CHAPTER 1 – ESCAPE VELOCITY
“… change sculpted our universe and our society …” By almost any measure society is changing faster than ever “a statement true for at least half a millennium, and mostly true since the agricultural revolution and the invention of writing over five thousand years ago.” This accelerated pace continues because of the products of technology which further speed up the process; humans struggle to keep pace: “the lessons of a technical education are often obsolete before the education is complete.”
If you rub wood is will get warm but will soon cool down—a rule for our ancestors—unless you rub it so hard it ignites—escape velocity (EV). Similarly our machines will achieve escape velocity and the rules will no longer apply. In Moravec’s (M) words: “the wood is already smoldering.”
M says we are like riders in an elevator who forget how high we are until we get an occasional glimpse of the ground—as when we meet cultures frozen in time. Then we see how different our world is compared to the one we adapted to biologically. For all of human history culture was secondary to biology, but about 5000 years ago things changed, as cultural evolution in the form of memes became far and away the most important means of evolution for humans. Today we are reaching the EV from our biology. But the world we will produce will be “unlike the villages, fixed and nomadic, in which human behavior evolved, …”
There is a mismatch between “our stone-age biology and our information-age…” This is apparent when you consider how long it takes to become specialized in the difficult and esoteric work we do. Still, despite our misgivings about contemporary life, few of us would be prepared to be stone-aged forest dwellers. Besides, M argues that we have substitutes for our tribal group in competitive sports, outdoor vacations, and BBQs. Certainly some groups reject all this, yet the ubiquitous nature of industrial society and the benefits it offers—medicine, food, clothing, etc—seem to be preferred by most. This disenchanted “are outvoted by the demands of billions for food, housing, and civilized comforts.”
As machine productivity rises, humans are left with less physical labor and more leisure time—which M says we can use to satisfy our hunter-gatherer instincts if we want. Furthermore, a real green revolution will be possible when we are sufficiently wealthy and technology sophisticated enough to move, for example, production to outer space. M has done a detailed analysis that shows that pollution increases as wealth increases since personal wants outweigh communal concerns—to a point. When wealth reaches a certain point, communal concerns are affordable and persons pay for them, “wealth increases the options available to individuals.”
CHAPTER 2 – CAUTION! ROBOT VEHICLE!
M begins by reminding us of how difficult robotics is. He reiterates the point that he has made several times—humans find glancing at a chessboard and seeing the pieces easy; while it takes much thought to make a good move. Machines have the opposite problem.
Cybernetics is “the science of control and communication in the animal and the machine.” Scientists have made artificial nervous systems and insect robots under this banner. The field began in the 1940s but by the 1960s challenging problems like building reading machines were stumping the field. The development of computers suggested a different approach to thinking machines. Alan Turing’s computer cracked the German code in WWII and later speculated about the development of intelligent machines. John von Neumann picked up where Turing left off and by the 1950s the idea of artificial intellects was in the air. The term “AI” was coined in 1956 and “Logic Theorist,” the first working program of AI, proved many of the theorems in Russell & Whitehead’s Principia Mathematica. But most of these programs were not very good and proved theorems no better or faster than a college freshman. Moreover there was no common sense in these programs.
The first attempts at AI s in the 1970s added arms and eyes to the robots but they picked things up less well than a six month old. M notes that this disparity between programs that calculate and reason versus programs that interact with the world “remains to this day.” Robots still don’t perform as well behaviorally as infants or non-human animals but play chess superb. So the order of difficulty for machines from easier to harder is calculating; reasoning; perceiving, and acting. For humans the order is exactly the reverse. The explanation most likely lies in the fact that perceiving and acting were beneficial for survival in a way that calculation and abstract reasoning were not. Machines are way behind in many areas and yet catching up: “In less than fifty years, inexpensive computers will match and exceed-in raw info processing power… the human brain.” But can we program them to intuit and perceive like humans?
COCKROACH RACE – Cybernetics tries to copy the nervous system by imitating its physical structure. After slowing down in the 60s, neural net technology reinvigorated the discipline when “computers became powerful enough to simulate interesting assemblies of neurons.” But copying a brain is difficult because examining a brain with present instruments is relatively primitive. M believes that both top down and bottom up methods will be used to create robots that interact with the world thus “recapitulating the evolution of biological minds …” M goes into detail “about the slow buildup of their [robots] arrival.”
CHAPTER 3 – POWER AND PRESENCE
Since computers have far to go to match humans, M will estimate future trends on the basis of analogy and inference and extrapolation. Computer vision can now follow roads but need to improve tenfold [1000 Million Instructions Per Second (MIPS] for reasonable 3D spatial awareness and another tenfold to find 3D objects in a clutter reasonably fast. There are lots of other technology that seek to replace brains and eyes like handwriting and speech recognition..
