Book summary “Out of control”: summary

Introduction

You face a world where the technologies you build and manage are growing too complex to control through top-down commands. Traditional mechanical logic fails as software systems, global networks, and artificial intelligence begin to behave more like wild organisms than predictable machines.

Kevin Kelly argues in Out of Control that the twenty-first century's most successful systems must abandon centralized control. You must adopt the decentralized, bottom-up logic of living nature to build adaptive, self-sustaining technologies.

The book investigates how complex systems emerge across several domains:

• How do simple decentralized agents form intelligent collectives without central direction?
• Why do autonomous machines outperform rigid programmed structures in uncertain environments?
• Can you engineer closed ecological loops and open-ended artificial evolution inside digital simulations?

1. The Made and the Born

The realms of nature and human technology are merging into a unified domain where mechanical systems acquire biological traits and life is engineered. As human-made environments grow increasingly complicated, you must turn to nature to understand how to manage them. The mechanical world must adopt biological traits to survive and function.

The realms of nature and human technology are merging into a single, unified domain where machines become biological and life is engineered.

Kevin Kelly calls this emerging reality a neo-biological civilization. You live at the intersection of two distinct realms:

• The realm of the born, representing organic nature;
• The realm of the made, representing human technology.

The Neo-Biological Civilization

Traditional machines rely on simple clockwork logic to operate. Highly complex systems like cells, meadows, economies, or brains demand nontechnological, biological logic to assemble and operate. You cannot manage massive technological networks with rigid mechanical rules.

Technicians can abstract the laws of life and apply them to technology. You can achieve several fundamental traits by transferring them from living systems to machines:

• Self-replication;
• Self-governance;
• Limited self-repair;
• Mild evolution;
• Partial learning.

The author describes living inside an experimental airtight glass cottage where a system of plants, soils, and machinery recycles his exhalations, urine, and excrement into fresh air, water, and food. This closed habitat demonstrates how manufactured technology and biological processes unite into a single operational cycle.

Vivisystems and Autonomy

Artificial and natural systems that exhibit lifelike properties can be collectively categorized as vivisystems. Human-made things are behaving more lifelike while life is becoming more engineered. This reveals a common soul between organic communities and manufactured counterparts.

Examples of artificial vivisystems include planetary telephone systems, computer virus incubators, robot prototypes, virtual reality worlds, synthetic animated characters, artificial ecologies, and whole Earth computer models. These complex creations possess autonomous behaviors that resist direct programming.

When you unleash living forces into artificial machines, you inevitably lose absolute sovereignty and control over your creations. When the union of the born and made is complete, fabrications will learn, adapt, heal, and evolve autonomously, acquiring the wildness and surprises inherent in living systems.

2. Hive Mind

Complex collective intelligence and emergent properties arise spontaneously from the simple local interactions of distributed members in a superorganism.

William Morton Wheeler demonstrated that an insect colony behaves as a unitary whole maintaining identity and spatial organization. When a beehive swarms, the queen does not lead. Worker bees democratically select a new home site through competing dances where popularity builds upon itself via the law of increasing returns.

Kevin Kelly opened a hollow tupelo tree filled with 100,000 bees. He found that the mass of cold-blooded insects generated a warm-blooded temperature of 95 degrees Fahrenheit, turning honey into flowing warm blood.

Emergence and Swarms

High levels of complexity and connectivity among simple parts produce entirely new qualitative properties that cannot be inferred from the parts alone. Complex systems exhibit a different kind of causation where quantitative accumulation transforms into qualitative novelty. Running a system in practice is the only way to reveal the patterns dissolved in its parts.

A single molecule of water has no temperature, and one grain of sand cannot form an avalanche. Collections of them produce distinct group characteristics like temperature and sand dunes.

Craig Reynolds developed computer algorithms where each virtual entity follows three basic rules:

• avoid bumping into neighbors;
• keep up with neighbors;
• do not stray too far away.

This proves that flocking behavior is an adaptive trick emerging from distributed rules rather than centralized life signs. Animator Craig Reynolds used these simple algorithms to generate realistic swarms of bats and marching penguins for the film Batman Returns.

Loren Carpenter used custom software to track thousands of audience members waving wands with red and green sides. By averaging individual inputs as votes, the crowd collectively controlled game paddles and flight simulators in real time. At a 1991 computer graphics conference, nearly 5,000 attendees collectively played Pong and successfully flew a simulated airplane, eventually executing a 360-degree roll as a single emergent mind.

