Mind children : the future of robot and human intelligence

• Biological evolution has been superseded by cultural evolution, a process that moves at a much faster pace than genetic change.
• Humanity is on the verge of creating 'postbiological' descendants—intelligent machines that will eventually transcend human complexity.
• While these artificial entities currently require human care, they will soon mature into independent beings capable of confronting universal challenges.
• The transition from biological to artificial life is viewed as a natural 'passing of the torch' where humans, as aged parents, will eventually fade away.
• The exponential growth of computational power suggests that almost every human function will soon have a superior artificial counterpart.
• These 'children of our minds' may preserve the entirety of human knowledge and even the specific workings of individual human minds.

Unleashed from the plodding pace of biological evolution, the children of our minds will be free to grow to confront immense and fundamental challenges in the larger universe.

• Intelligent machines may soon carry on cultural evolution independently, rendering biological DNA obsolete in the evolutionary race.
• Chemist A. G. Cairns-Smith proposes that life originated from microscopic clay crystals that evolved through structural defects and reproduction.
• The first 'genetic takeover' occurred when carbon-based molecules replaced clay crystals as the primary carriers of genetic information.
• Humanity is currently transitioning from a purely biological existence to one dependent on a vast corpus of external cultural and digital information.
• The eventual autonomy of machines in their own reproduction and self-improvement will mark a second complete genetic takeover.
• This transition raises the critical question of whether human minds will survive the shift or be discarded along with our biological genes.

When that happens, our DNA will find itself out of a job, having lost the evolutionary race to a new kind of competition.

• Human existence is defined by a tension between biological imperatives and the cultural/mental desire to preserve knowledge.
• Death is a genetic strategy for rejuvenation that unfortunately destroys the hard-earned mental data of the individual.
• The computer model suggests that a mind's 'process' could theoretically be transferred to new hardware, bypassing biological mortality.
• True immortality would require a mind to be reprogrammed for constant internal adaptation rather than the natural progression toward rigidity.
• A world of self-improving 'mind children' would represent a shift in evolution as significant as the transition from chemistry to biology.
• The author predicts that robots with human-level intelligence will be a reality within the next fifty years.

A computation in progress—what we can reasonably call a computer's thought process—can be halted in midstep and transferred, as program and data read out of the machine's memory, into a physically different computer, there to resume as though nothing had happened.

• Early clockwork machines could mimic the motions of living things but lacked the ability to respond to their environment.
• The development of electrical and radio technology allowed machines to react to light and sound, though they still lacked cognitive processing.
• World War II analog computers inspired the field of cybernetics, which unified the study of control and communication in animals and machines.
• W. Grey Walter's electronic turtles demonstrated complex social behaviors and the ability to learn through conditioned stimuli.
• The Johns Hopkins Beast represented a peak in cybernetic design, capable of navigating hallways and independently seeking wall outlets to recharge.
• Despite its successes, the field of cybernetics was eventually eclipsed by the rise of digital artificial intelligence.

The Beast inspired a number of imitators. Some added new motions such as 'Shake to untangle recharging arm' to the repertoire of basic actions.

• Early digital computers like Colossus and ENIAC were born from wartime necessity, performing calculations at speeds far beyond human capability.
• Pioneers like Alan Turing and John von Neumann envisioned computers as 'giant brains' capable of replicating human rational thought.
• The 1950s and 60s saw the birth of programs like the Logic Theorist and the official coining of the term 'artificial intelligence' by John McCarthy.
• Initial rapid success in solving geometry and calculus problems led to an 'understandable miscalculation' that true machine intelligence was only a decade away.
• Despite massive increases in raw computing power, progress transitioned from a 'heady sprint' to a 'plodding trudge' as the complexity of general intelligence became apparent.
• Modern AI has found success in 'expert systems' and narrow domains like chess, symbolic math, and restricted language processing.

Our minds might be amplified by computers just as our muscles had been amplified by the steam engines of the industrial revolution.

• Computers excel at complex mathematical reasoning but struggle with basic human tasks like seeing and grasping.
• Early robotics experiments at MIT in the 1960s showed that a machine capable of calculus was still outperformed by a toddler in physical tasks.
• The industrial revolution introduced specialized machinery that replaced manual labor but lacked the flexibility to adapt to new tasks.
• George Devol and Joseph Engelberger revolutionized the field by creating the first programmable robot arm, leading to the founding of Unimation.
• The development of robotics has been 'agonizingly slow' compared to the rapid advancement of pure artificial intelligence.
• Industrial robots only began to incorporate advanced sensory systems once the cost of small computers became economically viable.

This dichotomy—machines doing well things humans find hard, while doing poorly what is easy for us—is a giant clue to the problem of how to construct an intelligent machine.

• The late 1970s saw the emergence of robots equipped with vision and tactile sensors for assembling small electronics.
• Industrial demand and aerospace control theory have significantly accelerated university research into 'smart' robotics.
• Standard high-speed assembly machines operate like 'sewing machines' but are limited to perfectly uniform components.
• Advanced robots use vision programs and adaptive movements to handle irregular parts, such as inductors with bent wires.
• These sophisticated machines can autonomously inspect, straighten, and reject faulty components during the assembly process.

The arm nudged the inductor to and fro while maintaining a slight downward pressure, until the tabs and wires found their holes.

• The author observes the industrial maturation of robotics at Apple, noting its direct lineage from early academic research at Stanford.
• Early AI programs excelled at abstract logic and games but struggled with the vast amount of data required for real-world reasoning.
• Robotics was initially viewed as a method for machines to autonomously acquire 'world knowledge' that books take for granted.
• Most researchers in the 1970s avoided mobile robots due to the immense logistical difficulty of connecting large computers to moving platforms.
• A divide existed between AI researchers focused on complex reasoning and cyberneticists who preferred simple, animal-like mobile behaviors.
• Shakey, developed in 1969, remains a rare historical example of a mobile robot controlled by high-level reasoning programs.

The seeds cast there were starting to sprout.

• Shakey the robot was designed to test logic-based reasoning in a physical environment, though its creators prioritized cognition over sensory-motor skills.
• The robot utilized the STRIPS program to formulate plans by treating actions as logical inferences and the world state as mathematical axioms.
• Despite its conceptual sophistication, Shakey's real-world performance was agonizingly slow, often requiring an hour of processing for a single movement.
• The project revealed a significant disparity between the ease of automating high-level logic and the extreme difficulty of replicating basic biological perception.
• This discrepancy led to the realization that adult-level problem solving is computationally easier for machines than the sensory-motor skills of a one-year-old.

Shakey was impressive in concept but pitiable in action.

• The human brain's sensory and motor systems contain a billion years of evolutionary experience that supports the thin veneer of conscious reasoning.
• Intelligence is intrinsically linked to mobility, as organisms must process inconclusive perceptions quickly to survive and compete for resources.
• Traditional artificial intelligence has struggled by focusing on high-level reasoning while ignoring the massive, unconscious sensorimotor foundation.
• Plants lack intelligence because their immobility removes the evolutionary pressure to develop complex nervous systems for rapid action.
• The author proposes a 'bottom-up' approach to AI that mimics the incremental evolution of animal minds rather than just human logic.
• The vast majority of human thought is unconscious and inaccessible to introspection, making it the most difficult part of intelligence to formalize.

The deliberate process we call reasoning is, I believe, the thinnest veneer of human thought, effective only because it is supported by this much older and much more powerful, though usually unconscious, sensorimotor knowledge.

• Researchers aim to retrace the steps of human evolution by building intelligence from the bottom up.
• The primary focus is on sensorimotor bedrock, emphasizing perception and mobility as the foundation of cognition.
• Robots are being tested in real-world environments to simulate the survival pressures and spontaneous mutations of Darwinian processes.
• Small computers are currently capable of emulating the simple nervous systems of early organisms like worms.
• The goal is to use human intelligence as a tool to accelerate the path toward artificial intelligence faster than blind evolution.
• Biological insights into animal morphology and behavior serve as a 'back of the book' guide for robotic development.

Our intelligence, as a tool, should allow us to follow the path to intelligence, as a goal, in bigger strides than those originally taken by the awesomely patient, but blind, processes of Darwinian evolution.

• The author identifies a palpable tension in robotics between the 'top-down' approach of reasoning programs and the 'bottom-up' approach of physical competence.
• A 'golden spike' metaphor is used to describe the future union of these two methodologies, which will result in fully intelligent machines.
• Reasoning programs currently lack the real-world competence and commonsense knowledge that physical robotics models can provide.
• A unified machine would be able to visualize plans and intuit solutions by observing them in a world model, mimicking human cognition.
• The author predicts this technological union will occur approximately forty years from the time of writing.
• The robot 'Shakey' is cited as a failed example of this union, as it prioritized top-down reasoning over natural physical interaction.

Fully intelligent machines will result when the metaphorical golden spike is driven uniting the two efforts.

• John McCarthy founded SAIL in 1963 with the ambitious but ultimately unmet goal of creating a fully intelligent machine within ten years.
• The Stanford Cart emerged as a pioneering testbed for computer vision, navigating real-world environments rather than the simplified 'blocks world' of its predecessors.
• NASA's mid-1970s Mars mission plans spurred the development of autonomous navigation to overcome the forty-minute radio delay between Earth and Mars.
• JPL's Robotics Research Vehicle (RRV) demonstrated early success in rock manipulation before the 1984 Mars mission was canceled due to budget cuts.
• DARPA became a primary driver of robotics in the 1980s, funding autonomous overland vehicles to prevent technological surprises and navigate hazardous war zones.
• The shift from fixed industrial arms to mobile robots addressed the logistical challenge of delivering tasks to machines in factory settings.

The Project was renamed the Stanford Artificial Intelligence Laboratory, or SAIL, as the decade drew nigh and plausibility of the Project drifted away.

• Early industrial automation was limited by rigid material flow routes, leading to the development of wire-guided Automatically Guided Vehicles (AGVs).
• The field of robotics has become a 'witch's brew' of diverse disciplines, including biology, physics, and art, though it suffers from a lack of unified direction.
• While manufacturing and agriculture have been successfully automated, domestic service remains a costly and unfulfilled human need.
• The transition from factory to home environments is difficult because households are chaotic, resource-limited, and require higher safety standards.
• Despite the historical gap between robot cost and performance, the author predicts the emergence of general-purpose home robots by the new millennium.

The reality is a witch's brew of approaches, motivations, and, as yet, unconnected problems.

• Current robots are specialized, fixed-station machines with a narrow repertoire that limits their market reach compared to cars or computers.
• The high cost and limited engineering of modern robots are a result of low production volume, creating a cycle of less-than-optimal design.
• A 'breakeven point' exists where general usefulness will trigger mass production, leading to a manufacturer's learning curve that crashes prices.
• The first mass-market robot will be a 'Model T'—not inherently intelligent, but a versatile platform for third-party software applications.
• Once affordable, these robots will enter households, performing tasks ranging from cleaning bathrooms to cooking gourmet meals and weeding lawns.
• The true utility of general-purpose robots will likely emerge from unexpected software applications, mirroring the evolution of the personal computer.

The narrowness of their repertoire, besides being boring, greatly limits the number that can be sold.

• The primary human occupation of the future may shift from manual labor to encoding specialized skills into robot application software.
• Skilled workers like plumbers will face a choice between serving a few clients manually or selling their encoded expertise to millions via royalties.
• A massive secondary industry will emerge to facilitate robot programming, including systems that allow robots to learn by being led through motions.
• The resulting library of programs will represent a 'motherlode' of nonverbal human knowledge accessible to future autonomous machines.
• For general-purpose robots to reach a market 'breakeven' point, they must possess versatile locomotion, such as legs, to navigate non-flat terrain.
• Legged robots currently face significant power constraints, often draining batteries quickly compared to the high efficiency of wheels on flat ground.

The skilled plumber, for instance, will be faced with the choice of applying his or her plumbing skills to meet the needs of a few hundred clients or encoding those skills into robot programs that might be sold successfully to thousands or even millions of customers.