The key question is how much more computer power is needed to reach human performance. M suggests we relate nerve volume to computation. For example, the neural assemblies of the retina can be compared to what has happened so far with robot vision. The human retina has about 100 million neurons and processes about 10 images per second. Based on his extrapolation, M estimates that it will take 100 million MIPS of computing power to match human sight functionality.
Today’s most powerful super computers can do a few million MIPS; that is, they are within a factor of 100 of having the power to mimic a human mind. Of course such computers would have to cost about 1K for them to make economic sense. M extrapolates that “computers suitable for humanlike robots will appear in the 2020s.”
M is unconcerned over claims that the exponential growth of computing power will subside. He suggests a number of possibilities/ideas/technologies that will overcome difficulties including: single-electron transistors, quantum dots, quantum interference logic, molecular computers, and of course quantum computers. M believes that humanlike robots will arrive without these more exotic techniques. (M will detail in a later chapter how the evolution of robot minds will parallel the evolution of human minds but be 10 million times as fast. With robots achieving “humanlike intelligence in about forty years.” )
M counters the critics by arguing that the next fifty years will see change happening more quickly than the last fifty. Why? First of all the growth and competitiveness of the computer industry itself. Second, machine research has only been progressing since about 1990 with the funding necessary to double power every year. The result: Machine-read text, speech recognition, robots driving cars and crawling on Mars, and composing music. From the inside robots will be machines, from the outside they will appear intelligent.
M draws an analogy to topography. The human landscape of consciousness has high mountains like hand-eye coordination, locomotion and social interaction; foothills like theorem proving and chess playing; and lowlands like arithmetic and memorization. computers are analogous to a flood which drown the lowlands; has just reached the foothills, and well eventually submerge the peaks.
Turing anticipated the development of minds a half-century ago. He thought machines would pass the test in about 2000, and though they do for certain restricted topic tests, it will be a few more decades until they can truly be said to have passed the Turing test. In Turing’s famous paper “Computing Machinery and Intelligence” he responded to 9 classes of objections to the machines can be made to think issue. They were:
1) Theological – thinking comes from souls, machines don’t have souls, machines can’t think.
2) The “Heads-in-the-Sand” Objection – thinking machines aren’t possible because the consequences would be terrible.
3) Mathematical mechanical reasoning has limits that human reasoning doesn’t.
4) The Argument from Consciousness – machines have no inner experience to give meaning to what they do.
5) Arguments form Various Disabilities – machines can’t be kind, moral, joyous, etc.
6) Lady Lovelace’s Objection – computers can only do what they’re programmed to do.
7) The Argument from Continuity in a Nervous System – nerves respond to tiny signal differences, Cs work in fixed-size steps.
8) The Argument from Informality of Behavior – it isn’t possible to specify what a C should do in every possible situation a human might be in.
9) The Argument from ESP – humans sense things that deterministic computers can’t.
Turing (T) replies:
1) T was an atheist who rejected religious explanations. But for the sake of argument, T asked if a God couldn’t put a soul in a machine if she wanted to. [He assumes the answer is yes.] Furthermore, the soul is a name for subjective consciousness. The mechanistic idea is that consciousness arises from patterns of brain activity; no dualism necessary. This suggests that mechanisms could be made to produce consciousness. M tries not to settle the metaphysical issues but suggests that when robots interact with us and appear intelligent we will recognize them as such.
2) T thought arguments against AI came from a fear of being replaced. T suggested that the way around this problem would be to download our brains into robotic bodies. M argues that we do fear other “tribes” encroaching on our territory but robots will resemble us, acquire our values, and share our goals: “antisocial robot software would sell poorly…” We should look upon those who will inherit the world from us as our children.
3) The math objection revolves around Godel’s theorem, that axiomatic mathematical systems contain true statements that cannot be deduced. Analogously, T showed that universal T machines confront the same self-referential paradoxes. Since machines can be stumped by Godel questions that humans can answer, machines have limits that humans don’t. T thought this painted machines as rigid deterministic mechanisms and that this characterization was true for only the short-term
4) T pointed out that we don’t know if other persons have subjective experiences so we just assume they do because they act conscious and we’ll reach similar conclusions regarding intelligent machines.
5) T considered this a generalization from the experience of intelligent machines of the day. Fifty years later things look a bit different. And fifty years hence?
6) This is obviously false—programs do lots of unexpected things often producing solutions that would have taken humans lifetimes.