The Net as New Science Icon

Distributed swarm systems offer supreme adaptability and resilience at the cost of being nonoptimal, uncontrollable, unpredictable, and ununderstandable. While sequential clockware systems are efficient and easy to control, parallel swarmware systems provide redundancy and spontaneous order. However, their horizontal causality makes them inefficient, opaque to human understanding, and slow to boot up.

An engineer at Digital Equipment Corporation demonstrated the resilience of a distributed network by yanking a cable out of its computer network, causing the system to instantly route around the breach without faltering.

The dynamical Net replaces the Atom as the central icon of science, symbolizing complexity, interdependence, and control without authority. The planetary atom represents singleness, predictability, and reductionist power. The net represents multiples, distributed being, and the invisible hand.

Charles Darwin used the image of an entangled bank and a tangled net at the end of Origin of Species to describe how biological species emerge from the interconnected self-interests of many individuals. Networks allow for infinite rearrangements, the absorption of novelty, and the preservation of imperfection through small failures that prevent large catastrophes.

3. Machines with an Attitude

Traditional university labs built massive robots that rely on heavy central computers and complex internal world models, making them too slow, bulky, and fragile for practical deployment. Carnegie Mellon University's Field Robotics Center spent millions of tax dollars to build Ambler, a 19-foot-tall, two-ton robot that spent most of its time worrying about mapping the terrain rather than walking. In 1990, CMU graduate students wired together Ambler, and it managed to walk 100 feet across a courtyard.

Massive centralized robots fail in the real world, whereas decentralized, disposable machines using subsumption architecture achieve efficient autonomous movement.

Rodney Brooks proposed an invasion of space using armies of cheap, shoe-box-sized robots that rely on simple onboard logic rather than remote control or heavy central brains. In 1989, Rodney Brooks published his paper Fast, Cheap and Out of Control. NASA's Jet Propulsion Laboratory adopted this philosophy to create micro-rovers like the solar-powered six-wheeled dune buggy scheduled for the 1997 Mars Environmental Survey.

Bottom-Up Robotics

Artist Mark Pauline builds complex, autonomous iron machines that function as biological predators within civilization. Pauline creates massive, intricate machines out of military surplus and salvaged parts, staging destructive performances to reintroduce predatory dynamics into stable human ecosystems. In 1979, Pauline debuted his performance called Machine Sex. In 1991, Pauline staged a machine circus in San Francisco where mechanical gladiators like the Screw Throwbot and the 50-foot flamethrower demolished each other. The Screw Throwbot traveled at 30 miles per hour and hurled 50-lb. firebombs, using military surplus parts purchased for 65 dollars per pound under the name Obtainium through Survival Research Labs.

Subsumption Architecture

Complex behaviors emerge from the bottom up through modular layers of simple reflexes without a central controller. Rodney Brooks built Genghis, a six-legged cockroach-like robot with 12 motors and 21 sensors distributed in a network without a central brain, weighing 3.6 pounds, where walking naturally emerged. Subsumption architecture connects sensors to actuators through independent layers of behaviors where higher layers subsume lower ones without altering or destroying the working foundation.

You can build decentralized behaviors by following these engineering rules:

• Do simple things first.
• Learn to do them flawlessly.
• Add new layers of activity over the results of the simple tasks.
• Don't change the simple things.
• Make the new layer work as flawlessly as the simple.
• Repeat, ad infinitum.

A physical body equipped with sensory organs is essential to anchor the mind and prevent it from descending into hallucinations. Without the constant interruptions of urgent external data from senses like sight and touch, a brain's neural circuitry exhausts itself by playing games with its own internal logic. In D.O. Hebb's 1954 sensory deprivation experiments at McGill University, volunteers placed in dark, soundproof cells with restricted mobility quickly developed visual delusions and cognitive failure, reflecting the black patch psychosis documented in a 1923 hospital report on cataract patients.

4. Coevolution and Ecological Assembly

Complex ecosystems and artificial machines must be grown incrementally through feedback loops, temporary scaffolding, and specific developmental sequences.