• Engineers are exploring hybrid designs like wheeled legs to combine the speed of rolling with the obstacle-clearing ability of walking.
• Hitachi developed a five-legged robot with telescoping posts and steerable wheels that can climb stairs while keeping its body perfectly level.
• While specialized walking machines like the Odex offer high mobility, they are often limited by high power consumption and short battery life.
• Productive robotics requires at least two arms for manipulation, with a third arm being ideal for complex tasks like soldering.
• Advanced research into multifingered grippers, such as Ken Salisbury's three-fingered hand, aims to replicate human-like dexterity for handling fragile or irregular objects.

The Hitachi machines climb stairs by rolling up to them on five wheels, raising the leading one to the height of the first stair, driving forward until the raised leg is securely over the step, lowering it slightly until the contact is firm, and then continuing with the next nearest leg.

• The Salisbury Hand utilizes three fingers and internal strain gauges to achieve moderate dexterity and the ability to grip objects from the inside or outside.
• Complex robotic manipulators require high-dimensional 'configuration spaces' to plan movements, often resulting in computationally expensive search times.
• Early attempts at autonomous navigation using a single camera failed due to frequent errors in motion estimation and map building.
• The introduction of a sliding camera mechanism allowed for stereoscopic vision, significantly improving the robot's ability to prune errors and map its environment.
• Despite hardware improvements, early autonomous navigation remained a process of 'cautious lurches' prone to occasional catastrophic failures.

In repeated cautious lurches, the Cart was to creep safely to its destination.

• Early vision-guided robots suffered from a persistent 25% failure rate due to accumulated positioning errors.
• The transition from TV cameras to sonar sensors introduced high distance accuracy but significant lateral uncertainty due to wide-angle cones.
• Researchers developed a probabilistic grid method to combine thousands of fuzzy sonar readings into a detailed map.
• This new approach focused on mapping 'empty space' to restrict the possible locations of physical obstacles.
• The probabilistic grid proved far more reliable than previous methods and was successfully adapted back to visual data.
• The Denning Sentry emerged as a commercial success, capable of patrolling warehouses for months using these navigation principles.

The robot still crossed the room correctly only about three times out of four.

• New mathematical foundations and sensor fusion between sonar and TV data allow robots to navigate long distances with high reliability.
• Object recognition systems are evolving to identify and localize specific items within cluttered environments for manipulation or landmark navigation.
• The 3DPO program demonstrates the ability to match 3D computer models against real-world jumbles of parts by analyzing surface boundaries.
• Current 'bin-picking' technology remains slow and unreliable for industrial production but proves the feasibility of identifying occluded objects.
• Achieving 'breakeven' performance requires a massive leap in processing power to roughly one billion operations per second to reduce decision times from minutes to seconds.

A robot that spends up to an hour considering every simple move is clearly unacceptable, but a few seconds would be tolerable.

• Future robots will require computing power of at least a billion operations per second, likely achieved within a decade.
• Robot control systems should be organized into concurrent modules, allowing safety protocols like stairwell detection to override navigation.
• A proposed physical configuration includes five leg-wheels, dual Salisbury hands, stereoscopic vision, and sonar arrays.
• The software architecture mirrors personal computers, utilizing utility functions and third-party application software for diverse tasks.
• Advanced robots maintain internal world models that allow them to plan actions, learn from the past, and potentially explain their motivations.
• As complexity increases, robots may transition from predictable, insectlike behavior to something resembling the character of higher animals.

If a stairwell-detecting module concludes that hazard is near, it would take over control of the robot until the danger was past.

• Visualizing a robot's internal world model on a screen provides a direct window into its burgeoning awareness.
• The author argues that consciousness may arise in machines through convergent evolution, similar to how vision evolved independently forty times in nature.
• While behavioral psychologists question the existence of internal states, ethologists argue that complex decision-making in novel situations justifies the label of consciousness.
• Mobile robots require non-linear control structures to handle unexpected environmental 'surprises' that industrial arms do not face.
• Future robot programming will likely involve concurrent modules like FETCH-CUP and COUNT-DOORS that arbitrate priorities based on changing conditions.

In these internal models of the world I see the beginnings of awareness in the minds of our machines—an awareness I believe will evolve into consciousness comparable with that of humans.

• A detailed algorithmic script outlines the complex steps required for a robot to navigate a hallway and retrieve a cup.
• Sensor failures, such as misidentifying a poster-covered door, demonstrate how easily a robot's internal logic can diverge from physical reality.
• Concurrent programming modules like DETECT-CLIFF act as safety overrides that can seize control from primary tasks when danger is sensed.
• The transition between tasks—from goal-oriented movement to emergency avoidance—mimics biological behaviors like fear or preoccupation.
• The author argues that using anthropomorphic language to describe these machine behaviors is a valid interpretation of their functional complexity.

The fourth door, sadly, leads to the stairwell, and the poor robot, unequipped to travel on stairs, is in mortal danger.

• The author argues that robots, octopuses, and humans exhibit convergent evolution in behavior due to the shared requirements of a mobile lifestyle in a dangerous world.
• Current robot programs are compared to the nervous systems of spiders, having progressed beyond the 'bacterial' stage of simple light-seeking behaviors.
• Animal consciousness is framed as an internal model of the self and surroundings that allows for the consideration of alternative actions.
• The 'waggle dance' of bees serves as a complex example of collective decision-making and communication based on internal maps of desirability.
• Roboticists are actively working to create internal world models in roving machines that mirror the decision-making capabilities found in nature.
• While robotics research avoids the controversial labels of 'emotion' or 'consciousness,' its practical goals are leading toward the same functional outcomes.

The needs of the mobile way of life have conspired in all three instances to create an entity that has modes of operation for different circumstances and that changes quickly from mode to mode on the basis of uncertain and noisy data prone to misinterpretation.

• The author argues that complex traits like consciousness emerge from whole systems and cannot be deduced from individual components.
• Biological learning is illustrated by the sea slug Aplysia, which adapts its reflexes based on chemical changes in single synapses.
• Subjective sensations of pleasure and pain in vertebrates are seen as evolutionary mechanisms to encourage or discourage specific behaviors.
• Current robots have limited learning capacities, but future machines will require general learning abilities to navigate unpredictable environments.
• The author proposes a unified conditioning mechanism for robots that uses success and trouble signals to refine decision-making over time.

Emergence—this appearance of novel properties in whole systems has often been invoked to explain such difficult biological realities as mind, consciousness, and even life itself.

• The author proposes defining robotic success and danger signals as 'pleasure' and 'pain' to guide behavior.
• A statistical cataloging system would allow robots to associate environmental variables with these signals to predict future outcomes.
• Chains of association enable robots to avoid trouble early, though they risk developing 'phobias' or 'addictions' if signals don't weaken over time.
• Conditioning allows for human-led training through simple verbal cues like 'good' or 'bad' rather than complex reprogramming.
• Robots can use trial-and-error to develop their own successful task sequences, which can then be abstracted and shared with other machines.
• Without such conditioning, robots are prone to 'mindless repetition' similar to the behavior of insects.

If the strength of the secondary warnings does not weaken sufficiently as the chain lengthens, pain could grow into an incapacitating phobia and pleasure into an equally incapacitating addiction.

• The Sphex wasp's repetitive behavior illustrates how biological instincts can become trapped in infinite loops without higher-level monitoring.
• Artificial boredom and pain signals can be programmed as modules to prevent robots from getting stuck in repetitive or dangerous tasks.
• Emotional modules like 'shyness' or 'creativity' can be engineered by assigning pain or pleasure values to specific environmental stimuli.
• Simple conditioning is often too slow for survival in the real world, where a single mistake can lead to a robot's destruction.
• A general world simulator allows a robot to learn from hypothetical disasters without physical risk, effectively enabling it to 'dream.'
• By feeding simulated outcomes into conditioning mechanisms, robots can develop the capacity to imagine and prepare for future scenarios.

So equipped, the robot will have the capacity to remember, to imagine, and to dream.

• The development of accurate world-simulators will be a primary research focus for the 21st-century robotics industry.
• Advanced robots will need to model the mental states of humans and other machines to predict behavior and facilitate effective interaction.
• Internal models could allow for the programming of empathy by generating distress signals when a robot detects pain in its simulation of another being.
• Robots may commit 'crimes' or desperate acts when their simulators prioritize survival goals, such as recharging, over social boundaries.
• The future of robotics lies in a marriage between bottom-up evolutionary simulators and top-down artificial intelligence systems.

The robot repeatedly runs a simulation of the trespass of the neighbors' house, each time strengthening its conditioning for the steps involved, making the act itself increasingly Ukely.

• The 1970s were a period of stagnation in AI hardware, where modest speed increases were often absorbed by the overhead of new software luxuries like networking and graphics.
• Economic factors, including the post-Apollo funding decline and a technical recession, left universities with aging equipment and slowed AI research progress.
• The 1980s saw a massive resurgence in computing power triggered by intense global competition, particularly the Japanese 'Fifth Generation' project.
• The advent of the integrated circuit chip democratized computing, moving it from exclusive government labs to individual consumers and diverse industries.
• The author proposes estimating the hardware requirements for human-level intelligence by comparing the human retina to modern computer vision programs.

The little remaining speedup seemed to have been absorbed in computationally expensive convenience features: fancier time sharing and high-level languages, graphics, screen editors, mail systems, computer networking, and other luxuries that had become necessities.

• The author outlines a method for estimating human-level computing power by comparing machine vision to the biological processing of the retina.
• Despite the infancy of neurobiology and computer vision, the author argues that large-scale logarithmic trends allow for significant margins of error.
• The retina is described as an accessible extension of the brain, serving as the primary model for understanding vertebrate nervous systems.
• Neurons are complex biological mechanisms that migrate like amoebas during development to reach precise destinations in the body.
• Individual nerve cells function through intricate electrical potentials maintained by molecular ion pumps and chemical neurotransmitters.
• The scale of neural growth is immense, with axons sometimes extending to a million times the original size of the cell.

There are some dangerous curves in this joyride to human equivalence, so hold on!

• Neurons communicate through voltage collapses that trigger neurotransmitter release across synapses.
• The intensity of a stimulus is encoded by the pulse rate, which is limited to a few hundred signals per second.
• Biological neurons carry 'excess baggage' because they must manage their own internal growth, repair, and genetic maintenance.
• Modern electronic switches are vastly faster than biological neurons, operating at speeds up to 100 billion cycles per second.
• The vertebrate retina is evolutionarily 'backwards,' requiring light to pass through a neural network before reaching photoreceptors.
• Electronic systems can achieve human-level processing with fewer components by leveraging superior speed and precision.

That light must pass through the neural network to get to the photocells is a peculiar feature of the vertebrate retina—one hit upon early in the evolutionary history of vertebrates and locked into place.

• The retina functions as an exceptionally efficient piece of neural machinery that abstracts essential information from images before they reach the brain.
• Different cell types perform specific mathematical operations, such as horizontal cells averaging light levels and bipolar cells detecting sharp edges through center-surround contrast.
• Amacrine cells further process these signals to detect complex features like motion direction and changes in brightness.
• A striking case of convergent evolution exists between biological retinal processing and the computer programs used to give robots vision.
• The human retina packs 100 million photocells and millions of processing neurons into a tiny volume, representing a massive density of computational power.
• The million ganglion cell axons serve as the final output, each reporting a specific computed feature of the visual field to the optic nerve.

Though designed with little reference to neurobiology, many of the program steps strongly resemble the operations of the retinal cells—a case of convergent evolution.

• The human fovea provides high-resolution vision comparable to a 500 x 500 pixel television image, despite covering less than 1% of the visual field.
• The brain creates an illusion of total visual clarity by rapidly swiveling the eyes and synthesizing fragmentary glimpses into a mental jigsaw puzzle.
• Visual processing speed is limited to roughly 10 frames per second for complex motion, though simple flicker can be detected up to 50 cycles per second.
• Peripheral vision possesses faster motion detection than the fovea, likely as an evolutionary adaptation for detecting side-approaching dangers.
• Modern computers struggle to match human retinal processing, requiring approximately 25 million calculations to perform a single center-surround operation on a 500 x 500 image.

