The Endless Frontier and The Thinking Machine Hans P. Moravec Artificial Intelligence Lab Computer Science Dept. Stanford University Stanford, Ca. 94305 Copyright 1978 by Hans P. Moravec All Rights Reserved The first modern computers, appearing in the late 1940's, offered unprecedented opportunities for experiments in complexity. They raised the hope in influential scientists like John von Neumann and Alan Turing that the ability to think, our greatest asset in dealing with the world, might soon be understood well enough to duplicate. Our minds might be amplified just as our muscles had been by energy machines. Computers have become vastly more capable since then, but are still much stupider than people in most areas. Not for lack of trying. The last decade, in particular, has seen thousands of people years devoted directly to making them smarter. Much effort has been expended on computer programs which do mathematics, computer programming and common sense reasoning, are able to understand natural languages and interpret scenes seen through cameras and spoken language heard through microphones and to play games humans find challenging. There's been some progress. Arthur Samuel's 20 year old checker program occasionally beats checker champions, and chess programs have played at good amateur (class B) level for nearly as long. In August of 1978 a chess program written by David Slate and Larry Atkin of Northwestern University won one serious game and tied another against David Levy, a chess International Grandmaster. It happened in a tournament to settle a 1968 bet between Levy and several computer scientists that a computer couldn't beat him within a decade. Levy won the tournament and the bet, but by a narrower margin than he expected. An earlier version of the Northwestern program won the Minnesota Open Chess Championship in March 1977, earning a chess rating in the expert range. The program owes much of its success to the computer it runs on, a CDC Cyber 176. Executing 14 million instructions per second it is about 10 times faster than prior machines used to play chess. A ten year effort at MIT has gathered specialized knowledge about algebra, trigonometry, calculus and related fields from many sources into a wonderful program called MACSYMA. MACSYMA manipulates symbolic formulas the way pocket calculators manipulate numbers. With it a person can, for instance, solve a differential equation without thinking about the mechanics of doing an integral in the same way that someone with a pocket calculator can find the quotient of two numbers without knowing about long division. MACSYMA has been used by many to solve problems that would otherwise have been left untouched. Other semi-intelligent programs showing up in the real world can understand simplified typewritten English about restricted subjects, make elementary deductions in the course of answering questions and interpret spoken commands chosen from hundred word repertoires. Some can do simple visual inspection tasks, such as deciding whether or not a screw is at the end of a shaft. Computers are at their worst trying to do the things most natural to humans, like seeing, hearing, language and common sense reasoning. Large portions of our nervous systems are dedicated to these skills, and computers are unlikely to match human performance until they can process as much data as the neural centers. I calculate that a typical present day computer is a million times less powerful than a human brain. The computer can perform a million simple steps per second, but the 40 billion neurons in a brain can switch a thousand times in the same interval. The spectacular, continuing evolution of microelectronics should make the requisite power available at great cost in a decade, and inexpensively by the end of the century. In addition to being powerful enough, a computer able to perform like a human must be properly programmed. Just how long before the right programs are written is a matter of controversy. Many critical experiments can't be done now because computers are too small and too slow. I feel that infusing an adequately powerful machine with near-human intelligence is not as hard as some guess. Evolution has independently produced moderate intelligence several times, indicating that it develops naturally and does not require improbable leaps. The difficulty can be gauged by the evolutionary timescale. 300 million years passed between the invention of the neuron and development of the worm, a similar period between worms and dumb vertebrates, and 300 million more between vertebrates and us. Our technology has paralleled nature in many respects. It had developed electronic switching by 1920, electronic computers by 1950, and semi-intelligent machines by 1980. By analogy we ought to have truly smart machines by 2010. Human equivalent computers will have a profound effect on the nature and pace of space colonization. People are not built to live in space, but must be supported by massive, Earth simulating machinery to remain healthy. Robots can be made without this handicap. Tiny space probes cruise unprotected through the Solar System, their functioning unimpaired by vacuum, radiation and temperature extremes. Machines with these advantages and with human or above human intelligence have an overwhelming edge in space industrialization. We may be able to keep up, but only by profoundly changing ourselves. Natural