulator should be able to run a human mind-program.

Processes like addition and computer programs are “essentially immaterial”. To simulate them is to run them. Searle’s argument that at most computers can only simulate thought requires him to show that the functioning of the human mind is an “essentially material” process. He does not do so and I do not believe he can. In fact, I believe thought can be shown to be “essentially immaterial” under the above definitions.

Another argument made by Searle is that computers cannot think because “they are made of the wrong kind of substance to think”. This argument is really a form of the first one — that all computers can do is simulate thought. It is based on the same erroneous assumption that a mind is a physical process on the order of a chemical reaction — that it is “essentially material” by my definition. After making this assumption, Searle then establishes correctly that for essentially material processes simulation and reality are two different things. From this he deduces that computers may be able to simulate intelligence, but they can never actually be intelligent. In effect, for Searle, the mind is the brain and the brain is a physical protoplasmic object.

The fallacy in Searle’s line of reasoning is that it starts by assuming its conclusion. He assumes that the mind is essentially material, shows that simulation of essentially material processes is different in kind from the processes themselves, and then concludes that a simulation of a mind cannot itself be a mind.

The “The Mind is a Computer Program” Misunderstanding

On the opposite side of the intelligent program question there is also a misunderstanding that must be cleared up. This is the “radical dualist” view of the mind/brain question. This way of reasoning looks on the mind/brain system as directly analogous to a program/brain system. Proponents of this view hold, in essence, that the mind is a program running on the hardware of the brain.

Their syllogism goes as follows. Major Premise: Any computer program can be moved from one system to another system of sufficient power without essentially altering it. Minor Premise 1: The human brain is a computer. Minor Premise 2: The human mind is a program. Conclusion: the human mind can be moved from the computer that is the human brain to another computer of sufficient power.

The brain, if it is a computer at all, is pre-von Neumann. Brain structure at every level determines the functioning of the mind. A given brain, with its structure intact, simply cannot run a different program. The wiring is the program.

I believe this line of reasoning also to be in error, though not so seriously as Searle’s. Its primary failure is that it ignores how much more important the structure of the brain is to the functioning of the mind than the computer’s structure is to the functioning of the program.

In a computer, the physical layout and physical connectivity of the system does not change from one program to another. Many different programs can run on the same computer, with the only differences coming at the micron-scale level on which bits are stored in computer memory.

If a mind, on the other hand, is a program, it is a very peculiar sort of program indeed. Von Neuman invented computer programming because he got tired of changing the physical wiring on the ENIAC for each new program. Instead, he wired the computer a set way and modi

fied the data memory registers whenever he wanted to run a different program. Before von Neuman, the program ENIAC ran was determined by its wiring. In essence the wiring and the program were the same thing. The brain, if it is a computer at all, is pre-von Neuman. Brain structure at every level determines the functioning of the mind. A given brain, with its structure intact, simply can not run a different program. The wiring is the program.

The central question in human uploading, then, is can we provide a von Neuman-type modification to the brain? Can we change the essential form of the brain/computer from one where the structure is the program to one which will run on a general purpose computer? And can we guarantee that it is the same program when we are done?

The answer to these questions is yes. Such a von Neuman modification is possible in principle. Consider the actual change that he made to ENIAC. Before he learned how to program it, the memory registers were used to store initial conditions and intermediate result data for running the program. How that data was used was determined by which wires were connected to which other wires. After he changed the system, it was no longer necessary to physically modify the wiring by hand to make it run a program. The data in the registers, besides holding the same initial condition and intermediate result data, also held the program information — information that ENIAC read and used to flip electromechanical switches that in turn determined which wires were connected to which others. The programming had replaced the human, but the connectivity of the wiring still changed depending on what was being run on the system. All von Neuman changed was what was doing the rewiring.

It is, after all, strictly incorrect to say that a computer running one program has the same structure as one running another. Throughout the CPU (central processing unit) switches are being set and unset millions of times a second, with their configuration determined by the program steps as it operates. Even though we cannot see it, the physical structure of the computer is changing continuously as a program executes, and in ways that are precisely determined by the program. Even at the beginning, the bits that are the program on disk or in memory, before it starts to execute, are held by transistors in physically different states depending on whether a given bit is a one or a zero.

