Chris E. Stout
Immortality has been pursued by humanity through religion, science, and philosophy for centuries. Immortality has often been the domain of science fiction plots and fantasy. In the digital age, immortality mutates into emmortality: the use of electronic media to emulate a person in perpetuity. In the development of this model, there are several psychological concerns (e.g., consciousness, perception, thinking, reason, etc.) as well as philosophical quandaries (how would this change our cosmology?) that shall be addressed. The technology for this may be the easy part.
This proposal comes with a few basic caveats or limitations. First of all, this is not a concept of immortality that a subject gets to enjoy - it's for the "benefit" of others (which is, arguably, what immortality is). This is not a cryogenic, Golem-like, or Frankenstein-esque solution. Instead, it is a focus on technology, not biology.
Paul Ryan notes that "'Immortality' depends on the human practice of remembering the dead" (1991, p. 225). Allucquere Rosanne Stone said, "...it is important to remember that forgetting about the body is an old Cartesian trick, one that exacts a price from those bodies rendered invisible by the act of forgetting..." Humans can keep someone alive in their memories; memories can be prompted by an ideographic circumstance, photo, or video which leads into bittersweet remembrances. George Berkeley, an 18th-century British philosopher, similarly defined existence as being perceived.
But what if one could run a program that would simulate the advice a long-dead grandfather would have given to an unborn grandchild? What would a deceased spouse counsel in a time of crisis? I am suggesting the use of technology to create what I call the Emmortality Program. The Program would emulate its user, and, when the user no longer has a cooperative body (i.e., after death), the Emmortality Program would substitute for the departed human. It would be ideal to maintain the person in a functional state, but medical technology has, thus far, been unsuccessful in achieving this objective.
I am assuming that such medical technology would yield an inefficient use of resources in keeping the body alive. I'm suggesting consideration for maintaining the mind instead; hence, emmortality (an emulation of it). It may seem a little macabre, but personally, I would love to interact with a PC version of my grandparents. I know that an Emmortality Program would not be them, per se, but an emulation. Nevertheless, they'd seem almost as real as anyone else with whom I communicate via e-mail.
Most of what I discuss herein already exists, and what doesn't, will. Again, this is not science fiction, it's science proposition. We already have personal digital assistants (PDAs) that "learn" to make presumptions about numerous patterns of user activity. Chris Langton and Doyne Farmer, pioneers in a field known as Artificial Life, note that many basics have already been worked out. For example, John Holland has developed a classifier system, and as Langton notes, there are "...other systems that use genetic principles in order to search large problem space to help find optimal solutions or to help find better solutions than the ones we know. They are using principles imported from biology, those of mutation and genetic recombination - programs that are represented in such a way that most of the operations we do with them will result in viable programs... Another example that comes close to being alive is computer viruses, which satisfy a lot of criteria for living things" (p. 6).
The mathematician, Norbert Wiener, coined the term "cybernetics" to refer to self-regulating machines. In his 1948 book, Cybernetics: Control and Communication in the Animal and the Machine, he examines the likenesses between animals (including homo sapiens) and machines. Hardison pointed out that it is natural to compare cybernetic machines to humans. This raises the question, "would cybernetic machines possess self-awareness?" It's not that they would be self-aware per se, but the Emmortality Program could be designed to mimic self-awareness via some Goedel-type algorithm. Such machines already provide a handy "three-dimensional metaphor for self-awareness" (p. 294). This technology is nothing to fear. J.G. Ballard has iatrogenically defined personal computers as the brain's "...subcontracting of many of its core functions, creating a series of branch economies that may one day amalgamate and mount a management buy-out."
Farmer adds that, "Lifeness should perhaps be thought of as a continuous property. To me, a machine is a little more alive than a rock and probably less alive than a virus, which is less alive than a bacteria, which is probably less alive than [a human]. But nature can throw an Avogadro's number of computers at something because it's got zillions of molecules, all of which act like independent parallel processors. We really can't do that. We don't have that kind of computing power at our disposal, so we are forced to make these abstractions where we take an aspect of something out and build a little model around it that does what the original does, and so we have models of living things" (p. 7). It is one of these little models that can be made to emulate someone into emmortality. This may be considered the ultimate in inbreeding: self-regeneration without cloned progeny.