Intelligence My optimism about the future of intelligent machines is based partly on the evolutionary record. Nature holds the patents on high intelligence. It invented it not once, but several times, as if to demonstrate how easy it was. A billion years ago, before brains or eyes were invented, when the most complicated animals were something like hydras, double layers of cells with a primitive nerve net, our progenitors parted company with the invertebrates. Now both clans have intelligent members. Cephalopods are the most intellectual invertebrates. Most mollusks are sessile shellfish, but octopus and squid are highly mobile, with big brains and excellent eyes. Evolved independently of us, they are different. The optic nerve connects to the back of the retina, so there is no blind spot. The brain is annular, a ring around the esophagus. The green blood is circulated by a systemic heart oxygenating the tissues and two gill hearts moving depleted blood. Hemocyanin, a copper doped protein related to hemoglobin and chlorophyll, carries the oxygen. Octopus and their relatives are swimming light shows, their surfaces covered by a million individually controlled color changing cells. A cuttlefish placed on a checkerboard can imitate the pattern, a fleeing octopus can make deceiving seaweed shapes coruscate backward along its body. Photophores of deep sea squid, some with irises and lenses, generate bright multicolored light. Since they also have good vision, there is a potential for high bandwidth communication. Their behavior is mammal like. Octopus are reclusive and shy, squid are occasionally very aggressive. Small octopus can learn to solve problems like how to open a container of food. Giant squid, with large nervous systems, have hardly ever been observed except as corpses. They might be as clever as whales. Time before present Representative Creatures Significant events 0 (you are here) | | | | | computers massive technology 2.5 million years | | | | | | 10 | | | | elephants | tool use | | | whales | primates 40 | | | | | | | | | | | | 90 octopus squid | | | | | | | +-----+-----+ 160 +---+---+ birds mammals | | | learned behavior 250 early squid +------+------+ warm bloodedness | reptiles 360 | | cephalopods fish | 490 | | amphibians land vertebrates +---+ +----+---+ 640 mollusks vertebrates | | 810 | | complex nerve centers +------+------+ 1 billion years | invention of the neuron | old age death 1.21 | sex in animals perfected | 1.44 | multi-cellular animals animals 1.69 | plants | 1.96 | | oxygen to support animals +----+ 2.25 | | 2.56 blue-green | nucleated cells algae | 2.89 +-------+ | DNA genetics? 3.24 | photosynthesis earliest cells reliable reproduction 3.61 | invention of the cell | inorganic protein microspheres 4 billion years non-living chemicals amino acid formation FIGURE: Highlights in the evolution of terrestrial intelligence. The distance along the edge of the tree is proportional to the square root of the time from the present. This seems to space things nicely. Birds are vertebrates, related to us through a 300 million year old, probably not very bright, early reptile. Size-limited by the dynamics of flying, some are intellectually comparable to the highest mammals. The intuitive number sense of crows and ravens extends to seven, compared to three or four for us. Birds outperform all mammals except higher primates and the whales in "learning set" tasks, where the idea is to generalize from specific instances. In mammals generalization depends on cerebral cortex size. In birds forebrain regions called the Wulst and the hyperstriatum are critical, while the cortex is small and unimportant. Our last common ancestor with the whales was a primitive rat-like mammal alive 30 million years ago. Some dolphin species have body and brain masses identical to ours, and have had them for more generations. They are as good as us at many kinds of problem solving, and can grasp and communicate complex ideas. Killer whales have brains seven times human size, and their ability to formulate plans is better than the dolphins', who they occasionally eat. Sperm whales, though not the largest animals, have the world's largest brains. Intelligence may be an important part of their struggle with large squid, their main food. Elephant brains are five times human size. Elephants form matriarchal tribal societies and exhibit complex behavior. Indian domestic elephants learn over 500 commands, and form voluntary mutual benefit relationships with their trainers, exchanging labor for baths. They can solve problems such as how to sneak into a plantation at night to steal bananas, after having been belled (answer: stuff mud into the bells). And they never forget (really). Apes are our 10 million year cousins. Chimps and gorillas can learn to use tools and to communicate in human sign languages at a retarded level. Chimps have one third, and gorillas one half, human brainsize. Nervous System Size and Intelligence Animals exhibiting near-human behavior have hundred billion neuron nervous systems. Imaging vision alone requires a billion. The smartest insects have a million brain cells, while slugs and worms make do with a thousand, and sessile animals with a hundred. The portions of nervous systems for which tentative wiring diagrams have been obtained, including nearly all of the large neuroned sea slug, Aplysia, the flight controller of the locust and the early stages of