Thus, the difference between a pre- and a post-von Neuman computer is one of degree, not one of kind. The structure is still being modified as the program ex

EXTROPY #9 Summer 1992

14

ecutes. What is different is the level on which the modification takes place. At its most basic level, a program is dependent on the structure of the particular computer on which it is running. If the bits that specify the program are not on and off — physically charged one way or the other — then the program isn’t even there. If it doesn’t have a physical way of putting intermediate and final results into a memory, it cannot run. In essence two different program cannot be run on the same computer, because the programs modify the computer in essential ways when they run.

The real question, therefore, is can a human mind modify the general purpose switching of a general purpose computer instead of the very special purpose wiring of a human brain. Can there be a level of interface above which is the human mind and below which is an emulation of the human brain hardware?

In the mind plus brain-emulator plus computer system it is not really relevant where you draw the line and say from here up is the mind and from here down is the emulator. All that matters is that the emulation reach low enough down (probably at least to the individual neurons) so that it is emulating systems that are below the essential level of the brain. Remember that a mind is a ‘complex’ system. It arises when the basic constituents are such a small part of the overall system that exactly what they are, provided they work correctly, does not matter. I don’t believe anyone would argue that any given neuron or any given synapse is important for thought. Even at the most basic level, it is how that neuron interacts with the tens of thousands of other neurons to which it might be connected that matters.

We can successfully run a mind/brain on the computer if our program emulates the brain at the neuron/synapse level. Below that level, the computer is just transistors and switches — a general-purpose von Neuman computer. What runs above the level where emulation starts can be looked at as a brain emulator with a mind program running above it, but there is really only one program. By running the emulator all the way down to the level of the individual neuron and synapse, we are in effect running both the brain (as an emulator) and the mind (as a program) on the computer, regardless of where you consider the brain to leave off and the mind to begin — or even if you consider the question irrelevant. We have finessed the problem of mind/brain duality by going down to the level where the brain itself leaves off.

What Does it Mean to Upload a Human?

There is some dispute as to whether ‘uploading’ or ‘downloading’ is the proper term for moving a person onto a computer. In general computer parlance, ‘uploading’ refers to moving a file from your disk to your computer, or from your local computer to a remote system. Downloading usually refers to copying a file from a remote system to your local computer. If looked at from the perspective of the person doing the file transfer, if the movement is in some sense away from him, the process is called ‘uploading’, while if it is toward him, it is ‘downloading’. If it is the person himself being moved, the terms become confused. As the first elements of the personality are transferred, from the perspective of most of the person, the process is uploading. As the last elements come across, and most of the personality has already been transferred, it looks like downloading. Neither term is completely accurate. I suppose I could follow Max More’s lead and coin a new term, say ‘transloading’, but that seems unnecessary. I will adopt the viewpoint at the beginning of the process and call the procedure ‘uploading’.

The main reason people want to upload themselves onto a computer or other non-biological system is to escape death. The point of uploading is to transfer your personality from your body to the computer. However nice it might be to have a copy of yourself running around, that is not enough. It is not enough even if the copy thinks it is you. The goal is to complete a direct transfer of consciousness, and to do it in such a way that a person can be sure beforehand that he will make it across onto the computer. The only way for that certainty to occur, at least until enough people have made the transfer that anecdotal evidence of successful uploading is available and believed, is for the person to be conscious throughout the transference.

In deciding whether uploading is theoretically possible we have to make several careful distinctions, particularly between uploading and copying. There are a wide range of philosophies concerning what the self is. For a quite thorough discussion of the current theories and their applicability to uploading, I strongly recommend Max More’s Ph.D. dissertation$^{5}$. In discussing whether a consciousness can be uploaded, we need to be careful about directions in time. Were a person copied identically, both the original and the copy, looking backward in time, would feel that they were the original. They would remember equally well events that happened to the person, and they would both swear that

the other was the copy. It is important also to note that neither one of them would feel that they would continue were only the other one to survive.