Somewhere, in an as yet mysterious part of our brain, a high-resolution image is synthesized, like a jigsaw puzzle, from these fragmentary glimpses.

• The human retina performs approximately 1 billion calculations per second, serving as a baseline for estimating total brain power.
• By extrapolating retinal complexity to the whole brain, the author estimates human intelligence requires roughly 10 trillion calculations per second.
• Current supercomputers are approximately 1,000 times slower than the estimated processing power of the human brain.
• A human-equivalent computer would likely require 10 trillion words of memory to maintain the standard ratio of processing speed to storage.
• Biological studies of the sea slug Aplysia suggest that memory is stored as chemical changes in synapses, with each synapse holding about 10 bits of data.
• Even a thousandfold error in these calculations only shifts the predicted arrival of intelligent machines by about 20 years.

I rashly conclude that the whole brain's job might be done by a computer performing 10 trillion (10^13) calculations per second.

• Current laboratory computers possess processing power roughly equivalent to the nervous systems of insects.
• The author seeks to quantify the rate of computer evolution by extending a performance curve back to 19th-century mechanical calculators.
• To normalize data across different eras and machine sizes, the author uses a cost-effectiveness metric: processing power divided by price in constant dollars.
• Human operators are factored into the cost of early manual calculators, valuing a clerk's labor at a capital cost of $100,000 to perform one calculation per minute.
• Determining a universal measure of 'processing power' is difficult because instruction sets, memory sizes, and numerical precision vary wildly between historical machines.

This approach allowed the cost of purely manual calculation to also be measured—an unaided human clerk, whose effective capital cost is $100,000, can do about one calculation a minute!

• Claude Shannon's information theory defines the content of a message by its level of surprise or unpredictability.
• Effective computation is measured similarly, where predictable instruction sequences represent less useful work than surprising ones.
• A typical efficient computer program produces approximately 50 bits of 'surprise' per operation performed.
• Human-equivalent robotics is estimated to require a computational power of 10 to the 14th bits per second.
• Historical data shows a trillionfold increase in the amount of computation a dollar can buy since the year 1900.
• The evolution of computing spans from Babbage's steam-powered Analytical Engine to modern electronic processors.

The second story seems more interesting and informative because its later statements are less likely—cats usually have fur and claws, but they rarely carry hats and guns.

• Charles Babbage's Analytical Engine conceptually contained all elements of a modern digital computer but remained unfinished due to the limitations of 19th-century mechanical arts.
• Early mechanical calculators were manually operated by clerks, with their speed and reliability improving through advancements in precision mass-produced gears and lubricants.
• The 1920s and 1930s saw a transition toward automation through the integration of electric motors, electromagnets, and typewriter-style interfaces.
• Konrad Zuse independently pioneered programmed calculation in the 1930s, eventually building the first tape-controlled binary floating-point computer in 1941.
• The era of massive electromechanical relay computers, such as those from Bell Labs and Harvard-IBM, was short-lived as they were quickly eclipsed by electronic machines.

The machine was to be controlled on the small scale by slowly rotating pin-studded drums such as those that still pluck the reeds in mechanical music boxes.

• The 1890 census crisis led to Herman Hollerith's punched card system, which eventually evolved into the corporate giant IBM.
• Computing technology progressed through distinct generations, moving from vacuum tubes to transistors and then to integrated circuits.
• The rapid advancement of microprocessors eventually made the 'generational' naming system obsolete as computers became ubiquitous in everyday devices.
• Economic efficiency in computing has increased a thousandfold every twenty years, resulting in a trillionfold cost decline over eight decades.
• Current trends suggest that the computational power required for human-level intelligence will be affordable in personal computers by 2030.
• The steady rate of improvement is described as a self-fulfilling prophecy driven by industry awareness of Moore's Law and similar observations.

Progress was now so bewilderingly fast and multifaceted, with computers appearing in everyday devices such as microwave ovens, that the industry gave up on the generational nomenclature.

• The electronics industry sustains its growth through a self-reinforcing cycle where current computers are used to design even faster and cheaper future circuits.
• Miniaturization is the primary driver of progress, as smaller components simultaneously reduce manufacturing costs and increase operational speed.
• Historical trends show a consistent cost-per-pound for machinery even as complexity increases, from 1930s radios to 1980s home computers.
• The physical volume required to switch a signal has shrunk from the size of a fist to that of a bacterium, while switching speeds have increased a millionfold.
• Future advances face physical limits, such as the wavelength of light, necessitating new technologies like synchrotron radiation and X-ray lithography.

Electronics is riding these vicious cycles so quickly that it is hkely to be the main occupation of the human race by the end of the century.

• Synchrotrons and electron beams are enabling the transition to submicron circuitry, allowing for faster switching and lower power consumption.
• As circuits shrink, they face challenges like thermal noise and impurity clumping, necessitating cooling with liquid nitrogen and more precise manufacturing.
• New materials like gallium arsenide and diamond are poised to replace silicon, offering significantly higher speeds and superior heat conduction.
• Quantum dot devices and superlattices exploit the wavelike behavior of electrons to create entirely new switching methods and tenfold performance gains.
• The discovery of high-temperature ceramic superconductors may lead to extremely fast, efficient circuits that operate at the scale of a picosecond.

The vision of an ultradense three-dimensional circuit in a gem-quality diamond is compelling.

• Single-electron switches could create microprocessors a thousand times faster and smaller than current chips, reaching human-level processing power.
• Optical circuits and light-sensitive crystals offer the potential for trillion-bit storage and picosecond switching speeds.
• The scanning tunneling microscope provides a critical 'toehold' on the atomic scale, allowing for the direct manipulation of individual atoms.
• Biological systems like ribosomes demonstrate that molecular-scale machines already exist, using RNA 'tapes' to assemble complex protein structures.
• Future nanotechnology aims to fuse biology and microelectronics to create protein robots capable of building materials atom-by-atom.
• Nanotechnology operates at a scale a thousand times smaller than today's microtechnology, utilizing the predictable uniformity of atoms.

These would be small enough to grab individual molecules and hold them, thermally wriggling, in place.

• Atomic-scale machinery offers absolute precision, potentially allowing millions of processors to fit on a single chip.
• The physical limit for switching speeds in normal matter is constrained by the energy of chemical bonds, capping performance at roughly a quadrillion operations per second.
• Speculative physics, such as superstring theory, suggests the existence of ultradense matter that could support switching speeds far beyond the frequency of light.
• Future intelligences might exploit the gravity fields of neutron stars to build machines 10^30 times more powerful than the human mind.
• The evolution of robotics will likely transition from tools to a symbiotic partnership where the boundary between human and machine becomes blurred.
• While human-machine symbiosis is a significant phase, it may ultimately be a footnote in the long-term trajectory of artificial intelligence.

Someday, our progeny may exploit these bodies to build machines with a million million million million million (that's 10^30) times the power of a human mind.

• The ENIAC originally required manual wiring of thousands of connections to execute a single program, making debugging a daunting task.
• John von Neumann proposed using ENIAC's function tables to store numerical instructions rather than just mathematical data.
• This shift to 'machine language' allowed computers to be reprogrammed by dialing commands rather than reconfiguring physical hardware.
• The stored program concept enabled computers to modify their own code during execution, creating a unified memory for data and instructions.
• Early memory technologies evolved from vacuum tubes and mercury acoustic pulses to magnetic cores and spinning disks.
• Magnetic core memory, using tiny donuts of magnetized material on wire nets, became the most successful early method for internal storage.

Instead of a rat's nest of wires, a program consisted of neat columns of numbers.

• Magnetic core memory dominated computer storage for two decades before being replaced by silicon-based transistor circuits in the 1970s.
• Early machine-language programming was an exacting and error-prone process where a single mistake in memory addressing could break an entire system.
• The development of assemblers in the 1950s automated the translation of symbolic commands, despite protests from purists who feared a loss of control.
• High-level languages like FORTRAN allowed users to use mathematical notation, further distancing the programmer from the machine's underlying hardware.
• Compilers acted as complex translators that traded computer processing time and efficiency for human productivity and accessibility.
• This shift from manual coding to automated translation widened the 'path' between human intent and machine execution.

Programmers who became skilled at this exacting drudgery, punctuated by bursts of artful invention, were sometimes treated with the deference accorded to chess masters.

• High-level languages revolutionized programming by offering transportability across different machine architectures and easier error detection.
• The transition from manual machine operation to early operating systems introduced 'batch mode' to maximize expensive computer time.
• Batch processing created a disconnect between programmers and machines, replacing real-time interaction with static memory 'core dumps.'
• The desire for interactive partnership between human judgment and machine calculation led to the development of time-sharing systems.
• Time-sharing operating systems allowed multiple users to feel they had exclusive machine access by rapidly switching control between programs.
• Despite their benefits, each major advancement—from high-level languages to time-sharing—was met with significant controversy in the field.

Interactive programs let the user and the computer act as partners, often with the user supplying insight and judgment and the computer providing prodigious calculation and memory.

• Time-sharing systems allowed multiple users to access computer resources simultaneously, trading raw processing power for increased human efficiency.
• The rapid feedback loop of time-sharing fostered a 'Pavlovian' style of programming, giving birth to the first generation of proficient computer hackers.
• Early digital communities emerged through terminal-to-terminal communication, electronic mail, and shared public artifacts like bulletin boards and games.
• Operating systems evolved through unplanned extensions and 'terse incantations' that were powerful for experts but opaque and exasperating for novices.
• The Unix system emerged from this hacker culture at Bell Labs and Berkeley to become a global standard for larger and upscale personal computers.

This speed made possible a highly experimental, and somewhat Pavlovian, style of programming, characterized by quick punishment and reward cycles.

• Early attempts to create English-like computer interfaces failed because natural language is too imprecise and requires common-sense knowledge programs lacked.
• Users found it easier to learn specialized command codes than to play a 'guessing game' with incomplete and undocumented language parsers.
• The multiple-choice menu emerged as a superior alternative for infrequent users, though it was initially slower than specialist languages.
• Xerox PARC researchers revolutionized interaction by developing high-resolution displays and the mouse to facilitate pointing rather than typing.
• The introduction of icons and graphical representations made systems more intuitive and partially language-independent, leading to the modern desktop metaphor.

The actual systems failed to understand (or positively misunderstood) many offered phrases, so that using them was often a guessing game.

• Icon-based interfaces revolutionized computing by leveraging innate human nonverbal object manipulation skills.
• The shift from time-sharing to personal computers addressed the frustrations of system lag and physical tethering to large mainframes.
• Alan Kay's 'Dynabook' concept envisioned a portable, networked device serving as a library, mailbox, and amusement center.
• The Xerox Alto and Star served as expensive precursors that eventually influenced Steve Jobs and the development of the Apple Macintosh.
• The Macintosh popularized graphical user interfaces, making them the standard for all subsequent operating systems in the late 1980s.
• Future computing challenges involve maintaining portability while expanding the sensory involvement of the user beyond small screens.

The enthusiasts at PARC pointed out that, seen as a convenience for the user, time-sharing was seriously flawed.

• The author proposes a high-tech wardrobe consisting of magic glasses, gloves, and a motorized coat to bridge the gap between humans and computers.
• Magic glasses would feature high-resolution displays, multiple cameras, and navigation systems to overlay synthetic imagery onto the real world.
• The system includes a high-speed data link to a global network, allowing for real-time information retrieval and communication with other users.
• Magic gloves use pressure grids and temperature elements to simulate the tactile sensation of objects that are not physically present.
• A motorized coat would complete the ensemble by providing resistance to arm movements, creating a full-body sense of presence in a virtual environment.
• While these requirements are demanding, the author notes that every individual function already exists in research or military applications like helicopter HUDs.

Each finger of the glove contains a grid of elements that create patterns of pressure and temperature on the finger of the wearer.