vertebrate vision, reveal neurons configured into efficient, clever, assemblies. Measuring Processing Power The vertebrate retina has been studied extensively. Its 20 million neurons take signals from a million light sensors and combine them in a series of simple operations to detect things like edges, curvature and motion. Then image thus processed goes on to the much bigger visual cortex in the brain. Assuming the visual cortex does as much computing for its size as the retina, we can estimate the total capability of the system. The optic nerve has a million signal carrying fibers and the optical cortex is a thousand times deeper than the neurons which do a basic retinal operation. The eye can process ten images a second, so the cortex handles the equivalent of 10,000 simple retinal operations a second, or 3 million an hour. An efficient program running on a typical computer can simulate a retinal operation in about two minutes, for a rate of 30 per hour. Thus seeing programs on present day computers seem to be 100,000 times slower than vertebrate vision. Another measurement To compare the processing power of brains and computer programs more generally, note that devices which compute (or think) do things unexpectedly. Predictable entities like rocks do no computation, pocket calculators do a little, bees do a lot, and humans do even more. By this criterion the amount of computing done by a device is in the mind of the beholder. If you were very good at mental arithmetic and could predict a calculator's answer before it gave it, the calculator would do no real computation for you, and might as well be a rock. Information theory can make this idea precise. If an entity in a given state can change to one of N next states with equal probability, the information in the transition, which I will call the Compute Energy, is given by Compute Energy = log2 N where N is the number of next states. The measure is in binary digits, bits. Similarly the total compute energy of the entity is the log of the number of distinct states it can ever be in. Some people call this the memory size. A machine that computes faster is more powerful than a slower one. Compute Power is found by dividing the compute energy of a state transition by the time required for the transition. Compute Power = log2 N / t The units are bits/second. Slightly more complicated formulas, which give lower values, apply if the transitions probabilities and times are not all equal. These measures are highly analogous to the energy and power capacities of a battery. The compute power and energy of a system of two or more independent machines is the sum of the individual powers and energies. A device with a high power, able to reach a moderate number of states in a short time, can yet have a low energy, if the total number attainable in the long run is not high. Speeding up a machine by a factor of n increases the power by the same factor. A completely predictable system has zero power and energy. Computer Power The PDP-10 computer used by many researchers obeys simple instructions at the rate of one million per second. An instruction contains one of 2^5 different commands, involving one of 2^4 accumulators and one of 2^18 memory locations, most of these combinations resulting in distinct next sates. This gives a Compute Power of log2 (2^5 x 2^4 x 2^18) bit/(10^-6 sec) = 27 x 10^6 bit/sec The power is reduced because different instruction sequences can result in the same outcome and increased by information flowing in from high speed storage devices connected to the computer for a net of about 10^7 bit/sec. The power is also limited by the total compute energy, which is about 10^7 bits. The PDP 10 could execute at its maximum effectiveness for one second before reaching a state which could have been arrived at more quickly another way. Connecting the computer to the external world can increase this time indefinitely. 14| sperm whale * 10 | | human * 13| chimp * 10 | | human vision * 12| 10 | | 11| proposed NASA wind + 10 | tunnel simulator | 10| 10 | | 9 | Cray 1 + 10 | bee * | CDC 7600, IBM 360/195 + 8 | 10 | Power | 7 | PDP-10 + bit 10 | --- | sec 6 | 10 | slug * | 5 | 10 | * living sponge | 4 | 10 | | 3 | 10 | + pocket calculator | +--------------------------------------------------------------------------- 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 Energy (bits) FIGURE: Compute Power and Energy of various devices. Scales are logarithmic. The Cray machine is an extremely fast and large scientific computer. The NASA simulator would probably be a general purpose computer 100 times as powerful as the biggest existing machines. It has not been designed yet. Brain Power Making a number of simplifying assumptions, the human brain has 40 billion neurons, each able to change state 1000 times a second. Considering the space of all possible interconnections of these 40 billion (treating this as the search space natural evolution had, in the same sense that all possible programs are available to someone trying to make a computer smart), we note that the number of combinations reachable from a given state is 2^(40x10^9), giving a compute power of 40 x 10^9 bit / 10^-3 sec = 40 x 10^12 bit/sec which is about a million times more than a PDP-10 program. This ratio agrees with the visual system calculation, because the visual cortex is about ten percent of the brain. We got our figure by comparing all possible