7) Computers count rather than measuring and some feel that this discontinuity has less potential than continuity as in nervous systems. T believes this is human hubris.
8) Sophisticated machines will react sometimes by rules and sometimes unexpectedly.
9) Do we need to say anything?
CHAPTER 4 – UNIVERSAL ROBOTS
There are only a few thousand robots—some over ten years old and they aren’t very advanced. M predicts that the earliest general use robots may be the robot vacuum cleaner followed by robots that dust, pick up clutter, mow lawns, etc. If successful this should bring about a spiral effect for more research and better robots. Robots are to physical work what computers are to paperwork, and, since there is more of the former than the latter, M predicts that robots will eventually be much more numerous than computers.
First gen robots – 2010 – 3000 MIPS (lizard scale) – Distinguishing feature – general purpose perception, manipulation, and mobility. They will do light mechanical work, food preparation, household tasks, and car tune-ups.
Second gen robots – 2020 – 100,000 MIPS (mouse scale) – accommodation learning. They will adjust to an action’s past effectiveness, that is, use genetic algorithms. M thinks they will find jobs and become the largest industry on earth.
Third gen robots – 2030 – 3,000,000 MIPS (monkey-scale) – world modeling. will learn much faster than 2nd gen robots. Most importantly it will be able “to simulate its world in real time.” They will create simple programs of their own and do cool stuff.
Fourth gen robots – 2040 – 100,000,000 MIPS (human scale) – reasoning. The bottom up method slowly transfers perceptual and motor faculties while AI will transfer reasoning to robots. They will understand language “hey, the water is still running in the bathtub, get your little mechanism up there.” And of course, they will be smart enough to design their own successors—without us!
So four generations of robots will mimic the 400 hundred million year evolution marked by the brain stem, cerebellum, mid-brain, and neocortex. But will these things be conscious? Have emotions? M knows it upsets many to say yes. Just as the terrestrial and celestial was once a sacred distinction, so is the animate/inanimate thought to be a sacred distinction. Of course if the animating principle is a supernatural soul, then the distinction remains. But our current knowledge suggests that complex organization provides animation. We are, in effect, “inspiriting the dead matter around us.”
Naturally robots will manifest conscious/internal life as they advance. Fear, shame, and joy may be emotions valuable to robots so as to retreat from danger, reduce the probability of a future bad decision, or reinforce good decisions. M thinks there would be good reasons for rs to have platonic love for their owners and, since robots don’t have to be selfish to guarantee their survival, they will be nicer than most humans. Anger is more complicated, but M thinks it might be necessary to have robots capable of anger.
M notes that many reject the view that dead matter can give rise to consciousness. Philosopher Herbert Dreyfus argues that computers can’t “capture the ineffable intuitive subconscious,” while his colleague John Searle says that computers “may simulate thought, but will never actually think meaningfully.” Roger Penrose argues that consciousness is achieved “through gravitational collapse of the quantum wave function in individual neurons.” But M points to the accumulating evidence from neuroscience to disagree.
As robots become increasingly proficient, they could do almost all the work for us and support us from activity in the solar system: “leaving behind a nature preserve subsidized from space.” M sees this as a natural development with humans using one of their two channels of heredity. Not the slower biological DNA, but thru culture by books, language, and machines. For most of human history there was more info in our genes than in our culture but now libraries alone hold thousands of times more info than genes.
“Given fully intelligent robots, culture becomes completely independent of biology. Intelligent machines, which will grow from us, learn our skills, and initially share our goals and values, will be the children of our minds.”
CHAPTER 5 – THE AGE OF ROBOTS
A 100,000 years ago, our ancestors were supported by fully automated nature. With agriculture, we increased production but added work. Until recently everyone was a farmer, and most worked producing food. Farmers lost their jobs to machines and moved to manufacturing; but more advanced machines moved displaced farmers out of factories and into offices; where machines have put them out of work again. Soon machines will do all the work. As tractors and combines amplify farmers, computer workstations amplify engineers, resulting in productivity previously undreamt of. In the office layers of management and clerical help slowly disappear. The scribe, priest, seer and chief no longer are the repository of the sages wisdom—printing and mass communication ended that. There are no telephone operators to speak of and most queries are handled by voice recognition. Text readable machines sort mail, and phones and cash registers have been replaced by voice mail and ATMs. “Advancing automation and a coming army of robots will displace labor as never before.” In the short run this causes panic and the scramble to earn a living in new ways. In the med run it provides the opportunity to have a more leisurely lifestyle. In the long run, “it marks the end of the dominance of biological humans and the beginning of the age of robots.“
So there will be no work, when the robots become sufficiently advanced. M thinks this is good—less stress, urban strife, tribalism, war, etc. Humans will be able to live wherever they want—probably not cities—as robotic workers care for all our needs. The robots need to be constructed to enjoy serving us, along the model of the social insects for example. And it will be prosperity that will eliminate most instances of aggression. We will then become Exes(ex-humans or post-biological beings) and we will explore outer space.