Individuals develop via various cognitive inputs and by data processing. Little that contributes to one's attitudes, beliefs, or opinions, is innate, although such reductionism may be distasteful to some. A PDA-like program would act as an intelligent agent for a person - by reading what the person reads via e-news, having a repository for what was browsed on the net, and being provided with various psychological and social history data about the user (via automated psychological tests and programs). Interactive learning simulations would provide various neural connections and associations between the human user (or "domain expert"), the program, and the world. Thus, a database would be constructed from the individual human user's baseline (or history) and then be updated for as long as the user lives. Furthermore, it would continue to develop via new inputs of continuing world developments on the macroscopic level and family/community updates on the microscopic level. Periodically, the Emmortality Program and the person would be provided with random questions and situations, and a comparison of the Program's responses would be evaluated in comparison to the human's. Program differences within a certain tolerance would be tweaked to better match the user and provide opportunities for the Program to learn. In a sense, it would be an ongoing Turing test, but with benevolent cheating opportunities.
Thus, with the causal impact of ongoing environmental changes post-mortem, there would continue to be input - as the domain expert/user would have established the information drivers pre-mortem. The Program would be able to continue with a robust cognition - both aware of and responsive to change in the world. Psychology instructs us that associative memory (i.e., learning) does not require consciousness, just some good cause-and-effect scenarios.
Within artificial intelligence's domain, Neuron Data in Mountain View, California has a knowledge-based program that exploits various genetic algorithms and neural networking. It is being used for various tasks ranging from detecting bank fraud to making triage decisions in emergency rooms. There is also the Connection Machine, which as Farmer describes it, "is a physical system that is designed to simulate other physical systems...(which begs the question)...is there a threshold of complexity that we have to reach in order for something to behave as though it were alive?" (pp.10-11).
Almost a decade ago NETtalk was created by Sejnowski and Rosenberg. This parallel network program exploited a minuscule 231 "neurons" in a self-organizing algorithm. It taught itself to talk after being provided with rudimentary phonetic elements. Kinoshita and Palevsky described its linguistic development "like a child, the network starts out untrained, and produces a stream of meaningless babble... The continuous stream of babble first gives way to bursts of sound, as the network 'discovers' the spaces between words... After being left to run overnight... NETtalk is talking sense." That was almost a decade ago. Certainly the current ability with non-linear relationships (what these neural networks are made of) is a good starting point. Back-propagation allows for successful machine learning at our current level of sophistication. We can even add the "noise" of additional, non-sequetorial interests that may be subtle but nonetheless contribute to what makes a person unique. Katia Sycana, a professor at Carnegie Mellon University, has developed such noise-makers in a financial decision-making AI program via intelligent, and talkative, autonomous agents. Hardison provides a good tutorial:
The creation of an expert system is analogous to memorizing. Conversely, the learning that occurs in certain kinds of parallel systems is like the programming that the mind seems to do for itself as a result of interaction with the environment during infancy. This is because parallel systems can be designed so that the strengths of connections between their modes are created by the data received. For example, a connection used frequently can have its electrical resistance lowered; one used infrequently or not at all can have its resistance increased. The changes favor one set of connections while interdicting others. The process seems to resemble the creation of associative patterns in the brain. Through the development of these patterns, neural networks can be, to a certain degree, self-organizing, and what is organized is a crude internalization model of a fragment of reality. (p. 310)
You see, self-organizing, parallel systems are the ultimate cybernetic "do-it-yourselfers." Humans design such systems to a finite point, and the machine takes it from there.
Minsky (in The Society of Mind) argues that what occurs in human cyberspace - the mind - results from a culture "of special-purpose units and interdisciplinary controls. If so, many of the basic modules must be created by environmental stimuli that share neuron connections as the (human) brain develops. They are in this sense self-organized, and presumably they often operate in parallel rather than in serial ways" (p. 310).