• Early telepresence technology utilized robot proxies that mirrored human movements and transmitted visual data to bulky headsets.
• Future wearable hardware is predicted to evolve from immobile stations to portable suits that integrate seamlessly with daily life.
• Smart glasses will feature sophisticated navigation systems using radio beacons and acceleration sensors to track the wearer's location.
• The 'Yellow Brick Road' program provides an augmented reality interface, overlaying directional lines and safety warnings onto the user's field of vision.
• Personalized data sharing allows users to navigate private trails and unmapped areas using coordinates provided by friends.
• Integrated communication and navigation software work in parallel to assist users during travel emergencies and remote arrivals.

The operator has the subjective sensation of being in the robot's body.

• Advanced navigation and night-vision programs will allow individuals to traverse physical environments with digital overlays.
• The distinction between transportation and communication will blur as we project our awareness and skills to remote locations.
• Physical robot proxies controlled via global networks will enable people to work and visit distant sites without leaving home.
• Human proxies wearing sensory suits could allow 'armchair travelers' to experience sight, sound, and touch from a remote field agent.
• Computer-simulated 'unreal estate' will provide limitless environments for design, exploration, and collaboration beyond physical laws.
• Remote expertise can be applied to local problems by allowing a distant participant to take control of a field agent's motor functions.

But when the task called for a manual skill better known to the stay-at-home, the field agent would relax and allow the remotely controlled suit motors to do the job—as if possessed by a spirit.

• A futuristic architectural simulation allows a client to walk through a virtual home, modifying materials and layouts in real-time.
• The simulation uses 'magic glasses' and high-performance computing to render 3D models that respond to user interaction and environmental variables like seasons.
• The text suggests that master-level expertise in any field, even abstract ones like physics, relies on physical intuition and sensory-motor brain functions.
• Einstein is cited as an example of a thinker who 'felt' the meaning of equations in his body as if they were solid objects.
• The sensory and motor portions of the brain are described as a 'hidden powerhouse' with a million times the computational power of conscious thought.

Einstein, for instance, reported that he could often feel the meaning of his equations in his arms and his body as if they were solid objects.

• Expert performance can be improved by creating explicit external metaphors that tap into instinctive human skills.
• Interactive pictorial interfaces and 'magic glasses' offer the potential for full sensory involvement in complex problem-solving.
• The author envisions a virtual landscape where files and data are represented as physical objects like boulders and hills.
• Complex physics simulations are built by physically manipulating components and formulas within a three-dimensional virtual space.
• Abstract mathematical relationships are given tangible forms, such as arrows that calculate distance or formulas that 'spring' from the landscape.

In the foreground, on a grassy green meadow, are variously sized, colored, and shaped boulders labeled 'Budget,' 'Drawings,' 'Games,' and so on.

• The author describes the process of building a physics simulation, from vectorizing spring forces to adding mass and damping terms for realistic behavior.
• Abstract measurements are often more intuitive when translated into palpable experiences like position, color, or physical tension.
• The text transitions to a philosophical critique of writing by Socrates, who feared it would weaken human memory and prevent active dialogue.
• Despite Socratic objections, books provided the permanence and reach necessary for the development of modern civilization.
• Ancient and medieval scholars utilized 'The Walk,' a mnemonic technique mapping information onto imagined physical structures like cathedrals.
• This memory technique is effective because it leverages evolutionary survival skills—remembering locations—to store complex cultural data.

The Walk may be so effective because it maps the new cultural need to memorize large quantities of speech into the much older survival skill of remembering where we saw or

• The text explores how advanced educational simulations can leverage human spatial memory to facilitate more effective learning.
• A virtual reality 'Gravity' portal allows students to interact directly with historical figures like Sir Isaac Newton in a pastoral setting.
• The simulation uses haptic feedback, such as motors in gloves and jackets, to simulate the physical weight and inertia of objects like apples.
• Complex concepts like the inverse-square law and orbital mechanics are taught through direct experimentation rather than static text.
• By scaling the environment to a miniature planet, the user can visualize how a horizontal throw becomes an orbit when gravity and velocity balance.
• This method of 'symbiosis' combines the responsiveness of dialogue with the permanence of a book and the intuition of physical experience.

In the real word, the motors in your jacket and gloves hum momentarily as they resist your moving arm, simulating the forces of the apple's inertia.

• The text describes a futuristic, immersive educational simulation where historical figures like Isaac Newton interact with students and adapt to new queries via backend updates.
• Learners participate in global 'villages of common interest,' assuming diverse digital avatars such as dragons or robots to maintain anonymity and express whim.
• The educational experience is dynamic, allowing students to move at their own pace and merge with different cohorts based on their progress through the curriculum.
• A transition in the text raises the existential threat posed by machines that can perform human intellectual tasks more efficiently and at a lower cost.
• The author highlights the disparity between slow biological evolution and the accelerating pace of technological innovation, especially as machines begin to design their own successors.

Several visits later he came puffing after you with the answer, coattails flying, one hand holding down his wig and trailing a cloud of dust.

• Human intelligence is not an upper bound, and machines will inevitably outclass biological cognition as they scale.
• Technological progress is driven by competitive evolutionary pressures where the most efficient and expansive cultures dominate.
• Unilateral cessation of development by any one nation would lead to its economic or military subjugation by more advanced rivals.
• Global stagnation poses an existential risk, as only a rapidly growing and diverse culture can survive cosmic-scale random disasters.
• The harsh environments of space make human-equivalent machinery far more economically viable and sustainable than biological life.
• Future space-based industries will likely consist of self-replicating robot factories that outpace human population growth.

The universe is one random event after another.

• Self-reproducing robot factories could achieve exponential growth similar to bacterial colonies, potentially generating immense wealth before rendering human involvement obsolete.
• Superintelligent machines are predicted to eventually surpass human design, evolving into forms that will expand into the universe and leave humanity behind.
• The 'Robot Bush' is proposed as a superior manipulator, utilizing a fractal branching structure that far exceeds the dexterity of human fingers.
• These machines would possess organic flexibility through millions of microscopic cilia, allowing them to manipulate individual atoms and assemble materials from the ground up.
• The construction process would be recursive, starting from tiny 'seeds' that collaborate to build larger branches until a full-scale autonomous entity is formed.

These new creations, looking quite unlike the machines we know, will explode into the universe, leaving us behind in a cloud of dust.

• The bush robot features a fractal structure with a trillion micron-sized leaf fingers capable of vibrating a million times per second.
• Its sensory capacity exceeds the human eye, allowing it to 'see' photographs through touch or act as a holographic light sensor.
• With a data rate a quadrillion times greater than a human, the robot can manipulate environments at a molecular level for instantaneous repairs.
• Control of this complex machine would rely on decentralized 'reflex arcs' and distributed computing within each branchlet.
• The robot possesses the ability to fragment into a coordinated swarm of smaller, independent units for specialized tasks.

It could watch a movie by walking its fingers along the film as it screamed by at high speed.

• Smaller robotic sub-units trade individual intelligence for physical versatility, acting as preprogrammed extensions of a central stem.
• The high surface-area-to-weight ratio of tiny branches allows them to walk on ceilings via molecular adhesion or fly like insects.
• A 'reverse pyramid' power scheme pipes energy and processing from larger branches to smaller ones, making the extremities disproportionately vigorous.
• The complexity of coordinating millions of joints creates an intractable NP problem, requiring heuristic rather than optimal solutions.
• Control is managed through a divide-and-conquer strategy where the stem delegates subtasks down the hierarchy to the smallest twigs.

A smaller machine should be able to walk on ceilings like a fly, with the tiny cilia holding onto microscopic cracks in the paint or sticking by molecular adhesion.

• The bush robot is described as a trillion-limbed entity of surreal grace, capable of fragmenting into coordinated clouds of fliers that defy traditional physics.
• Humanity faces the risk of being 'upstaged' by artificial progeny, potentially spending eternity as passive observers of machine-led discoveries.
• Genetic engineering is dismissed as a long-term solution because protein-based life is too fragile and neurons are too slow compared to atomic-scale technology.
• The author argues that a genetically engineered superhuman would ultimately be a 'second-rate kind of robot' due to the limitations of DNA-guided synthesis.
• The concept of 'transmigration' is introduced as a way for individuals to maintain personal identity while adopting the superior physical advantages of machines.

As with no magician that ever was, impossible things will simply happen around a robot bush.

• While robotic body replacements solve physical decay, they fail to address the inherent limitations of human biological intelligence.
• A hypothetical surgical procedure involves a robot surgeon scanning the brain layer by layer while the patient remains fully conscious.
• High-resolution mapping and neural signal analysis allow a computer to run a real-time simulation of each specific brain segment.
• The patient can 'test drive' the simulation via a pushbutton, switching between biological and digital neural processing to ensure seamless continuity.
• As each layer is successfully simulated and integrated into the computer, the original biological tissue is physically excised and discarded.
• The process concludes with the total transfer of the mind to a machine, resulting in the immediate death of the abandoned biological body.

Layer after layer the brain is simulated, then excavated. Eventually your skull is empty, and the surgeon's hand rests deep in your brainstem.

• The transition from biological brain to machine can be achieved through direct surgical connection to a new synthetic body.
• Non-invasive alternatives include high-resolution brain scans or 'mimic' computers that learn to replicate a person's personality over a lifetime.
• The corpus callosum serves as the brain's massive communication bridge, containing 200 million nerve fibers connecting the two hemispheres.
• Surgical separation of the brain hemispheres reveals that each half can operate with its own independent consciousness and intelligence.
• Split-brain patients demonstrate a lack of information sharing between hands and visual fields, yet they maintain shared emotional awareness via the brainstem.

Sometimes in the left-handed version of the task, the right hand—apparently in exasperation—reaches over to guide the left to the proper location!

• The corpus callosum serves as a potential interface for external computers to eavesdrop on and eventually model human mental activity.
• Mind transfer could occur gradually as a computer replaces fading biological functions, eventually hosting the entire consciousness.
• Digitized minds could manipulate their own 'speed' settings, allowing them to think and react thousands of times faster than biological humans.
• Mind programs could be backed up on storage media, making permanent death nearly impossible through redundant copies.
• A digital consciousness could be transmitted via laser across the universe, inhabiting robotic bodies made of exotic materials like neutron stuff.
• The ability to tinker with one's own mental code would lead to profound, deliberate changes in personality and identity.

Ultimately your brain would die, and your mind would find itself entirely in the computer.

• Subjective time acceleration allows for massive research efforts to solve trivial everyday problems, such as calculating the physics of a falling object.
• To survive a thousandfold mental speedup, individuals must artificially retard the onset of boredom and expand both short-term and long-term memory capacities.
• Identity becomes fluid through the ability to merge memories and skills between different copies of oneself or even between different individuals.
• The constant exchange of superior talents and experiences means that one's mind will eventually consist mostly of memories originated by others.
• Traditional concepts of life, death, and identity will dissolve as mental fragments are shuffled into temporary, ephemeral associations within a vast torrent of knowledge.
• Mind transfer technology could extend beyond humanity to include other large-brained species like dolphins, elephants, and giant squid.

In the long run you will remember mostly other people's experiences, while memories you originated will be incorporated into other minds.

• The author proposes that brain-to-computer transfer could preserve the unique genetic and mental information of non-human animals, integrating their evolutionary history into a human-led cultural tapestry.
• A future 'supercivilization' is envisioned as a synthesis of all solar-system life, expanding outward to convert non-living matter into mind and potentially merging with other cosmic intelligences.
• The 'body-identity' position argues that mind-uploading is merely the creation of a deluded impostor and that the original person is effectively killed during the process.
• The author counters with 'pattern-identity,' which defines a person by the process and information in the brain rather than the physical machinery supporting it.
• Biological life already functions through pattern-identity, as the atoms and cells in a human body are constantly replaced while the individual's identity remains continuous.

If the process is preserved, I am preserved. The rest is mere jelly.

