The Cybernetics Family Tree
I wondered where cybernetics would sit on my O'Reilly shelf. It has no shelf of its own. It sits under all of them. A checked family tree, from the governors to Kubernetes, with your shelf on it.
I was looking over my bookshelf on O'Reilly Learning, the one with the data science and the computer science and the engineering on it, and I asked myself where cybernetics would sit.
I put the question to an agent, and the answer that came back was that it doesn't get a shelf. It's the shelf under the shelves. Several of the categories on that screen split off from it after a run of meetings in New York in the late forties. The children took new names, and the parent dropped out of the catalog.
Then it drew me a rough chart, and the chart opened with four questions:
| The field | The question it asks |
|---|---|
| Computer science | What can be computed? |
| Data science | What does the data say? |
| Engineering | How do we build it so it holds? |
| Cybernetics | How does a system keep steering toward a goal when the world pushes back? |
That last one is regulation, not computation. A thermostat hardly computes anything. It watches the room, compares it with the setting, nudges the heat, and watches again. That loop is the whole field in small.
I liked the chart so much I wanted the real one. Who came from whom, which lines are written down in the record and which are only claimed, and for every shelf on today's catalog, what to read first. This post is that map.
Checked, not yet read
The rule from my reading list holds here too: checked is not read. Every box and every line below was checked against a source, the papers themselves and the histories of the field, by an agent working for me. Most of those I have not read yet. I've read Gleick, and some of Donella Meadows, and watched some of Paul Pangaro's videos. I'm nearly through the companions and primers I made for the papers, and now I'm attempting the papers themselves. Yesterday I began Maxwell's paper on governors, and once again I found I need to learn his language to understand what he is saying: calculus and differential equations, and the algebra of the roots that his question about stability turns into. So the reading path has a math layer now too. So this is a map drawn before the walk, the way the primers were, and the histories it leans on are listed at the bottom.
The book of Wiener's I read as a teenager is in here, near the middle. I wrote in the culture essay that I did not think much about its subtitle then, Control and Communication in the Animal and the Machine. I think about it a lot now.
The tree
The map goes on to the right: swipe it sideways.
documented descentpartial: shared ideas, some contactcontested: claimed and disputeda roota shelf todayone I've read or watched
Read it as a history of ideas: the colored bands on each box are the older disciplines it came out of, my own rough sort. Pick one to see where it went.
Lineages run down, time runs across. Tap a box, or tab to it and press Enter, and its whole line lights up, where it came from and what came from it, with its entry under the drawing: the people, the idea, what to read first, and where the line is contested. Every entry is also in the whole tree in words. On a phone the drawing scrolls sideways.
Time runs across, from the engineering and physiology the founders drew on, through the roots and the first branches, to the shelves they turned into. The lineages run down the page in bands: control, regulation, learning machines, information, people and computers, organizations, systems thinking, life, mind, complexity. A solid line means the descent is in the record, a student, a citation, a seat at the same table. Dashed is partial, shared ideas and some contact. Dotted is contested: someone claims the line and someone else says no, and the entry below says who.
A few boxes have no line at all. Ampère used the word for the art of governing long before Wiener, and Wiener coined it again on his own from the same Greek, the steersman. A namesake is not a parent, so it sits there alone.
Where the name went
The parent dropped out of the catalog partly because one of the children changed its name on purpose. John McCarthy, who named artificial intelligence for the Dartmouth summer project, wrote years later that one of his reasons for the new term was "to escape association with 'cybernetics'," and to avoid having either to accept Wiener as a guru or to argue with him. Information theory kept its distance too. Shannon himself warned in print about the cybernetics bandwagon.
Other branches kept the word and went somewhere else with it. In the Soviet Union it was condemned as a reactionary pseudoscience and then brought back almost at once, and it grew into the plan for a national computer network to run the economy, which was never built. In Chile, Stafford Beer and a team in Santiago built Project Cybersyn, a telex network and an operations room for the nationalized industries, until the coup ended it.
Then the field turned around and looked at itself. Margaret Mead asked the new American Society for Cybernetics to turn its tools on its own workings. Heinz von Foerster gave it a name, second-order cybernetics, the cybernetics of observing systems, where the one doing the describing is inside the description. That branch is where my listening is now. It's where a page of my freewriting caught me in the act, listening to the primer on Pask while working with an agent, in Arguing and Agreeing:
this all seems very much like Paskian conversation theory where I seek to reach agreement with my cognitive partner. Arguing and agreeing.
It has been out of the public light for a long time, and its children are everywhere.
Find your shelf
If you work on one of today's shelves, here is your branch: the cybernetic roots it grew from, the question it inherited, the one thing to read first, and the primer here on the site where there is one.