What kind of machine could run an Emmortality Program? Let's examine the nomological hardware needed to run cognitive processes. Typically accepted contemporary conventions for brain speed are that when 1%-10% of a given brain's neurons are firing at any one time they do so at a rate of about 100 times/second. The speed of such things is measured in FLOPS (Floating Points Operations per Second). This refers to the time needed to add, subtract, divide, or multiply two numbers that are expressed in scientific notation, e.g., if calculating 3.82x10^7 + 4.57x10^6 = 4.28x10^7 takes ten seconds, then ten seconds =1/10 FLOPS. If one neuron is equivalent to one FLOP, then 1% = about 10 gigaFLOPS. If one synapse is equivalent to one FLOP per firing, then 10% is about 10 tera (trillion) FLOPS. About ten million FLOPS would be the upper limit for the power required to simulate a single neuron, therefore 100,000 teraFLOPS would be needed to simulate an entire brain. Moravec's estimate is about 10 teraFLOPS. However, one would not need a whole brain emulated for an Emmortality Program - what's the use of occipital lobes or structures maintaining primitive bodily functions such as respiration, heart rate, glandular activity, etc.?
As for storage, if one neuron codes one bit and the brain has 10^10 neurons, then 10^10 bits would be needed. If every neuron has about 10^5 connections with other neurons, and every connection codes one bit, then the number of synaptic connections in the cortex and cerebellum is 10^15.
Tipler postulates that 10^15 bits processed at a speed of 10 teraFLOPS is a brain. Contemporary machines can handle 10^15 bits, but it's the cruising speed that's the problem. A Cray-2 (c. 1986) could do one gigaFLOP, by 1990 we had 10 gigaFLOP machines. In 1992, Thinking Machines had a super-computer that ran at 100 gigaFLOPS. So-called ultra computers could clock two teraFLOPS just a year later. With current exponential industry advances, a 1000 teraFLOP machine should be here by 2000 or so, and a 10^17 at 100,000 teraFLOPS model wouldn't be too far behind. This bodes well for the Strong AI Postulate and development of an exosomatic brain emulator.
Julian Jaynes presented the bicameral mind as the converse of our "own subjective conscious minds." It is true that the Emmortality Program is not conscious. I echo Jaynes findings that consciousness is not the sine qua non for reconciling experience, or for concept formation, learning, reason, or even thinking. So, then, does the Emmortality Program need consciousness? Not at all. Can a machine or a program be "conscious"? John Searle says "no." I say "not yet."
Of course, there is the risk of the Emmortality Program being better than the domain expert/user. It could evolve into being more empathic and interested in others, wiser, etc. It would certainly be smarter, since it started off that way. But intelligence is not what makes us human, it is not a uniquely human quality. But, then, what is? Some argue consciousness is. So, if the Emmortality Program or another such system was "asked" if it was conscious and it said "yes" how would (or could) one argue it was wrong? The thrill - or horror - of a machine or program thought to be conscious is likely rooted in humanity's narcissistic habit of anthropomorphizing. As can be documented, this dates back to Darwin, Titchener, and Binet, to books such as The Psychic Life of Micro-Organisms and others. Well-intentioned but misplaced sympathies cause projections of consciousness upon everything from protozoa to earthworms.
Hardison notes how we do the same with computers. Doesn't a computer use language in order to reason logically? It understands C++ and Fortran. It has memory (both short-term [RAM] and long-term [ROM]). It plays games and can talk. A computer can get a virus as well as be inoculated. We can now move from jingoistic anthropomorphic metaphors to reality. Maybe the human-mechanistic metaphor is a two-way street. Bi-directionally, one could accurately equate a machine with the body, an information-processing device with the brain, and a program with intelligence or thinking processes. (Aren't bad habits for which we wish to claim no volitional responsibility "hardwired" in?)
Farmer argues that even pure science is anthropomorphic, "when you jump from the Ptolemaic view to the Copernican view of the solar system, you've taken a small step toward making our human view of the universe less anthropomorphic... (but) when we assign magical properties to ourselves, such as intelligence, that we refuse to assign to something else, then I think that as we are confronted with things that are overtly intelligent we will have to begin to accept that they are intelligent." Perhaps this may be the contemporary Copernican shift of humans from the center of a metaphorical universe.