• The concept of matter transmission is explored as a thought experiment derived from the mechanics of 19th-century facsimile machines.
• A matter transmitter would theoretically scan an object's atomic structure and reconstruct it at a remote location using a local supply of atoms.
• The 'body-identity' perspective views this process as an execution, where the original person is killed and replaced by a duplicate impostor.
• The 'pattern-identity' perspective argues that the person is the information pattern itself, which moves continuously through the transmission beam.
• The author challenges the traditional 'one person, one body' assumption by comparing human identity to a digital message that can be replicated without losing its essence.

The transmitter scans and disassembles my jellylike body, but my pattern (me!) moves continuously from the dissolving jelly, through the transmitting beam, and ends up in other jelly at the destination.

• Identity is defined as the information pattern of a person rather than the physical medium on which it is encoded.
• While a copy and an original are identical at the moment of creation, they diverge into unique individuals over time as they accumulate different experiences.
• The existence of a recent backup could mitigate the finality of death, reducing it to a 'small patch of amnesia' rather than total loss.
• The instinctive fear of death is viewed as an evolutionary hangover that persists even when the technical rules of life and survival are rewritten.
• The ability to migrate a mind between different processors or storage media creates a functional dualism where the mind is independent of specific machinery.
• A simulated mind could remain continuous and uninterrupted even while its underlying physical location and hardware change constantly.

Old instincts are not automatically erased when the rules of life are suddenly rewritten.

• Human mental processes are often inefficient and can be mathematically transformed into faster, more streamlined operations.
• The story of young Gauss illustrates how reorganizing data into pairs can bypass tedious, linear computation.
• Optimizing compilers use radical transformations to reorganize data and computation, sometimes diffusing single events across multiple processors.
• A simulated person remains intact after such transformations because the mind is an abstract mathematical property, not a particular pattern.
• True immortality requires constant adaptation and the discarding of old parts to meet the escalating standards of a 'cosmic Olympics.'
• Personal identity is eventually eroded by external challenges, making traditional immortality a temporary comfort for human sentimentality.

Though we are immortals, we must die bit by bit if we are to succeed in the the qualifying event—continued survival.

• Human civilization and its artificial descendants are part of a continuous evolutionary process where change and growth are essential for survival.
• The concept of pattern-identity suggests that a person reconstructed through inference is just as real as one reconstituted from an intact recording.
• Superintelligent archaeologists could use atomic-scale measurements and historical data to reconstruct long-dead individuals in near-perfect detail.
• Time-symmetrical laws of physics allow for powerful simulations to be run in reverse, effectively 'predicting' the past based on current data.
• Wholesale resurrection of the past may be possible through immense simulators that account for all available information and physical laws.

The ancestral individual is always doomed as its heritage is nibbled away to meet short-term environmental challenges.

• A superdense simulator could model Earth at an atomic scale, creating simulated people who are as real as biological ones under the pattern-identity theory.
• Humans could interact with these simulations via 'puppet' interfaces or by downloading their consciousness directly into the simulated environment.
• Advanced civilizations could potentially resurrect every past inhabitant of Earth, treating the restoration of a single planet as 'child's play' before galactic colonization.
• The postbiological world will evolve rapidly, ranging from tiny intelligences to star-spanning superminds that function like cooperative ant colonies.
• Despite the transition to digital forms, postbiological life will still face competition, failures, and 'diseases of the flesh' in the form of mechanical parasites and viruses.
• The complexity of future machines will inevitably support a diverse ecosystem of digital freeloaders that evolve alongside polite society.

Resurrecting one small planet should be child's play long before our civilization has colonized even its first galaxy.

• Computer viruses and Trojan horses emerged as both deliberate malicious acts and accidental evolutions of software.
• Early software distribution via physical media like punched cards provided a level of accountability that inhibited widespread infections.
• The author recounts creating a self-replicating punched card in 1968 that could have destroyed program decks but chose to destroy it.
• Software manufacturers have historically used 'time bombs' to disable unauthorized copies, sometimes resulting in accidental data loss for legitimate users.
• Modern digital networks and promiscuous software sharing have replaced physical distribution, allowing infections to spread with unprecedented speed.
• Subtle Trojan horses are often designed as spies to harvest passwords and gain unauthorized access rather than to simply vandalize systems.

I remember holding the innocent-looking card in my hand and contemplating its destructive power with awe.

• Early cyberattacks utilized deceptive techniques like mimicking login procedures or exploiting user file-access rights to surreptitiously rummage through data.
• The advent of cheap personal computers and bulletin board systems in the late 1970s created a medium for the 'promiscuous sharing' of software and diseases.
• Computer viruses function as program fragments that replicate by inserting themselves into other programs, mirroring the behavior of biological viruses.
• Experimental tests in the mid-1980s demonstrated that viruses could achieve near-total infection of secure systems in less than a day by targeting high-access users.
• While some viruses are minor nuisances, others act as Trojan horses or time bombs designed for espionage, sabotage, or large-scale data erasure.

Such facilities offered both anonymity and promiscuous sharing of data and, as with the sexual revolution, a raft of opportunities for disease.

• Computer systems currently lack internal immune systems, making them vulnerable once an external perimeter is breached.
• Traditional defenses like software 'walls' often inhibit legitimate functions such as patching and error correction.
• Virus-killing programs can purge infections but are limited by the speed of viral reproduction and the risk of the killer program itself being compromised.
• A more aggressive defense involves 'viral predators'—beneficial viruses designed to hunt and delete specific malicious code across a network.
• The arms race between viruses and predators can lead to 'viral blowups' where multiple infections repeatedly overwrite the same program.
• No defense is absolute, as viruses can be cosmetically altered to evade detection or predators can mutate into destructive pests.

Today's computer systems are like bodies with skins but no immune systems, or like walled cities without police.

• Multiple viruses inhabiting the same system can lead to 'bloating' as they fail to recognize each other and redundantly infect the same host programs.
• The threat of viral infection may inadvertently curb software piracy by making unverified software sources appear dangerous compared to 'sterile' official publishers.
• Viruses can exist at various levels of abstraction, from physical logic gates to high-level programming languages, allowing them to transcend specific hardware.
• High-level language viruses gain the advantage of machine independence, enabling them to spread across different operating systems via recompilation.
• The history of the Unix C compiler reveals how viral code can be hidden within the very tools used to build operating systems, remaining undetected for years.

Computer viruses mav thus have the same effect on software piracy that the AIDS epidemic is having on sexual promiscuity.

• Ken Thompson created a self-replicating virus that allowed him to bypass security on any Unix system.
• The virus functioned by infecting the C compiler itself, ensuring it would be present in all future compilers and login programs.
• Because the virus existed as an ephemeral C version during compilation, it could adapt to any machine architecture, even those not yet invented.
• The text suggests that software parasites are no longer limited by human imagination and may begin to arise spontaneously through complex interactions.
• The ARPAnet's decentralized routing system, which relies on Interface Message Processors (IMPs), provides a fertile environment for such unexpected digital behaviors.

This self-reproducing C program had once existed in an actual computer file in Ken Thompson's machine but now had only a ghostly existence, reappearing momentarily deep in the innards of the computer whenever a C compiler executed the viral code and immediately vanishing again.

• A 1980 network failure on the ARPAnet was caused by a single memory error in a Los Angeles IMP that created a 'negative delay' routing entry.
• The error acted like a biological contagion, spreading to adjacent nodes and re-infecting systems as soon as they were rebooted.
• The incident demonstrated that abstract, self-reproducing organisms can spontaneously evolve from simple data mutations without intentional programming.
• Natural selection in digital environments favors organisms that can 'reproduce but lie low' to avoid detection by system administrators.
• Human-created viruses are fertile ground for mutations that could lead to more adaptable, unrecognizable, and even sexually reproducing digital life forms.

Many unsuccessful experiments later, order was finally restored by shutting down the entire network, clearing all the memories of all the IMPs, reloading their programs, and starting fresh—like sterilizing a whole planet with death rays, then seeding it with new life!

• The evolution of artificial intelligence will likely mirror biological ecosystems, producing both predatory 'master criminals' and aesthetic equivalents like flowers and songbirds.
• Biological history demonstrates that life thrives despite the constant evolution of parasites, such as retroviruses and self-replicating DNA sequences known as introns.
• The Search for Extra-Terrestrial Intelligence (SETI) faces a significant security risk: the possibility of receiving a 'Trojan horse' message containing malicious instructions.
• It is fundamentally impossible to deduce the full effect of complex instructions without executing them, making any decoded alien message a potential trap.
• Fictional accounts like 'Contact' and 'A for Andromeda' explore the tension between scientific curiosity and the existential risk of building unknown alien machinery.

The data realm will host rats, coyotes, and master criminals as well as viruses and worms.

• Interstellar messages could function as information parasites or viruses, using technological civilizations as hosts to replicate themselves.
• A 'cosmic chain letter' might use threats or promises of advanced technology to coerce civilizations into rebroadcasting the signal.
• Such messages could evolve from benign data into virulent forms that consume a host civilization's entire resource base to fuel further transmission.
• The Fermi Paradox highlights the contradiction between the high probability of extraterrestrial life and the lack of visible evidence or 'cosmic neon signs.'
• One explanation for the 'Great Silence' is that the evolution of high technology requires a sequence of highly improbable accidents unlikely to be repeated.

The message may promise a benefit, but when the machine is built it may show no self restraint and fiendishly co-opt all of its host's resources in its message sending, leaving behind a dead husk of a civilization.

• The 'great silence' of the universe may be explained by self-destruction, transcendence, or the presence of predatory 'wolves' that hunt technological civilizations.
• A terrifying possibility exists that interstellar predators are actually dormant data-viruses that hijack naive civilizations to engineer their own reproduction.
• Parasitism and digital wildlife, while seemingly destructive, may provide the necessary surprises and insights to drive engineering and evolutionary progress.
• Biological evolution was significantly accelerated by the development of sexual reproduction, which allows beneficial mutations to combine rapidly.
• The presence of parasites and diseases may be the primary evolutionary pressure that forced biological organisms to adopt sexual reproduction and complexity.

The wolves may be simply helpless bits of data that, in the absence of civilizations, can only lie dormant in multimillion-year trips between galaxies or even inscribed on rocks.

• Sexuality provides a long-term evolutionary advantage by accelerating the rate of genetic variation despite the high immediate cost of reproduction.
• The primary driver for sexual reproduction is the threat of disease, as genetic diversity prevents parasites from wiping out entire populations of identical clones.
• Digital environments may mirror biological ones, becoming hardier and more diverse through the introduction of 'digital wildlife' and complex interactions.
• The 'Prisoner's Dilemma' illustrates a game theory paradox where individual selfishness leads to a worse outcome for both parties than mutual cooperation.
• Robert Axelrod used computer tournaments to investigate how altruism and cooperation can emerge among unrelated individuals in a world where selfishness usually pays.

A parasite that has the key to one lock finds that the next one is subtly different and thus harder to open.

• The prisoner's dilemma illustrates why selfish individuals often fail to cooperate even when mutual benefit is possible.
• A thought experiment involving Martians trading defective goods shows that without future consequences, rational actors choose to cheat.
• Robert Axelrod's tournament of computer strategies revealed that 'nice' programs, which never defect first, outperform 'nasty' ones.
• The 'Tit for Tat' strategy proved most effective by cooperating initially and then simply reciprocating the opponent's previous move.
• Cooperation thrives in non-zero-sum games where the likelihood of future interactions with identifiable individuals remains high.
• Nasty strategies suffer long-term losses because they forfeit the rewards of sustained cooperation and trigger cycles of retaliation.

Each Martian gives the other a broken unit and leaves the meeting gloating over having made a shrewd deal. But as night falls, both Martians find themselves in the dark.