Ecosystem Restoration

Aldo Leopold attempted to recreate a prairie in 1934 by planting seeds on an old farm, but the plot eventually devolved into an overgrown weedy lot because a crucial missing element—fire—was absent. Fire clears intruding tree saplings and triggers specific seeds, proving that ecological restoration requires the correct ingredients and operational forces. Aldo Leopold initially argued fiercely against letting wildfires burn in 1920, only to discover a decade later that organic fire is vital for prairie survival. Leopold nursed a 300-acre young emerging prairie using buckets of water at the University of Wisconsin arboretum.

Steve Packard struggled to restore suburban Chicago plots until he realized the encroaching plants belonged to a savanna ecosystem rather than a prairie. By collecting and sowing lumpy, oozy, glop-like seeds of savanna species from old cemeteries and railway tracks, he successfully catalyzed a self-sustaining system. Steve Packard discovered unknown blazing star flowers and rare plants like the white-fringed orchid and pale vetchling spontaneously appearing once the right savanna conditions were met. In 1988, a drought shriveled non-native weeds while reseeded natives flourished. By 1989, a pair of eastern bluebirds settled into their familiar restored habitat.

Feedback and Sequence

Stuart Pimm and Jim Drake conducted laboratory experiments with microecosystems in flasks and computer simulations to test random assembly. They discovered that while systems always find a persistent and structured state rather than wandering indefinitely, changing the introduction sequence of even a single species shifts the system to a completely different endpoint. Stuart Pimm and Jim Drake used microecosystems in flasks containing 15 to 40 different pure strains of algae and microscopic animals to chart the importance of sequence. Aquatic mixtures took 10 to 15 days to form a stable slime ecology. Computer models using small mixes of 10 to 20 components yielded tens to hundreds of stable peaks through anti-chaos.

David Wingate worked to save the nearly extinct cahow seabird in Bermuda by moving to an uninhabited island. Because you cannot wait for a full forest to grow naturally, you must build artificial underground nest boxes, plant 8,000 cedar trees, and use casuarinas as temporary windbreak scaffolding. David Wingate planted fast-growing casuarinas as a windbreak to protect vulnerable cedar trees, which eventually fostered night herons that controlled pest land crabs and allowed rare sedges to grow. The cahow seabird was rediscovered in 1951 on the outer cliffs of the Bermuda archipelago.

When given a slim foothold, interconnected green things launch the law of increasing returns, causing a snowballing effect where adding subsequent biological pieces becomes progressively easier. Wendy Campbell-Purdy planted a green wall of 2,000 trees across 45 acres in Tiznit, Morocco. She later financed 130,000 more trees on a 260-acre desert dump in Algeria, successfully transforming a desert dump into a microhabitat suitable for citrus, vegetables, and grain.

5. Closed Systems and Synthetic Ecology

Stable closed ecologies depend on microbial foundations, species redundancy, and synthetic ecology principles that surrender direct human control.

You build stable closed systems by recognizing that microbes do the lion's share of breathing, generating air, and supporting larger organisms. Clair Folsome sealed brackish water microbes in glass flasks in Hawaii and showed that they maintained stable oxygen and carbon cycles for up to 25 years without any outside air or food.

Microbial Foundations

Artificial complex ecosystems must be assembled in a specific developmental sequence, starting with foundational microorganisms before adding higher species. Lloyd Gomez spent five years developing a 16,000-gallon synthetic coral reef display at the Steinhart Aquarium in San Francisco, supported by heavy industrial plumbing, sand beds, and halide lamps.

You must follow these rules when you build synthetic ecologies:

• Microorganisms do most of the work.
• Soil is an organism. It is alive. It breathes.
• Make redundant food webs.
• Increase diversity gradually.
• If you can't provide a physical function, you need to simulate it.
• The atmosphere communicates the state of the whole system.
• Listen to the system; see where it wants to go.

Biosphere Construction

Creating a functioning large-scale biosphere requires combining multiple diverse biomes and vast species redundancy. Walter Adey built a one-million-gallon ocean and coral reef inside Biosphere 2 by hauling in bags of Instant-Ocean salt and a starter microbial solution from the Pacific Ocean, alongside chunks of Florida Everglades mangrove swamps.

You can manage the carbon economy of such a system with specific actions:

• Rouse a dormant savanna or desert into spring by rising temperatures and artificial rains to suck up CO2.
• Prune old growth to stimulate new CO2-consuming growth.
• Pile dried grass clippings on soil as mulch and wet them to release CO2 via bacterial decomposition.