If it is just looking backwards in time — I feel that I was uploaded is all that matters — then I could let someone copy me and then kill my original and not care. However strong might be philosophical arguments that the original self survives in such a situation, to a person looking ahead in time to the transfer such a situation is not much preferable to just being killed. It certainly does not feel like a transfer.

There has been substantial discussion that one of the desirable consequences of uploading is the possibility of making backup copies of yourself to be activated in event of the destruction of the original. A backup, as normally envisioned, is a copy of the original. While, looking backwards in time, it would likely believe that it was the original, it would not be the original, since, at the time the copy was made, the original still existed, and presumably had its own opinion about who was there first. By keeping a backup up to date using techniques similar to those discussed below for uploading, however, it should be possible to ‘upload’ yourself to a new program in all but the very worst of circumstances.

It is easy to picture a copying process where it is not possible from the perspective of either the person being copied or people outside to tell which is the original and which is the copy. We really have no way of knowing when we awoke this morning that we were the same person who went to sleep the evening before. Had someone made a copy while we slept and then destroyed the original, we would never know it.

Nonetheless, because we have lots of experience of continuity of self while conscious, we are willing, looking forward in time, to accept periods of unconsciousness — whether sleep or anesthesia — without worrying about whether it will be ‘us’ that makes it through. But, as in lots of other ways, once copying and uploading become possible, our common sense can easily fail us. I submit that while there may be artifacts of the copying process that enable us to distinguish an original and a copy (someone reliable witnessed the process, say, and states that the copy was made at a location remote from the original without the original being disturbed), if the person is not conscious, then except for those artifacts, there is no way for a person looking forward to such a procedure to be sure he would survive.

Being conscious through the process, however, should be sufficient for the person involved (who, after all, is the one who

15

EXTROPY #9 Summer 1992

matters) to be assured beforehand that he will feel that he will make it through each part of the process to the end with his personality still intact.

How Do You Upload a Human?

If the arguments presented above are correct, we do not need to argue directly that an entire mind can be transferred to computer hardware. All we have to argue is that a nerve and all its synapses can be. To upload a mind we just ‘upload’ neurons one at a time until we’ve uploaded everything. It isn’t even necessary to understand the functioning of the brain as a whole to do this. As long as the system can recognize what a neuron is and replace it with a nerve-replacement structure (NRS), uploading can be accomplished.

This NRS is the key to uploading. It has two purposes. First, when it replaces a neuron, it must function just like the neuron it replaced. Externally, this means that it must interact with the remaining neurons with which it is in contact just like the neuron did. This must be done through physical actuators that release and absorb synaptic chemicals and electrical impulses.

Internally, the replacement structure must be a program running on a computer that controls the actuators. This program is, in effect, a neuron emulator that learned, in the process of replacing the neuron, how to successfully emulate it.

The second purpose of the NRS comes into use only when all neighboring neurons have also been replaced. At that point instead of interacting with neighboring NRSs through actuators, it emulates the synapse as well. Now, it is all program.

This procedure continues until the entire brain and all sensory structures have been replaced. Since seeing the transfer first-hand is important, let’s look at it from the perspective of a person having it done.

Uploading Jason Macklin

Jason Macklin turns his head slightly but cannot see the tube connected to his neck through his carotid artery. Still, he knows it is there, and he knows what it will do. For years he has resisted the urgings of family and friends to get rid of his natural body and upload his mind onto the Web, to become a creature in Cyberspace like them. Unlike nearly everyone else, he doesn’t have even a neural tap to communicate instantly with any person or database on the world-wide Cyberspace Web. He sighs. Maybe he has been old-fashioned. If he’d already acquired artificial senses through direct neural implants, the

idea of replacing his entire brain with such structures wouldn’t be so frightening. Still, the doctors say he’ll feel absolutely nothing while the process is going on.