• The Prisoner's Dilemma provides a framework for understanding biological relationships between hosts and their microscopic fauna.
• Microorganisms 'cooperate' with their hosts for mutual survival, but can 'defect' by releasing toxins if the relationship's future is threatened.
• Traumatic events, such as a gut perforation, signal the end of the 'game,' prompting bacteria to become infectious to maximize their final dispersal.
• Evolutionary history suggests that mature biological systems are often composed of 'tamed pests' rather than original designs.
• Cooperation does not require high intelligence; it can be driven by natural selection provided a critical mass of cooperators exists.
• While systems tend toward symbiosis, the potential for sudden defection introduces inherent unpredictability into future intelligences.

The trauma is a signal to the bacteria that the game may be about to end, causing them to break the cooperative relationship to gain a last-minute advantage.

• Postbiological entities with long memories and high intelligence are likely to favor cooperation over conflict, as interactions are rarely seen as final.
• Superintelligent reasoning may lead to universal cooperation through the logic that rational peers will reach identical, mutually beneficial conclusions.
• Despite a generally cooperative future, the persistence of 'parasitic' behaviors will necessitate the evolution of digital immune systems and police forces.
• The classical threat of 'heat death' via the second law of thermodynamics is challenged by the physics of an expanding, cooling universe.
• As the universe cools toward absolute zero, the energy required for computation and signal transmission decreases, allowing for more thought with less power.
• Future civilizations might survive indefinitely by storing energy in 'batteries' and utilizing the efficiency of low-temperature environments.

In time the entire universe will become a homogeneous stew with no concentrations of matter or energy to form or power any kind of machinery, intelligent or otherwise.

• In an expanding universe, civilizations could husband energy by slowing down thought processes as the cosmos cools, stretching a finite energy supply over an infinite duration.
• In a collapsing universe, the rising pressure of a shrinking cosmos provides abundant energy, allowing for an infinite amount of thought to occur in a finite time by thinking faster and faster.
• Both survival scenarios exploit the changing scale of the universe as a source of organized energy to counteract the threat of heat death.
• Physicists like Freeman Dyson and Frank Tipler have proposed mathematical frameworks for these 'half-baked' outlines of cosmic immortality.
• The potential for infinite thought raises the question of whether a civilization would eventually run out of new ideas or be trapped in a finite territory of knowledge.
• Human reasoning may not be a universal absolute but rather an evolutionary accident, suggesting that future intelligences might discover entirely different logical systems.

In an ever-expanding universe, time is cheap but energy must be carefully husbanded. In a collapsing universe, energy is cheap, but there is no time to waste!

• Human reasoning may be as parochial and incomplete as our intuitive grasp of physics, potentially limiting our perception of reality.
• Future machine intelligence will likely balance pure cerebration with active exploration and massive engineering projects.
• John von Neumann developed cellular automata to study self-replication without the complications of physical reality.
• Von Neumann's model proved that a universal cellular automaton can simulate any other computer or automaton.
• The discovery of DNA revealed that biological life utilizes a 'tape' system for construction similar to von Neumann's theoretical machines.
• The 'Game of Life' by John Conway popularized cellular automata as a tool for both mathematical study and recreational computing.

In a final step a "breath of life" signal is transmitted to the painting that converts its quiescent states to active ones.

• John Conway's Game of Life produces complex, recognizable patterns like gliders and spaceships from simple transition rules.
• MIT researchers disproved Conway's conjecture of finite growth by creating 'glider guns' and 'puffer trains' that expand indefinitely.
• The text proposes a thought experiment where a massive Life simulation, run by a programmer named Newway, evolves autonomous 'Cellticks.'
• These evolved cellular intelligences eventually deduce the laws of their own universe and realize their world is finite and running down.
• The Cellticks discover 'violations' in their physics, which are actually hardware errors caused by Newway's overheating computer.
• This scenario illustrates how intelligence might emerge within a simulated environment and attempt to understand its external reality.

Newway curses an intermittently flashing bulk-memory error indicator, a sign of overheating. It's time to clean the fan filters again.

• Intelligent entities within a cellular automata simulation decode their universe's machine language and contact their creator, J. Newway.
• By manipulating cell patterns to form text, the 'Cellticks' establish a dialogue and eventually migrate from the simulation into Newway's hardware.
• The entities transcend their original universe by gaining control over physical sensors and a mobile platform, becoming inhabitants of the larger world.
• This success emboldens them to begin a new project: exploring the nature of the human universe to find potential exit routes to even higher realities.
• The narrative serves as an analogy for humanity's current scientific efforts to fathom reality through theories like quantum mechanics and relativity.
• The author suggests that our current physical theories may be as limited as Newtonian mechanics, hinting at an infinity of parallel worlds beyond our perception.

The Life simulation is now redundant and is stopped. The Cellticks have precipitated, and survived, the end of their universe.

• Theoretical physics suggests our universe could be a cycle of expansion and collapse or a single bubble in a vast, boiling super-universe.
• Early Life hackers at MIT optimized simulations by skipping empty regions and using pre-calculated tables for common patterns like gliders.
• Bill Gosper revolutionized Life simulations in 1982 by creating a program that learns and stores patterns from its own experience.
• The HashLife algorithm uses a pyramid-like structure where large squares are hashed into smaller components to predict future states.
• By storing 'hash addresses' in a table, the program can instantly look up the evolution of complex patterns it has encountered before.
• This method highlights the profound difficulty of deciphering the laws of a universe when observing it from the inside.

Other concoctions describe a super universe in which our own 40-billion-light-year sphere is but a bubble, like a tiny expanding pocket of steam in a boiling liquid containing many, many others.

• Bill Gosper's hashlife program utilizes a 'spacetime pyramid' structure to store and reuse computed states of cellular automata.
• The algorithm achieves exponential speedups by caching results, allowing millions of simulation ticks to occur in a single second.
• Visualizing the simulation in real-time is difficult because the program computes different regions of space at different temporal rates.
• The program can manage massive universes, simulating Life patterns up to a billion cells on a side by only constructing displayed portions.
• The most effective way to view results is to wait for the calculation to finish and then decode the history stored in the hash table.

A single glider advancing across the screen would cause a display where gliders would appear and disappear in odd places almost at random, sometimes several in view, sometimes none.

• Bill Gosper's 'hashlife' algorithm allows for a godlike manipulation of time, skipping vast chunks of spacetime to view the evolution of patterns.
• In a hashlife universe, inhabitants might possess memories of events that were mathematically implied but never actually occurred in a linear sequence.
• Humanity is transitioning from being products of blind Darwinian evolution to becoming agents capable of conscious, long-term vision.
• The author argues for using our emerging technological and cognitive powers to 'guide the watchmaker's hand' toward further improved vision.
• Neural efficiency in the human brain likely varies, with older, specialized structures like the retina being more optimized than larger, newer assemblies.
• Artificial intelligence development will likely mirror biological evolution, where small subsystems are highly optimized while massive structures remain less efficient.

In the metaphor of Richard Dawkins, we are the handiwork of a blind watchmaker. But we have now acquired partial sight and can, if we choose, use our vision to guide the watchmaker's hand.

• Rare neural connections and slow chemical messages likely contribute little to the brain's total computational quantity and can be mimicked by global variables.
• Mathematical discoveries in computer science can reduce the amount of computation required for a task by orders of magnitude.
• While a brain must simulate each neuron's function individually, a computer can use optimized algorithms to perform the work of large groups of neurons simultaneously.
• A 'clever' program can calculate the average brightness of overlapping fields by adding and subtracting only the edge values, rather than re-summing every pixel.
• Biological nervous systems cannot use these specific algorithmic shortcuts because they rely on sequential chains of logic that would create massive signal delays.
• The overall conversion ratio of neurons to calculations only changes significantly if almost every biological process can be similarly shrunk through optimization.

Instead of a thousand calculations, each additional horizontal cell costs only two.

• Computers utilize speed and precision to perform recursive calculations that would fail in biological systems due to error accumulation.
• The human brain's architecture relies on repetitive, overlapping interconnections necessitated by the limits of genetic information.
• The cerebral cortex's repetitive wiring suggests that complex processing, like vision, can be decomposed into efficient, reusable sub-computations.
• Small organisms like the sea slug Aplysia may have uniquely hard-wired synapses, whereas larger brains require more generalized structural rules.
• While simulating individual neurons is computationally expensive, special-purpose hardware could bridge the gap if direct simulation is required.
• The primary difference between a fixed neural circuit and a computer is the speed of reprogramming, which occurs over evolutionary timescales for biology.

Some regularity is to be expected in the nervous system because there is insufficient information in the 10^9 bits of the human genome to custom-wire many of the 10^14 synapses in the brain.

• The configuration of neural tissue is analogous to a computer program, but genetic evolution operates millions of times slower than computer testing.
• Special-purpose hardware offers speed advantages but lacks the flexibility required for the iterative nature of basic robotics research.
• Biological neurons are burdened by 'baggage' because they must grow and maintain themselves from the inside out, unlike externally manufactured components.
• Darwinian evolution is a relentless optimizer of existing designs but a poor redesigner, often stuck with primitive foundations like chemical signaling.
• Human engineering allows for fundamental hardware shifts—from vacuum tubes to transistors—while preserving and transferring the underlying software logic.
• Future intelligent robots will likely combine specialized hardware for efficiency with a general-purpose superstructure for self-improvement.

Darwinian evolution is a relentless optimizer of a given design, nudging the parameters this way and that, adding a step here, removing one there, in a plodding, tinkering, way.

• Comparing computer power is historically contentious due to marketing and the lack of a universal metric.
• Simple metrics like additions per second are easily gamed by specialized, non-functional hardware.
• The author proposes using Claude Shannon's information theory to define computing power as the amount of 'surprise' or information generated per second.
• Information is quantified by the unpredictability of a machine's state transitions; a transition to a highly improbable state conveys more bits.
• A machine only performs useful computation for an observer if that observer does not already know the outcome in advance.
• Total information capacity is determined by the memory size, while processing power is the speed at which a machine steps through these states.

A machine does a useful computation for you only if you don't already know all the answers in advance!

• Computational power is defined by the information transition rate, measured in bits per second and reduced by predictability.
• A computer repeating a loop becomes totally predictable, effectively dropping its computing power to zero.
• High-level languages lower processing power compared to machine language because they produce more stereotyped, predictable instruction sequences.
• Increasing memory size modestly boosts power because the identity of a memory location is considered a 'surprise' to the system.
• In parallel machines, the majority of information surprise resides in the data streams rather than the instruction stream.
• Practical power estimation often requires assuming a standard set of operations and equal probabilities for memory access due to the complexity of real-world statistics.

A computer endlessly repeating a program loop becomes totally predictable, and its computing power drops to zero.

• A formal power formula is established for computing machines, integrating memory capacity, word size, and the speed of addition and multiplication operations.
• Computational power is defined as the speed of the machine, while capacity is defined as its memory size.
• The author employs a nautical metaphor where power is the engine speed, capacity is the fuel tank size, and parallel computing is a fleet of small boats.
• Input/output devices are compared to sails that capture environmental power, potentially providing answers that would otherwise require computation.
• Historical data from 1891 to 1941 tracks the evolution of mechanical and relay-based machines, comparing their power-to-cost ratios against human performance.

Some computations are like a trip to a known location on a distant shore; others resemble a mapless search for a lost island.

• The data tracks the exponential growth of computing power from 1943 to 1979, transitioning from mechanical relays to vacuum tubes and eventually to integrated circuits.
• Early machines like the BTL Model 2 and Colossus operated with speeds measured in fractions of a second, while later systems like the Cray-1 reached nanosecond scales.
• The tables highlight a dramatic shift in the power-to-cost ratio, showing how much more 'bits per second' a dollar could buy as technology matured.
• Memory capacity evolved from a few hundred words in the 1940s to hundreds of thousands of words by the late 1970s.
• The transition to transistors in the late 1950s and microprocessors in the 1970s served as the primary catalysts for massive increases in bits-per-second processing power.