Complex closed systems undergo an unpredictable transition called a pop when microbes and species finally self-organize into a stable community. Saltwater aquarium hobbyists observe that it takes about six weeks of balancing water and bacteria before a tank suddenly pops to zero ammonia.

6. Artificial Evolution

Simulated computer worlds running self-replicating code, mutation, and death can spark open-ended evolution and generate solutions surpassing human engineering.

As software systems grow to millions of lines of code, traditional engineering correctness must be traded for artificial evolution, sacrificing total control to ensure perpetual survivability. Long-distance phone call-routing programs span 2 million lines of code where a few faulty lines caused national outages. Combat computers aboard the Navy's Seawolf submarine reach 3.6 million lines of code, and Microsoft's NT operating system released in 1993 required 4 million lines of code.

Tierra and Virtual Life

Tom Ray built a shadow computer running inside a real machine to contain his experiments without crashing the host. He seeded it with an 80-byte self-replicating computer virus and added two key features: a mutator that occasionally scrambled digital bits during copying, and an executioner program named the Reaper that killed the oldest or most broken creatures, thereby introducing natural selection and death.

Tom Ray seeded his Tierra world with a single 80-byte creature. Within minutes the RAM was saturated, and creatures mutated into smaller, more efficient forms, including a 45-byte parasite that depended on 80-byte hosts, and eventually a fully self-replicating 22-byte creature that surpassed human engineering skills.

Genetic Algorithms

Invented by John Holland, genetic algorithms combine sexual mating and mutation across huge populations of code processing in parallel. This implicit parallelism allows the system to sample thousands of combinations concurrently, avoiding local peaks and successfully locating global optima in complex problems.

John Holland spent years developing genetic algorithms after being inspired by R. A. Fisher's mathematical approach to population genetics, creating a foundation that was initially ignored due to the dominance of von Neumann serial programming. His book Adaptation in Natural and Artificial Systems was published in 1975 and sold 2,500 copies until it was reissued in 1992.

You can use evolution as a practical tool across three domains:

• How to get somewhere you want but can't find the route to.
• How to get to somewhere you can't imagine.
• How to open up entirely new places to get to.

7. The Edge of Chaos

Complex decentralized networks achieve maximal adaptability and open-ended evolution when balanced precisely on the phase transition between order and chaos.

Stuart Kauffman modeled genetic systems where 10000 genes were connected at random to turn each other on or off, finding that the ensembles consistently fell into predictable basins of behavior. Order arises for free from the topology of heavily interconnected networks without requiring external design.

Spontaneous Order

Chris Langton derived the lambda parameter to predict the likelihood that rule sets will produce interesting behavior and locate the phase transition sweet spot. Systems balanced at this edge can evolve, learn, and adapt most readily.

Kauffman tested generic networks and found that large networks with thousands of members adapted best with less than ten connections per member, and sometimes less than two. Ultra-connected networks suffer from a complexity catastrophe where conflicting influences paralyze the system.

Poised Systems

You can apply these principles to guide adaptation in complex environments:

• Seek persistent disequilibrium to balance stability and change on the edge of chaos.
• Change changes itself by allowing bottom-up interactions to alter the rules for changing entities.

10 Key Ideas

1. The realms of nature and human technology are merging into a single, unified domain where machines become biological and life is engineered.
2. Complex collective intelligence and emergent properties arise spontaneously from the simple local interactions of distributed members in a superorganism.
3. Massive centralized robots fail in the real world, whereas decentralized, disposable machines using subsumption architecture achieve efficient autonomous movement.
4. Complex ecosystems and artificial machines must be grown incrementally through feedback loops, temporary scaffolding, and specific developmental sequences.
5. Stable closed ecologies depend on microbial foundations, species redundancy, and synthetic ecology principles that surrender direct human control.
6. Simulated computer worlds running self-replicating code, mutation, and death can spark open-ended evolution and generate solutions surpassing human engineering.
7. Complex decentralized networks achieve maximal adaptability and open-ended evolution when balanced precisely on the phase transition between order and chaos.
8. When given a slim foothold, interconnected green things launch the law of increasing returns, causing a snowballing effect where adding subsequent biological pieces becomes progressively easier.
9. Artificial and natural systems that exhibit lifelike properties can be collectively categorized as vivisystems because human-made things are behaving more lifelike while life is becoming more engineered.
10. The dynamical Net replaces the Atom as the central icon of science, symbolizing complexity, interdependence, and control without authority.