When human uploading first became possible, there was substantial debate over whether the uploaded person was the original or just a copy. A person uploading himself to avoid dying along with his body needed to know that he would really make to transition to Cyberspace. It might be nice to know that a copy of you would survive your death, but not nearly as nice as knowing you would. And the only way to be sure that it was really you who made the transfer was to do it fully conscious.

While Jason ponders his decision, the process is already underway. Nanomachines have crossed the blood-brain barrier and are systematically replacing each neuron with a functionally equivalent artificial structure. This structure is very special. Part of it is a series of nanomachine actuators that interact with neighboring cells just as though the replaced neuron were still in place. The rest of his body cannot tell that anything has changed. The remainder of the structure is a program that emulates the neuron and controls the actuators.

A person is not just brain cells. The nanomachines invading Jason’s body will replace all sensory neurons as well, and then replace all the parts of his body that influence the neurons with programs to do the same thing. Thus, muscles that stretch and flex will have their program equivalents, and neither the replaced nor the remaining neurons will be able to tell which is which.

On the main computer there exists an artificial world that, as far as sensory input is concerned, is identical to what Jason sees and hears and feels as he lies on his bed. All of the sensory input — and this includes the food in his stomach, the orientation of his limbs, his sense of breathing, as well as sight and sound — is duplicated in Cyberspace. When every neuron and its surrounding tissues have been replaced by their temporary physical counterparts, the structures on the computer that control each replacement part begin to interact among themselves directly, in direct synchronization with how they perform in his body. Gradually, each synapse in his brain is absorbed into the program structure of the emulation program, its functionality retained but its physical structure gone.

Jason’s eyes are open, staring at the ceiling, but even as the cells within his eyes are replaced, he has no sense of disorientation. For a time, the unreplaced cells are getting input from photons in the room while the replaced cells are getting

input from the artificial world in Cyberspace. Because the synchronization is complete, he cannot tell which part gets which input. After his eyes have been completely uploaded, everything still looks the same.

After a while, a doctor comes into the room and removes the tap into his neck. She holds out her hand and tells Jason to stand. For a moment, Jason wonders why he feels so normal. The doctor’s touch, the pressure of the floor against his feet, all feel completely physical. And yet he knows that all the sees and feels now is part of Cyberspace.

The doctor leads him over to a curtain at the side of the room and draws it back. Through the revealed window, Jason looks into another room, identical to the one he is in. On a bed in the middle of the room lies his body, still connected to its cable. For a moment, he watches it breathe. All of its neurons have been replaced, but his new autonomic nervous system still controls his old body. Just as in a dream, however, his conscious movements no longer make his old body move. The doctor hands him a switch which he knows will turn off his old body. He represses the feeling that he is committing suicide and throws the switch. In the next room the body — he no longer thinks of it as himself — releases its last breath and seems to relax. All nerve and muscle connections are severed at once and the body dies instantly.

He feels less emotion than he thought he would. He knows that if he doesn’t like it here in Cyberspace, he can always have another physical body constructed, grown from his original DNA, if he wishes. But right now that is not his concern. The room around him appears ordinary and familiar in order to ease his transition into Cyberspace, but outside the door is a vast new world. It will take time to learn all it has to offer. But time is what he now has. As an entity running in Cyberspace, with proper backup programs, he is assured of essentially unlimited life. □

References

¹Hans Moravec, Mind Children: The Future of Robot and Human Intelligence. (Harvard University Press, 1988.)

²Claude E. Shannon, “A Mathematical Theory of Communication.” Bell System Tech. J. (July, 1948, October 1948.)

³Paul M. Churchland, “Functionalism, Qualia, and Intentionality.” Philosophical Topics 12, no.1: 121-145. Daniel C. Dennett, Brainstorms. (Montgomery, 1978.)

⁴John Searle, “Minds, Brains, and Programs.” Behavioral and Brain Sciences 3, no.3: 417-457.

⁵The Diachronic Self: Identity, Continuity, Transformation.

EXTROPY #9 Summer 1992

16