brain interconnections with all possible PDP-10 programs, measuring the relative computational richness of the media available to nature and a programmer. We concluded that a million PDP-10's would be needed to support a program that thinks like a human. But a computer can be reprogrammed more conveniently than a brain can be rewired, and this is a source of hidden power. The human-equivalent program on a million machine system could be replaced by single-minded programs to play superhuman chess or blindingly solve differential equations, an option not available with a real human. This generality has a price. The PDP-10 computer is just one instance of all possible ways of connecting its components. The extra options give the raw circuitry about 1000 times the compute power of the finished computer. But there's a tradeoff between efficiency and design effort. It's much easier, faster and cheaper to write a program than to build a circuit for a complex function. Computer instructions do useful things with few unexpected side effects. In circuit design the basic operations are more primitive, and the side effects many. A program that can be written in a few days by one person may take months to implement in special hardware. It's for this reason that the best machine chess player is a general purpose computer rather than a special gadget. I imagine that the first near-human machines will similarly be implemented on general purpose computers, and converted later into more efficient hardware. The Growth of Processing Power A millionfold increase in processing power seems a tall order, but the gap is not quite as wide as our simple analysis implies. PDP-10 computers, still used, are a decade old. Today's similarly priced machines are ten times as fast. And for ten million dollars one can buy scientific computers with more than 100 times the speed. 10,000 of these supercomputers operating together with the right program should be as intelligent as a human. At a hundred billion dollars, it's an offer easily refused. The cost of computing has dropped steadily since the 1950's by a factor of ten every five years. Each price drop expanded the market and fueled further advances. In the last few years computers have crept into the consumer area, disguised as calculators, watches, smart appliances and automotive electronics. Computer use will continue to expand in the forseeable future. The conventional applications of micro-computers will continue to grow. Calculators and computer games will merge with television and the phone system to become a universal information utility, giving bidirectional access to the world's knowledge. Until now assembly line robots have been deaf, blind and very dumb. Like the sorcerer's apprentice's broomsticks, they repeated a sequence of motions over and over, oblivious to the world. Computers, seeing through cameras and feeling with switches and strain gauges, are begining to provide the senses and the sense robots have lacked. As they become more capable and cheaper, and human labor becomes more expensive, they will be increasingly in demand. Not only the number of computers, but the capability of each one, will grow. Homes are less structured than assembly lines, and home robots are more difficult to make, but they will appear, and be in great demand, when computers become smart enough. They will be cheap partly because other robots will assemble them. Outer space is prime grazing land for machinery. Robot mines and robot factories making more robots on the moon or in the asteroids ought to be a fast growing and lucrative investment, once the machines are good enough. Integrated circuit technology is nowhere near its physical limits, and the science to support a few decades of continuing improvement already exists. In production are new semiconductor techniques, I^2L, ten times as efficient as older TTL, and super fast D-MOS, CCD for large sensors and fast bulk memory, and magnetic bubbles for mass storage. The new 65K memory chips use V-MOS, where transistors trade depth in the silicon for surface area, pointing towards much denser three dimensional circuits. The structures on IC surfaces are becoming smaller than one micron, and electron beams and X-rays are displacing longer wavelength light in their manufacture. Industrial and university labs are full of potential major improvements. Exotic semiconductors like gallium arsenide can be made into IC's several times faster than silicon. Thermal disturbances show up as electrical noise, and the amount of energy needed to unambiguously signal a circuit is proportional to its temperature. Cooling semiconductor circuits to liquid nitrogen temperatures allows them to deal with much smaller signals, and consequently be smaller and less power hungry, and up to 100 times faster. If we go to 4 degree absolute liquid helium temperatures, even more astounding things are possible. A Josephson junction gate is a superconducting logic element in which the magnetic field of one current can switch off another supercurrent tunneling through a very thin insulator. The transition takes as little as a picosecond, 1000 times faster than existing semiconductors. Josephson junctions use 1000 times less energy per switching function than neurons. New discoveries, unpredictable in detail, will further fuel the fire. Magnetic monopoles weighing 1000 protons, as yet undiscovered because they are so massive and hard