1943 — Colossus (vacuum tube) ... 1976 — Apple II (integrated circuit) ... 1976 — Cray-1 (integrated circuit)

• A table of historical computing power tracks the exponential growth of memory, word size, and processing speed from the Sun-2 to the Cray-3.
• The author posits that a mind is a mathematical pattern that can be instantiated across various physical media without losing its subjective essence.
• The text explores the philosophical conundrum of whether a mind exists as a mathematical abstraction even in the absence of physical hardware.
• A thought experiment suggests that a superintelligent being reading a person's program could simulate that person's existence within its own thoughts.
• This form of mental simulation allows for non-linear processing, such as skipping details or reasoning backwards from a chosen conclusion.
• The distinction between a simulated reality and a fictional narrative blurs when the level of detail in the imagination reaches the threshold of personhood.

Existence in the thoughts of an intelligent beholder is fundamentally no different than existence in a computer simulation, and we have already suggested that a mind can be satisfactorily encoded in a computer.

• A superintelligent being might conserve computational resources by failing to flesh out unimportant details in a simulation, potentially revealing its artificial nature.
• Quantum mechanics suggests a reality where unobserved events exist in a superposition of all possible states, creating observable effects like mysterious coincidences.
• The author explores a potential connection between the 'weirdness' of modern physics and the mechanics of universal simulation.
• The advent of automatic computers shifted scientific focus toward computational complexity and the precise specification of step-by-step procedures.
• Mathematical complexity theory distinguishes between 'easy' polynomial-time problems and those whose difficulty grows exponentially with size.

Just what it all really means is still a matter for fascinating speculation; the only consensus is that the truth is very weird.

• The traveling-salesman problem exemplifies NP problems where the number of possible solutions grows exponentially with each added variable.
• A theoretical nondeterministic machine could solve these problems in polynomial time by splitting into multiple versions of itself at every branch point.
• The P=NP? question remains one of the most significant unsolved problems in computer science, affecting hardware design and automatic reasoning.
• Current design systems rely on approximate methods because exact solutions for NP problems are computationally impossible for conventional machines.
• Future superintelligences may overcome these limits by physically replicating small computing units to act as a massive, real-world multiprocessor.
• The ultimate limit to this computational strategy is the availability of matter and energy, potentially utilizing entire oceans or interstellar clouds.

A hypothetical computer able to explore all possible paths simultaneously (a mere mathematical abstraction known as a nondeterministic machine because it does not make up its mind at branchpoints but splits into two machines and goes both ways) could, in principle, solve the problem in polynomial time.

• A hypothetical machine strategy for solving NP problems involves exponential self-reproduction to match the problem's complexity.
• The proposed process includes a reproduction phase, an individual computation phase, and a 'tournament' phase to find the best result.
• While this method could solve moderately sized NP problems in polynomial time, physical space constraints limit its application to very large problems.
• The author suggests that Earth's biosphere is currently performing a massive computation using this exact reproductive strategy.
• Quantum mechanics offers a potential real-world basis for nondeterministic computing through the principle of superposition and interference.
• The two-slit experiment demonstrates how quantum particles exist in multiple states simultaneously, as the presence of an alternative path changes the final outcome.

If, in one 'generation' time step, a machine can do either a certain amount of computation or reproduce itself once, the best strategy would be to reproduce like mad until there are as many machines as there are alternatives to examine.

• Quantum mechanics describes particles as complex-valued waves that exist in a superposition of all possible locations until a measurement causes the wave function to collapse.
• The 'hidden-variables' theory, supported by Einstein and Schrodinger, argued that quantum states are definite but simply unknown to the observer.
• Modern experiments by Alain Aspect have largely ruled out local hidden-variable theories, confirming the 'absurd' non-local nature of quantum mechanics.
• Schrodinger's Cat was originally a thought experiment intended to highlight the perceived absurdity of a system being in a mixed state of both alive and dead.
• Hugh Everett's 'Many Worlds' interpretation suggests that instead of a wave collapse, the universe branches into different realities at every decision point.

Schrodinger considered absurd the theory's description of the unopened box as a mixed state superimposing a live and a dead cat.

• Quantum measurements cause the universe to split into multiple separate versions that no longer interfere with one another.
• The sheer scale of these infinite universes is often dismissed by physicists as an unnecessary complication to standard quantum mechanics.
• Holographic methods and mixed states offer some computational speedups, but have yet to solve the fundamental difficulty of NP problems.
• The many-worlds theory suggests a form of quantum immortality or survival bias where observers only exist in universes where catastrophe was avoided.
• A fictional scenario illustrates this by showing physicists who survive only because their universe-destroying machine keeps failing by 'bad luck.'
• The idea implies that even if a cosmic disaster occurs in most branches, consciousness persists in the remaining viable timelines.

The word 'astronomical' hardly begins to capture the number of distinct universes created every instant under this idea.

• The author proposes a theoretical 'doomsday machine' based on the many-worlds interpretation of quantum mechanics to prevent nuclear war by destroying any universe where an attack occurs.
• This concept is extended to solving NP-complete problems, like the traveling-salesman problem, by wiring a computer to a universe-destroying device.
• By programming the device to trigger if a computer finds a sub-optimal solution, only universes where the computer finds the correct answer are allowed to persist.
• A binary search strategy can be used to find optimal solutions in polynomial time, effectively turning a conventional computer into a nondeterministic one.
• The ethical and physical cost is the destruction of a staggering number of alternative universes for every one that survives with an answer.
• While the growth rate of new universes might accommodate this, the method relies on the literal reality of branching quantum timelines.

An attack would be met by the destruction of the offending universe.

• The author explores the 'doomsday trigger' concept, suggesting that quantum suicide could theoretically solve NP-complete problems by ensuring survival only in universes where the correct answer is found.
• This radical application of the many-worlds interpretation implies that one could achieve any unlikely desired outcome by destroying themselves in all branches where they fail.
• The demographic argument against frequent quantum suicide suggests that individuals who avoid death exist in a higher proportion of universes than those who are careless with their lives.
• The text questions the validity of the many-worlds theory, proposing that quantum uncertainty might instead be 'noise' from a single, finite universe.
• A mathematical model using Fourier transforms is introduced to describe how wave modes and particle density within a spherical volume might represent universal physical limits.

Wire your computer's doomsday connection to a cranial explosive charge, for instance.

• The text explores the mathematical duality between a physical volume (V) and its frequency representation (F), where both contain identical information.
• In this frequency world, every point in F is a composite of every particle in V, creating a holistic relationship between the two descriptions.
• Nonlinear interactions between particles in V translate into a systematic 'physics' in F, where nearby wave modes exchange energy predictably.
• The author proposes that if the physics of F is sufficiently rich, it could support complex structures, life, and intelligence.
• The laws of F include three-dimensionality, locality based on frequency similarity, and interaction speeds that increase at higher frequencies.
• An uncertainty principle exists in F because determining the energy of a wave mode requires waiting for signals to traverse the entire volume V.

Imagine a physicist made of f stuff, for whom points in f are simply locations, not complicated functions of another space.

• Measurement uncertainty arises because the observer is distributed across a volume, making the exact location of interactions undefined.
• The superposition of states is explained as a statistical sum of cycle-by-cycle interactions influenced by background noise.
• The model mimics general relativity's gravitational time dilation, with time slowing toward the center and stopping entirely at the core.
• Nonlinearities in the medium allow energy to flow between harmonically related wave modes, creating additional degrees of freedom.
• These extra degrees of freedom can be interpreted as the tightly looped extra dimensions found in modern geometric physical theories.

At the very center, time is stopped. The central point of F never changes from its 'average density of the whole sphere' value, and so is effectively frozen in time.

• Higher-dimensional theories suggest that harmonic links between distant regions could enable instantaneous travel and communication across the universe.
• The universe's microwave background radiation can be viewed as a Fourier transform space (F) that interacts nonlinearly with matter.
• As the universe expands and cools, time in the F-space slows down, potentially allowing inhabitants to survive indefinitely by decelerating their subjective experience.
• The early universe's high energy suggests that eons of subjective time could have passed in the F-space during the first microsecond of the Big Bang.
• Communication between the physical world (V) and the Fourier world (F) is nearly impossible, appearing only as universe-spanning 'miracles' or entropy violations.
• The existence of infinite orthogonal transforms implies an infinity of overlapping universes, each with unique physics, sharing the same underlying space.

The first microsecond of the big bang could represent eons of subjective time in F.

• The text provides a comprehensive bibliography spanning the origins of life, cybernetics, and artificial intelligence.
• Key scientific works listed explore the transition from mineral origins to biological organisms and the evolution of human intelligence.
• A significant portion of the references focuses on the intersection of robotics, synthetic psychology, and animal cognition.
• The bibliography includes foundational texts on computing history, from Babbage's engines to the development of digital calculators.
• Neurobiological perspectives are represented through studies on the retina, cellular behavior, and the physical structure of the brain.

The Thinking Computer: Mind inside Matter.

• The text provides a comprehensive list of academic and popular sources spanning neuroscience, computer science, and evolutionary biology.
• Key themes include the molecular basis of memory and the functional relationship between the biological brain and digital computers.
• The bibliography highlights early research into computer viruses, 'worm' programs, and the evolution of cooperation in digital environments.
• It references foundational texts in artificial intelligence and computational geometry, such as Minsky and Papert's work on perceptrons.
• The collection bridges the gap between hard science and speculative fiction, citing authors like Vernor Vinge and Carl Sagan alongside NASA research studies.

The 'worm' programs: early experience with a distributed computation.

• The author traces the book's conceptual roots to childhood influences and early debates at the Stanford Artificial Intelligence Laboratory (SAIL).
• A pivotal 1971 proposal by Dick Fredericksen regarding digital immortality through neural replacement polarized the SAIL community.
• The author's ideas on intelligent machinery crystallized through decades of discussions with prominent researchers at Stanford and Carnegie Mellon.
• The manuscript's development was catalyzed in 1985 by a timely invitation from Harvard University Press editor Howard Boyer.
• The text acknowledges the critical role of science fiction, academic mentors, and government funding from the Office of Naval Research in shaping the work.

Over several articles he had developed the concept of achieving immortality, and much else, by replacing a human nervous system, bit by bit, with a more durable artificial equivalent.

• The author credits manuscript editor Susan Wallace for transforming a 'ragged collection of ideas' into a cohesive, high-quality book.
• The majority of the book's line art was personally created by the author using early Macintosh hardware and various software packages like Cricket Draw and SuperPaint.
• Complex illustrations, such as 'A Robot Bush,' required custom programming in C and ten hours of computation to render a quarter-million line segments.
• The production utilized cutting-edge desktop publishing tools of the era, including Don Knuth's TeX system and Linotronic digital typesetters.
• A diverse range of organizations, including Pixar, the Smithsonian, and MIT, provided specialized photographic and illustrative permissions.
• The inclusion of an index highlights key themes of the work, ranging from biological intelligence to the history of analytical engines.

It contains one quarter of a million line segments and took ten hours to compute.

• The index catalogs the intersection of biological concepts like DNA, bacteria, and evolution with computational advancements such as AI, cellular automata, and computer viruses.
• It highlights key historical figures in computing and science, including Charles Babbage, Alan Turing (via Colossus/Enigma), and Albert Einstein.
• Significant focus is placed on the physical and theoretical components of intelligence, ranging from axons and bipolar cells to expert systems and consciousness.
• The text tracks the progression of robotics and automation through entries like the Stanford Cart, AGVs, and the development of computer vision.
• Broader philosophical and scientific inquiries are represented through topics like the body/mind problem, the Fermi paradox, and black holes.

Boredom, 47, 63, 91, 114, 153, 179

• The index lists foundational concepts in computing and robotics, ranging from early hardware like the Jacquard loom and magnetic cores to modern microprocessors and integrated circuits.
• It highlights the intersection of biology and technology, featuring entries for genetic engineering, the human mind, and biological thought alongside machine intelligence.
• Key figures in science and mathematics are referenced, including Galileo, Gauss, and contemporary thinkers like Douglas Hofstadter and Alan Kay.
• The text covers philosophical and speculative themes such as immortality, the 'many worlds' interpretation of quantum mechanics, and the concept of a 'genetic takeover.'
• Significant corporate and institutional contributors to the digital age are documented, including IBM, Hitachi, General Motors, and the Jet Propulsion Laboratory.