In Understanding Computers and Cognition, Winograd and Flores postulate that computers have an inherent Heideggerian blindness. That is, computers are unable to see the wide spectrum of stimuli that we humans can. It echoes another human-centric philosopher's contention that computers "only deal with facts, but man - the source of facts - is not a fact or set of facts, but a being who creates himself and the world of facts in the process of living in the world" (Hubert Dreyfus). Arguments against to these myopic complaints of cybernetic blindness are:
1) Is such blindness "bad"? Oedipus had to put out his eyes in order to truly see. "Blindness" brings with it decreased distraction as well as insulation from accidental illusion, or purposeful sleight-of-hand. Psychology's school of operationalism would offer that experience/conditioning is the road to perfection;
2) If human abilities are used as a yardstick, then humans are bound to mis-measure and thus mis-perceive or misunderstand silicon (NETtalk is a good example of this), and;
3) As Hardison puts it "the problem is not what computers are in some Platonic sense but how they are perceived, which is closely related to how they are incorporated into the web of human culture" (p. 323).
This leads to another concern: semantics versus syntax. Such is the crux of John Searle's argumentative example of the Chinese Room, in that following certain procedures can produce correct results, but doing so does not prove knowledge or learning, and most certainly not consciousness.
Certainly a computer does not know what "mom" means in the same way a person recalls the meaning of the term. Thus, the computer has the syntax down pat but is void of semantics. Good prose doesn't equal good poetry is Searle's point. It parallels Winograd and Flores's blindness concept. Hardison's solution to this supposed dilemma "...is interesting because we have no way of knowing about the subjectivity of anything except by what we observe. If somebody said, 'I am conscious,' and you replied, 'I can't prove you are not conscious, but I know you are unconscious anyway,' your attitude would seem a bit churlish. Since you honestly don't know what is going on in the head of the person who says, 'I am conscious,' you have to take the person's word for it. Who, after all, knows better than the person whether or not he is conscious? Who knows, really, whether anybody is conscious in Searle's sense? Who knows what thought is? Perhaps consciousness is a matter of procedures - a syntax - and semantics is an illusion created by the syntax" (p. 331). Perhaps this is Searle's attempt to put a new spin on credo quia impossible?
So what's "better" and who's to say? If a computer is blind (a la Winograd and Flores) to what it is to be human, we are similarly impaired when it comes to an empathic understanding of what it means to be silicon. We may never know if a machine is conscious in the way we conceptualize it and in the way we perceive/believe our senses to be, and vice-versa. Maybe it is simply "a semantic quibble: machines cannot acquire human abilities because they are machines."
Cognitive psychology has long noted the fact that humans rarely receive data at 100% accuracy. When various types of noise get in the way - especially emotional contaminants - it is referred to as an "apperception." At this non-emotive stage of computer evolution such would be absent. But this lack may be problematic, as it would subtract a helpful human-like realism. Sycara's work may add some helpful connections with some fabricated noise, and perhaps along with non-entropic chaos or even digitized apperceptions.
It is unusual to think of having accurate data/memory as a problem. Furthermore, our brains likely do not store memories or all that it remembers. The human body is constantly deteriorating and being renewed. Every 10 to 23 seconds our personal, subatomic quarks and gluons (the stuff of neutrons and protons - which make up one's atoms) "die" and are replaced. Aging is a cruel example of the inexactness of such replications. And brain cells don't even replicate. "What probably happens (with human memory) is that any part of a memory is stored and recall of the entire memory involves the inverse of data compression techniques: the memory is 'flushed' out. Such a storage mechanism is probabilistic; errors can be made in this process, so perfectly sane people occasionally 'remember' events that never occurred" (Tipler, p. 237). I would offer that probabilistic occurrences can be coded into the Emmortality Program quite easily. Randomizers would not be difficult to introduce.
Turing suggested that the inclusion of "random elements" would be a necessary aid in the development of a computational machine that would pass his test. As Tipler put it, "...although deterministic algorithms may exist to solve a problem, often these require such an enormous amount of computer capacity that systematic 'guessing' - making choices among equally weighted possibilities at random - is almost always more efficient" (p. 194). This has led to the concept of "heuristic programming."
In human decision-making random selection may be most efficient due to the "information cost" phenomenon. That is, it may take an inordinate amount of resources to determine a solution by exhaustively examining all relevant data prior to arriving at a solution. Thus avoiding the Buridan's Ass dilemma in that a random choice is better than indecision. Game theories would also tend to support this. Heuristic programs themselves use a pseudo-random number generator to act then as a randomizer. [A pseudo-random number is produced by a deterministic algorithm, but it is so complex that one cannot tell the difference between a pseudo-random number and an actual random number.]