to make, could drastically expand our options. Matter made of monopole atoms would be extremely dense and strong, and solid state magnetic circuits could be millions of times faster than conventional electronics. The lowest transition in a magnetic atom is so energetic that monopole wire would magnetically superconduct at the plasma temperatures of conventional matter. The cost of computing will continue to fall, by a factor of 10,000 in the next 20 years, and a ten milion dollar general purpose computer powerful enough to be programmed for human equivalence will be exist by the 21st century. But that's not the whole story. In measuring the processing power of the PDP-10 we assumed it could do nothing but obey its instructions. Processing power expresses the possibilities per unit time, and this restriction was expensive. If we instead think about rewiring the computer's circuitry, the number of options is much greater, and the processing power is about 1000 times as large. Or, carrying it to absurdity, if we considered how the atoms of the PDP-10 could be reconfigured, the compute power would be astronomical. Hardware design is becoming more automated. Most new computers, and all new integrated circuits, were designed with the help of other computers. Programs exist that make printed circuit or IC patterns from circuit diagrams, and others can simulate designs and check for errors. As computers get smarter these automatic design aids will get better, and the complexity bounds on special hardware will rise. Some specialized operations useful for human equivalence are already in hardware. Hughes has built an IC that locates edges in TV images a hundred times faster than typical programs. Some air traffic control systems have associative memory units able to look up data dozens of times more quickly than an unaided computer. Within twenty years this kind of development, coupled with the general cost reductions, should allow human equivalent machines costing a year's salary. Even before then idiot savant machines like the Northwestern/CDC chess computer mentioned earlier, better than humans at a few things, though stupider on the whole, will be slaving for our short term well being, and long term obsolescence. It has already begun. The Future What happens when increasingly cheap machines can replace humans in any situation? What will I do when a computer can write this article, and do research, better than me? These questions face some occupations now. They will affect everybody in a few decades. By design, machines are our obedient and able slaves. But intelligent machines, however benevolent, threaten our existence because they are alternative inhabitants of our ecological niche. Machines merely as clever as human beings will have enormous advantages in competitive situations. Their production and upkeep costs less, so more of them can be put to work with given resources. They can be optimized for their jobs, and programmed to work tirelessly. Intelligent robots will have even greater advantages away from our usual haunts. Very little of the known universe is suitable for unaided humans. Only by massive machinery can we survive in outer space, on the surfaces of the planets or on the sea floor. Smaller, intelligent but unpeopled, devices will be able to do what needs to be done there more cheaply. The Apollo project put people on the moon for forty billion dollars. Viking landed machines on Mars for one billion. If the Viking landers had been as capable as humans, their multi-year stay would have told us much more about Mars than we found out about the moon from Apollo. As if this weren't bad enough, the very pace of technology presents an even more serious challenge. We evolved with a leisurely 100 million years between significant changes. The machines are making similar strides in decades. The rate will quicken further as multitudes of cheap machines are put to work as programmers and engineers, with the task of optimizing the software and hardware which makes them what they are. The successive generations of machines produced this way will be increasingly smarter and cheaper. There is no reason to believe that human equivalence represents any sort of upper bound. When pocket calculators can out-think humans, what will a big computer be like? We will simply be outclassed. Then why rush headlong into the intelligent machine era? Wouldn't any sane human try to delay things as long as possible? The answer is obvious, if unpalatable on the surface. Societies and economies are as surely subject to evolutionary pressures as biological organisms. Failing social systems wither and die, to be replaced by more successful competitors. Those that can sustain the most rapid expansion dominate sooner or later. We compete with each other for the resources of the accessible universe. If automation is more efficient than hand labor, organizations and societies which embrace it will be wealthier and better able to survive in difficult times, and expand in favorable ones. If the U.S. were to unilaterally halt technological development, as a vociferous minority urges, it would soon succumb either to the military might of the Soviets, or the economic success of its trading partners. Either way the social ideals which led to the decision would become as unimportant on the world scale as the opinions of the religious cults. If, by some evil and unlikely miracle, the whole human