CHAPTER 6 – THE AGE OF MIND
M maintains that the future actually exceeds the imaginings of Verne, Franklin, Da Vinci, etc. Exes will compete and eventually “be transformed into intelligence-boosting computing elements. … Physical activity will gradually transform itself into a web of increasingly pure thought, here every smallest interaction represents a meaningful computation.” M thinks high-energy physics has only scratched the surface and we “may learn to tailor spacetime…” Exes will arrange spacetime and energy “into forms best for computation” with the result that “the inhabited portions of the universe will be rapidly transformed into a cyberspace, where overt physical activity is imperceptible, but the world inside the computation is astronomically rich.” Beings wont be defined by physical location but will be patterns of info in cyberspace. Minds, pure software, will interact with other minds. The wave of physical migration into space will have long given way to “a bubble of Mind expanding at near lightspeed.” 
As for the “state of Mind” M sees boundaries of personal identity as breaking down but still remaining. Ineffective thoughts will still be weeded out by a Darwinian evolution. Exes will have more future since they will “cram more events” into physical time. And cyberspace will be “much bigger and longer lasting than the raw spacetime it displaces.” After some difficult calculation, M determines that the 1045 bits of a single human body “could contain the efficiently encoded biospheres of a thousand galaxies-or a quadrillion individuals each with a quadrillion times the capacity of a human mind.” And the “expanding bubble of cyberspace” will recreate all it encounters, “memorizing the old universe as it consumes it.” 
M wonders if some ind mind might escape from its small role in a godlike mind to be… independent? To consider this M turns to a discussion of telepresence and VR. Imagine you are in a good simulated reality, your brain attached to a simulator and then eventually replaced altogether by artificial hardware. ” … our essences will become patterns that can migrate the information networks at will.” Surprisingly, M thinks we might still want a body. Well, the illusion of a body at least for a while, but he speculates that the slow substitution of AI we will finally be liberated from any sense of our original brain/body. ” … the bodiless mind that results … would hardly be human. It will have become an AI.” 172-3
M moves next considers theorist who think time-travel is possible: the Kerr-Newman solutions of the 1960s; Tipler’s work in the 70s; and the Wheeler-Feynman model of the 1940s which sends signals to the past. M goes into detail about time-loop logic. The warping effects of time loops “may be magic paddles to navigate the alternative worlds in powerful ships.”
CHAP 7 – MIND FIRE
M is a “physical fundamentalist” and yet Descartes beckons—simulated worlds with simulated persons seem possible. “A possible world is as real … as conscious observers, especially inside the world, think it is!” But what is consciousness? The prescientific idea of soul has been successful socially, but not scientifically. M thinks that con is a byproduct of “a brain evolved for social living.” We told stories of physical and psychological events including the teller’s state of mind. Thus consciousness is “the continuous story we tell ourselves … about what we did and why we did it.” It is often inaccurate and it is subjective. Objectively it “is just a pattern of electrochemical events…”
M suggests that consciousness is ubiquitous, even in those things that appear to lack consciousness. Furthermore he believes a future universal mind might be able to give our lives meaning. Cosmic mind, he believes, will be subjectively infinite and self-conscious.
Quantum mechanics is not common sense. A particular problem is how to reconcile the commonsense view that things are in particular positions with the a scientific cosmology that asserts that the universal behaves as a wave function that has not collapsed? In other words, why does the universe appear the way it is, when it isn’t really any way? The many worlds interpretation of quantum mechanics may resolve thIS above question. The world is what it appears to be only because we see it that way. But our mind children will be able to “transcend our narrow notions of what is.”
Since our existence is largely self-produced; our descendents will exist in different realms. They may be able to traverse in and thru other possible worlds. Consciousness may be able to exist in many possible worlds and our existence in this one may be a consequence of it being the made up of the simplest rules that could have produced consciousness. As long as we are alive we are governed by the laws of the universe. But after death things will change. And in the future they will too; when our mind children may learn how to proceed thru other worlds. But for now, there is only Shakespeare’s lament:
To die, to sleep;
To sleep: perchance to dream: aye, there’s the rub;
For in that sleep of death what dreams may come,
When we have shuffled off this mortal coil …