Quantum non-locality (aka the No Clone Theorem) would prevent the Emmortality Program from true human emulation because of its inability to mimic or download quantum mechanically entwined human relationships and such inputs. Many assume that human life should not be considered a quantum state. But my point is that the Emmortality Program is a mimicked version of its user, not the user her or himself. (The Bekenstein Bound would actually support an even more radical postulate [a la Tipler] that "...using computer memory capability of the amount indicated by the Bekenstein Bound, a computer simulation of a person, a planet, [or even] a visible universe will not merely be very good, it will be perfect [emphasis in original], it will be an emulation" (p. 223). [It is beyond the limitations of this article to venture off into discussions about emulated quarks and their capability to reconstitute ontological free will, as has been done elsewhere.])
Forgetting can be quite therapeutic for humans in certain circumstances. It can be a powerful analgesic (if not opiate). It can help free people from non-productive fixations. The paradox is being able to choose what is forgotten. The Emmortality Program then raises a new cosmological conundrum: Is it desirable to keep interacting with the dead? Psychics and mediums have supposedly been doing it for centuries. There are possible concerns that merit examination, but the occult will be omitted.
Certainly monuments, mausoleums, Indian burial mounds, memorial wings in buildings, endowments, various statues, and other such technologies and symbols act to preserve the memory of the dead. So do paintings, audio and video recordings, photographs, and letters. Perhaps the closure and finality of death is not so absolute as one may think within many cultures. Certainly, Judaic, Christian, and Islamic traditions would concur.
It is Ryan's opinion regarding the examination of such issues vis-a-vis video "...that only in the context of... a new (cosmology) can we invent stable (and suitable) rituals that allow us to replay the dead live...Given the flexibility of electronic information technologies, we have the possibility of flexing the story in a non-narrative way that avoids the patterns of dominance associated with egocentric 'master narratives.' Another way of saying this is that we encode cosmology in a way that it is sensitive to chaos and responsive to local knowledge" (pp. 225-226).
Computers are simulators - they simulate typewriters, musical instruments, drawing boards, etc. A simulation, digitally speaking, is a model of bits arranged in a pattern that mimics the object/ procedure in question. This arranged code yields what one recognizes as a program. Running a program is analogous to putting the model into action (e.g., typing with a word processor). An emulation is a perfectly modeled simulation of space or task. The Bekenstein Bound would support that, with adequate computer power, a person (or at least the mimicked bits that make up a person) could indeed be emulated. But one need not go to this extreme as simulated levels would be quite adequate.
Hypothetically, if there were a working Emmortality Program - at an emulation level, would an emulated mind then "exist"? That's for Descartes to determine. Such scenarios conjure philosophic conundrums, such as "how does one know he/she is not already an emulation?". Leibniz may offer help in this matter, via his "Identity of Indiscernibles" rule. That is, "...entities which cannot be distinguished by any means whatsoever, even in principle, at any time in the past, present, and future have to be considered identical" (p. 208).
Was Hans Moravec right? Will humans disappear into these machines, perhaps via the route offered by an Emmortality Program? A better conceptualization is that humans will reappear out of such machines. Silicon is already immortal. Perhaps it is time we move outside of ourselves to see what we can learn from it.
Ballard, J.G., "Project for a glossary of the Twentieth Century", in J. Crary & K. Winter, eds., Incorporations. New York: Zone, 1992.
Dreyfus, H.L., What Computers Can't Do: A Critique of Artificial Reason. New York: Harper & Row, 1972.
Hardison, O.B., Disappearing Through the Skylight. New York: Viking, 1989.
Minsky, M., The Society of Mind. New York: Simon & Schuster, 1986.
Stone, A.R., "Virtual Systems", in J. Crary & K. Winter, eds., Incorporations. New York: Zone, 1992.
Tipler, F.J., The Physics of Immortality. New York: Doubleday, 1994.
Wiener, N., Cybernetics: Control and Communication in the Animal and the Machine. Boston: MIT Press, 1948.