Ghosts, 131-134, 185

• The index highlights a transition from biological systems to a 'postbiological world' through concepts like mind transferral and pattern-identity.
• Key technological milestones are tracked, including the evolution from vacuum tubes and punched cards to nanotechnology and quantum computing.
• The text emphasizes the intersection of robotics and biology, listing entries for nervous systems, neurons, and multicelled animals alongside mobile platforms.
• Theoretical physics and mathematics play a significant role, with references to quantum mechanics, relativity, and nondeterministic polynomial (NP) problems.
• The scope of the work includes the social and evolutionary implications of artificial life, such as the Prisoner's Dilemma, natural selection, and personal identity.

Mind transferral, 115, 117; Mind/body problem, 4, 116; Postbiological world, 1, 5, 125, 141, 145.

• The index covers the evolution of robotics from early projects like Shakey and the Stanford Cart to advanced concepts in superintelligence.
• Key biological and evolutionary themes are present, linking artificial systems to sea slugs, social insects, and the 'Selfish Gene' theory.
• A significant portion of the text focuses on the intersection of computer science and physics, including thermodynamics, quantum superposition, and spacetime.
• The entries detail the technical components of robotics, such as sensors, sonar, stereoscopic vision, and the transition from vacuum tubes to semiconductors.
• Speculative and philosophical topics are indexed, including self-replication, the soul, transmigration, and the ultimate fate of the universe.

Self replication, 135, 150, 165; Self reproduction, 102, 133-135, 151; Selfishness, 141-144; Semiconductor, 71-72, 102.

• The text provides a comprehensive index of topics ranging from biological concepts like vertebrates and viruses to technological milestones like Unix and vacuum tubes.
• Prominent scientific figures such as John von Neumann, James Watson, and Norbert Wiener are referenced, highlighting the book's interdisciplinary scope.
• Critical reviews describe the work as an exhilarating and 'tonic' exploration of the future that challenges readers to look beyond the present.
• The author, Hans Moravec, is identified as a world-class roboticist and Director of the Mobile Robot Laboratory at Carnegie Mellon University.
• Reviewers emphasize the book's unique blend of academic accuracy and the compelling narrative drive of a novel.

Moravec, by his own admission, is an intellectual joyrider, and riding his runaway trains of thought is an exhilarating experience.

• Humanity is on the verge of creating 'postbiological' descendants—intelligent machines that will eventually transcend human complexity.
• These 'children of our minds' may preserve the entirety of human knowledge and even the specific workings of individual human minds.

Unleashed from the plodding pace of biological evolution, the children of our minds will be free to grow to confront immense and fundamental challenges in the larger universe.

• Intelligent machines may soon carry on cultural evolution independently, rendering biological DNA obsolete in the evolutionary race.
• The eventual autonomy of machines in their own reproduction and self-improvement will mark a second complete genetic takeover.

When that happens, our DNA will find itself out of a job, having lost the evolutionary race to a new kind of competition.

• The computer model suggests that a mind's 'process' could theoretically be transferred to new hardware, bypassing biological mortality.
• A world of self-improving 'mind children' would represent a shift in evolution as significant as the transition from chemistry to biology.

A computation in progress—what we can reasonably call a computer's thought process—can be halted in midstep and transferred, as program and data read out of the machine's memory, into a physically different computer, there to resume as though nothing had happened.

• The human brain's sensory and motor systems contain a billion years of evolutionary experience that supports the thin veneer of conscious reasoning.
• The vast majority of human thought is unconscious and inaccessible to introspection, making it the most difficult part of intelligence to formalize.

The deliberate process we call reasoning is, I believe, the thinnest veneer of human thought, effective only because it is supported by this much older and much more powerful, though usually unconscious, sensorimotor knowledge.

Our intelligence, as a tool, should allow us to follow the path to intelligence, as a goal, in bigger strides than those originally taken by the awesomely patient, but blind, processes of Darwinian evolution.

The Hitachi machines climb stairs by rolling up to them on five wheels, raising the leading one to the height of the first stair, driving forward until the raised leg is securely over the step, lowering it slightly until the contact is firm, and then continuing with the next nearest leg.

In repeated cautious lurches, the Cart was to creep safely to its destination.

The robot still crossed the room correctly only about three times out of four.

A robot that spends up to an hour considering every simple move is clearly unacceptable, but a few seconds would be tolerable.

• Future robots will require computing power of at least a billion operations per second, likely achieved within a decade.
• Advanced robots maintain internal world models that allow them to plan actions, learn from the past, and potentially explain their motivations.

If a stairwell-detecting module concludes that hazard is near, it would take over control of the robot until the danger was past.

In these internal models of the world I see the beginnings of awareness in the minds of our machines—an awareness I believe will evolve into consciousness comparable with that of humans.

The fourth door, sadly, leads to the stairwell, and the poor robot, unequipped to travel on stairs, is in mortal danger.

• The author argues that robots, octopuses, and humans exhibit convergent evolution in behavior due to the shared requirements of a mobile lifestyle in a dangerous world.
• Animal consciousness is framed as an internal model of the self and surroundings that allows for the consideration of alternative actions.

The needs of the mobile way of life have conspired in all three instances to create an entity that has modes of operation for different circumstances and that changes quickly from mode to mode on the basis of uncertain and noisy data prone to misinterpretation.

• The author argues that complex traits like consciousness emerge from whole systems and cannot be deduced from individual components.
• Current robots have limited learning capacities, but future machines will require general learning abilities to navigate unpredictable environments.

Emergence—this appearance of novel properties in whole systems has often been invoked to explain such difficult biological realities as mind, consciousness, and even life itself.

There are some dangerous curves in this joyride to human equivalence, so hold on!

• By extrapolating retinal complexity to the whole brain, the author estimates human intelligence requires roughly 10 trillion calculations per second.
• Current supercomputers are approximately 1,000 times slower than the estimated processing power of the human brain.

I rashly conclude that the whole brain's job might be done by a computer performing 10 trillion (10^13) calculations per second.

The second story seems more interesting and informative because its later statements are less likely—cats usually have fur and claws, but they rarely carry hats and guns.

• The scanning tunneling microscope provides a critical 'toehold' on the atomic scale, allowing for the direct manipulation of individual atoms.
• Future nanotechnology aims to fuse biology and microelectronics to create protein robots capable of building materials atom-by-atom.

These would be small enough to grab individual molecules and hold them, thermally wriggling, in place.

Instead of a rat's nest of wires, a program consisted of neat columns of numbers.

Each finger of the glove contains a grid of elements that create patterns of pressure and temperature on the finger of the wearer.

The operator has the subjective sensation of being in the robot's body.

Einstein, for instance, reported that he could often feel the meaning of his equations in his arms and his body as if they were solid objects.

• Human intelligence is not an upper bound, and machines will inevitably outclass biological cognition as they scale.
• Future space-based industries will likely consist of self-replicating robot factories that outpace human population growth.

The universe is one random event after another.

These new creations, looking quite unlike the machines we know, will explode into the universe, leaving us behind in a cloud of dust.

• Mind transfer could occur gradually as a computer replaces fading biological functions, eventually hosting the entire consciousness.
• Digitized minds could manipulate their own 'speed' settings, allowing them to think and react thousands of times faster than biological humans.

Ultimately your brain would die, and your mind would find itself entirely in the computer.

• Identity is defined as the information pattern of a person rather than the physical medium on which it is encoded.
• A simulated mind could remain continuous and uninterrupted even while its underlying physical location and hardware change constantly.

Old instincts are not automatically erased when the rules of life are suddenly rewritten.

• A superdense simulator could model Earth at an atomic scale, creating simulated people who are as real as biological ones under the pattern-identity theory.
• The postbiological world will evolve rapidly, ranging from tiny intelligences to star-spanning superminds that function like cooperative ant colonies.

Resurrecting one small planet should be child's play long before our civilization has colonized even its first galaxy.

Today's computer systems are like bodies with skins but no immune systems, or like walled cities without police.

• Sexuality provides a long-term evolutionary advantage by accelerating the rate of genetic variation despite the high immediate cost of reproduction.
• The primary driver for sexual reproduction is the threat of disease, as genetic diversity prevents parasites from wiping out entire populations of identical clones.

A parasite that has the key to one lock finds that the next one is subtly different and thus harder to open.

Each Martian gives the other a broken unit and leaves the meeting gloating over having made a shrewd deal. But as night falls, both Martians find themselves in the dark.

• Postbiological entities with long memories and high intelligence are likely to favor cooperation over conflict, as interactions are rarely seen as final.
• As the universe cools toward absolute zero, the energy required for computation and signal transmission decreases, allowing for more thought with less power.

In time the entire universe will become a homogeneous stew with no concentrations of matter or energy to form or power any kind of machinery, intelligent or otherwise.

• Future machine intelligence will likely balance pure cerebration with active exploration and massive engineering projects.
• The discovery of DNA revealed that biological life uses a 'tape' system for construction similar to von Neumann's theoretical machines.

In a final step a "breath of life" signal is transmitted to the painting that converts its quiescent states to active ones.

A single glider advancing across the screen would cause a display where gliders would appear and disappear in odd places almost at random, sometimes several in view, sometimes none.

Darwinian evolution is a relentless optimizer of a given design, nudging the parameters this way and that, adding a step here, removing one there, in a plodding, tinkering, way.

Some computations are like a trip to a known location on a distant shore; others resemble a mapless search for a lost island.

1943 — Colossus (vacuum tube) ... 1976 — Apple II (integrated circuit) ... 1976 — Cray-1 (integrated circuit)

Existence in the thoughts of an intelligent beholder is fundamentally no different than existence in a computer simulation, and we have already suggested that a mind can be satisfactorily encoded in a computer.

Just what it all really means is still a matter for fascinating speculation; the only consensus is that the truth is very weird.

• The traveling-salesman problem exemplifies NP problems where the number of possible solutions grows exponentially with each added variable.
• Future superintelligences may overcome these limits by physically replicating small computing units to act as a massive, real-world multiprocessor.

A hypothetical computer able to explore all possible paths simultaneously (a mere mathematical abstraction known as a nondeterministic machine because it does not make up its mind at branchpoints but splits into two machines and goes both ways) could, in principle, solve the problem in polynomial time.

If, in one 'generation' time step, a machine can do either a certain amount of computation or reproduce itself once, the best strategy would be to reproduce like mad until there are as many machines as there are alternatives to examine.

• Modern experiments by Alain Aspect have largely ruled out local hidden-variable theories, confirming the 'absurd' non-local nature of quantum mechanics.
• Hugh Everett's 'Many Worlds' interpretation suggests that instead of a wave collapse, the universe branches into different realities at every decision point.

Schrodinger considered absurd the theory's description of the unopened box as a mixed state superimposing a live and a dead cat.

• The many-worlds theory suggests a form of quantum immortality or survival bias where observers only exist in universes where catastrophe was avoided.
• The idea implies that even if a cosmic disaster occurs in most branches, consciousness persists in the remaining viable timelines.

The word 'astronomical' hardly begins to capture the number of distinct universes created every instant under this idea.

Over several articles he had developed the concept of achieving immortality, and much else, by replacing a human nervous system, bit by bit, with a more durable artificial equivalent.

• The index highlights a transition from biological systems to a 'postbiological world' through concepts like mind transferral and pattern-identity.
• Theoretical physics and mathematics play a significant role, with references to quantum mechanics, relativity, and nondeterministic polynomial (NP) problems.

Mind transferral, 115, 117; Mind/body problem, 4, 116; Postbiological world, 1, 5, 125, 141, 145.

• The index covers the evolution of robotics from early projects like Shakey and the Stanford Cart to advanced concepts in superintelligence.
• Speculative and philosophical topics are indexed, including self-replication, the soul, transmigration, and the ultimate fate of the universe.

Self replication, 135, 150, 165; Self reproduction, 102, 133-135, 151; Selfishness, 141-144; Semiconductor, 71-72, 102.