race decided to eschew progress, the long term result would be almost certain extinction. The universe is one random event after another. Sooner or later an unstoppable virus deadly to humans will evolve, or a major asteroid will collide with the earth, or the sun will go nova, or we will be invaded from the stars, or a black hole will swallow the galaxy. The bigger, more diverse and competent a culture is, the better it can detect and deal with external dangers. The bigger events happen less frequently. By growing sufficiently rapidly it has a finite chance of surviving forever. Even the eventual collapse or heat death of the universe might be evaded or survived if an entity can restructure itself properly. The human race will expand into the solar system soon, and Gerry O'Neill's little Earths will be part of it. But the economics of automation will become very persuasive in space even before machines achieve human competence. I visualize immensely lucrative self-reproducing robot factories in the asteroids. Solar powered machines would prospect and deliver raw materials to huge, unenclosed, automatic processing plants. Metals, semiconductors and plastics produced there would be converted by robots into components which would be assembled into other robots and structural parts for more plants. Machines would be recycled as they broke. If the reproduction rate is higher than the wear out rate, the system will grow exponentially. A small fraction of the output of materials, components, and whole robots could make someone very, very rich. The first space industries will be more conventional. Raw materials purchased from Earth or from human space settlements will be processed by human supervised machines and sold at a profit. The high cost of maintaining humans in space insures that that there will always be more machinery per person there than on Earth. As machines become more capable, the economics will favor an ever higher machine/people ratio. Humans will not necessarily become fewer, but the machines will multiply faster. When humans become unnecessary in space industry, the machines' physical growth rate will climb. When machines reach and surpass humans in intelligence, the intellectual growth rate will rise similarly. The scientific and technical discoveries of super-intelligent mechanisms will be applied to making themselves smarter still. The machines, looking quite unlike the machines we know, will explode into the universe, leaving us behind in a figurative cloud of dust. Our intellectual, but not genetic, progeny will inherit the universe. Barring prior claims. This may not be as bad as it sounds, since the machine civilization will certainly take along everything we consider important, including the information in our minds and genes. Real live human beings, and a whole human community, could be reconstituted if an appropriate circumstance ever arose. Since we are biologically committed to personal death, immortal only through our children and our culture, shouldn't we rejoice to see that culture become as robust as possible? An Alternative Some of us have very egocentric world views. We anticipate the discovery, within our lifetimes, of methods to extend human lifespans, and look forward to a few eons of exploring the universe. We don't take kindly to being upstaged by our creations. The machines' major advantage is their progress rate. We evolve by DNA + nucleated cell + sex + personal death, they develop by the much faster intelligence + language + culture + science + technology technique. If we could somehow learn the new way, we might be able hold our own. Genetic engineering is an option. Successive generations of human beings could be designed by mathematics, computer simulations, and experimentation, like airplanes and computers are now. But this is just building robots out of protein. Away from Earth, protein is not an ideal material. It's stable only in a narrow temperature and pressure range, is sensitive to high energy disturbances, and rules out many construction techniques and components. Anyway, second rate superhuman beings are just as threatening as first rate ones, whatever they're made of. What's really needed is a process that gives an individual all the advantages of the machines, at small personal cost. Transplantation of human brains into manufactured bodies has some merit, because the body can be matched to the environment. It does nothing about the limited and fixed intelligence of the brain, which the artificial intellects will surpass. Transmigration You are in an operating room. A robot brain surgeon is in attendance. By your side is a potentially human equivalent computer, dormant for lack of a program to run. Your skull, but not your brain, is anaesthetized. You are fully conscious. The surgeon opens your brain case and peers inside. Its attention is directed at a small clump of about 100 neurons somewhere near the surface. It determines the three dimensional structure and chemical makeup of that clump non-destructively with neutron tomography, phased array radio encephalography, and ultrasonic radar. It writes a program that models the behavior of the clump, and starts it running on a small portion of the computer next to you. Fine wires are run from the edges of the neuron assembly to the computer, providing the simulation with the same inputs as the neurons. You and the surgeon check the accuracy of the simulation. After you are satisfied, tiny relays are inserted between the edges of the clump and the rest of the brain. Initially these leave brain unchanged, but on command they can connect the simulation in place of the clump. A button which activates the relays when pressed is placed in your hand. You press it, release it and press it again. There should be no difference. As soon as you are satisfied, the simulation connection is established firmly, and the now unconnected clump of neurons is removed. The process is repeated over and over for adjoining clumps, until the entire brain has been dealt with. Occasionally several clump simulations are combined into a single equivalent but more efficient program. Though you have not lost consciousness, or even your train of thought, your mind (some would say soul) has been removed from the brain and transferred to a machine. In a final step your old body is disconnected. The computer is installed in a shiny new one, in the style, color and material of your choice. You are no longer a cyborg halfbreed, your metamorphosis is complete. Advantages become instantly apparent. Your computer has a control labelled speed. It had been set to slow, to keep the simulations synchronized with the old brain, but now you change it to fast. You can communicate, react and think a thousand times faster. But that's just a start. The program in your machine can be read out and altered, letting you conveniently examine, modify, improve and extend yourself. The entire program may be copied into similar machines, giving two or more thinking, feeling versions of you. You may choose to move your mind from one computer to another more technically advanced, or more suited to a new environment. The program can also be copied to some future equivalent of magnetic tape. If the machine you inhabit is fatally clobbered, the tape can be read into an blank computer, resulting in another you, minus the experiences since the copy. With enough copies, permanent death would be very unlikely. As a computer program, your mind can travel over information channels. A laser can send it from one computer to another across great distances and other barriers. If you found life on a neutron star, and wished to make a field trip, you might devise a way to build a neutron computer and robot body on the surface, then transmit your mind to it. Nuclear reactions are a million times quicker than chemistry, so the neutron you can probably think that much faster. It can act, acquire new experiences and memories, then beam its mind back home. The original body could be kept dormant during the trip to be reactivated with the new memories when the return message arrived. Alternatively, the original might remain active. There would then be two separate versions of you, with different memories for the trip interval. Two sets of memories can be merged, if mind programs are adequately understood. To prevent confusion, memories of events would indicate in which body they happened. Merging should be possible not only between two versions of the same individual but also between different persons. Selective mergings, involving some of the other person's memories, and not others would be a very superior form of communication, in which recollections, skills, attitudes and personalities can be rapidly and effectively shared. Your new body will be able to carry more memories than your original biological one, but the accelerated information explosion will insure the impossibility of lugging around all of civilization's knowledge. You will have to pick and choose what your mind contains at any one time. There will often be knowledge and skills available from others superior to your own, and the incentive to substitute those talents for yours will be overwhelming. In the long run you will remember mostly other people's experiences, while memories you originated will be floating around the population at large. The very concept of you will become fuzzy, replaced by larger, communal egos. Mind transferral need not be limited to human beings. Earth has other species with brains as large, from dolphins, our cephalic equals, to elephants, whales, and giant squid, with brains up to twenty times as big. Translation between their mental representation and ours is a technical problem comparable to converting our minds into a computer program. Our culture could be fused with theirs, we could incorporate each other's memories, and the species boundaries would fade. Non-intelligent creatures could also be popped into the data banks. The simplest organisms might contribute little more than the information in their DNA. In this way our future selves will benefit from all the lessons learned by terrestrial biological and cultural evolution. This is a far more secure form of storage than the present one, where genes and ideas are lost when the conditions that gave rise to them change. Our speculation ends in a super-civilization, the synthesis of all solar system life, constantly improving and extending itself, spreading outwards from the sun, converting non-life into mind. There may be other such bubbles expanding from elsewhere. What happens when we meet? Fusion of us with them is a possibility, requiring only a translation scheme between the memory representations. This process, possibly occuring now elsewhere, might convert the entire universe into an extended thinking entity, a prelude to even greater things.