| Your shelf today | Its cybernetic roots | The question it inherited | Read first | On this site |
|---|---|---|---|---|
| Robotics and Control Systems | Ratio Club, Grey Walter's tortoises; Wiener, Cybernetics | How does a body keep acting sensibly when the world will not hold still? | Rodney Brooks, Intelligence Without Reason (1991) | David Silver's robot bug |
| DevOps and SRE | Wiener, Cybernetics; Ashby, requisite variety | How does a system keep what is running matched to what was asked for, when things keep breaking? | The Kubernetes project, Controllers (current docs) | The Regulator Was Fed Intents A Loop Is a Path That Closes |
| AI and Machine Learning | A Logical Calculus of neurons | Can a network of simple units learn from examples what no one could write down as rules? | Rumelhart, Hinton and Williams, Learning representations by back-propagating errors (1986) | not yet |
| AI and Machine Learning | Wiener, Cybernetics; Ratio Club, Grey Walter's tortoises | How does something learn to steer when the only feedback is whether things went well? | Sutton and Barto, Reinforcement Learning: An Introduction, chapter 1 (2018) | not yet |
| Data Science | A Mathematical Theory of Communication | How much does a message, or a dataset, actually tell you? | James Gleick, The Information (2011) | my review of The Information |
| UX and Design | The Macy conferences; Wiener, Cybernetics; Ashby, requisite variety | What makes a machine a partner in a conversation instead of a vending machine? | Dubberly and Pangaro, How cybernetics connects computing, counterculture, and design (2015) | the primer for Dubberly and Pangaro, 2015 |
| Software Architecture and Organizations | Ashby, requisite variety | How does an organization stay whole while its parts act on their own? | Dan Davies, The Unaccountability Machine (2024) | Gentler On-Ramps to Cybernetics |
| Systems Thinking | Wiener, Cybernetics (contested); The Macy conferences | Why does a system keep producing the behavior nobody wants? | Donella Meadows, Thinking in Systems (2008) | first on the gentler shelf |
| Psychology and Therapy | The Macy conferences | What loop keeps a problem alive, and what small change would break it? | Bateson, Jackson, Haley and Weakland, Toward a Theory of Schizophrenia (1956) | not yet |
| Cognitive Science | The Macy conferences; Ashby, requisite variety; Craik, The Nature of Explanation | Where does a mind stop: at the skin, at the hand, at the notebook? | Andy Clark and David Chalmers, The Extended Mind (1998) | not yet |
| Neuroscience | Ashby, requisite variety; Craik, The Nature of Explanation | Is a brain a regulator that predicts its world in order to keep itself steady? | Anil Seth, The Cybernetic Bayesian Brain (2015) | not yet |
| Complexity Science | Wiener, Cybernetics (contested); The Macy conferences; Ashby, requisite variety; A Mathematical Theory of Communication | How does order arise from many parts that no one is steering? | Melanie Mitchell, Complexity: A Guided Tour (2009) | Making Complexity Visible the cybernetics reading path |
The shelf names are written the way a bookshop would file them, in O'Reilly's style. A few of them, cognitive science and neuroscience and therapy, aren't on O'Reilly at all, and they still belong to the family.
My own work sits on the regulation row and the organizations row, the work I described on the reading list as systems design, DevOps and architecture modeling. Both of them come down from Ashby. I hadn't made the connection yet between cybernetics and the work I was doing: the loops, the AI, Kubernetes and its controllers. Now it all feels very good, like a coming home, and also an awakening. It is very fun, cerebrally.
Where the lines are contested
Some of the most tempting lines are the weakest ones, and I drew one of them myself.
Kubernetes and Ashby. In The Regulator Was Fed Intents I set a Kubernetes reconciliation loop beside Ashby's law, and the law didn't apply the way I expected. As an idea the pairing still helps me. As descent it's partial. The Kubernetes docs explain a controller as a control loop, with a thermostat for the example, so its line runs through control engineering. What it shares with cybernetics is the name, the same Greek word for the helmsman.
Data science and Shannon. Data science's main ancestor is statistics. What it got from this family is Shannon's measure of information, the bits and the entropy, and the cross-entropy most models are trained on.
System dynamics. Jay Forrester came out of the servomechanisms lab at MIT, and George Richardson argues his kind of feedback thinking came down a separate line from cybernetics. Donella Meadows and the systems-thinking shelf sit on that line, so their link to the movement is dotted, while their link to Forrester is solid.
Conway's law. It sits on the organizations shelf and arrived there on its own. I found no line to draw.
Complexity science. The Santa Fe Institute's founders came from physics and Los Alamos and don't claim cybernetics as a parent. Some historians read complexity science as its continuation anyway. Dotted.
Luhmann and ecosystems. Luhmann borrowed autopoiesis for society, and Maturana, one of the two people who coined it, rejected that use. Ecologists argued in the same journal over whether an ecosystem is cybernetic at all, since nothing in it holds a set point.
Then there are the lines the descendants drew themselves. Karl Friston cites Conant and Ashby's good regulator theorem by name in his work on the free energy principle. Rodney Brooks gave Grey Walter's tortoises a section of their own when he looked back at where his robots came from. Those are solid, and the people at the far end of them made them so.
If you only read three
- Rosenblueth, Wiener and Bigelow, "Behavior, Purpose and Teleology." A short paper, and the place where the idea that purpose is feedback first gets said out loud. The primer walks it step by step.
- Ashby, An Introduction to Cybernetics, chapter 11. The law of requisite variety, which nearly every branch on the regulation side inherits. Free to read, with a primer.
- Ronald Kline, The Cybernetics Moment. The history of how one conversation scattered into these shelves, which is this tree told as a story.
The first two are at the top of my own reading list, and the third is the history I want open beside them. For a shortcut that costs nothing, Hugh Dubberly and Paul Pangaro's essay How cybernetics connects computing, counterculture, and design draws a family tree of its own, from the design side.
Going in through the history
I come into a new field through its history and its people first. I said it on the gentler shelf, "I think that is why the history is the right door for me," and the freewriting from the same month says it louder:
Given my love for history of science I am perfectly "primed" and filled with excitement to move to the actual papers and books and lectures.
My degree is in Comparative History of Ideas, and it stuck. For me the history and the people make a field more interesting: how they came to their thoughts, what their lives were like, how they came to be. It all fits together. I like seeing the whole picture and building the map, and I find it fun. And for me anyhow, it gives the more abstract things an anchor. What pulls me in most is the link between them all: computers and the mind, neurons and mathematics, electronics and systems. That is the table at the Macy conferences, a neurophysiologist and a logician and a mathematician talking about one loop.
There's a longer story behind that habit, and it gets its own post.
Where this sits on the site
This map is the overview for the cybernetics reading path, which has everything else in order. If you want a gentler way in than the papers, start on the gentler shelf. The papers themselves, with my guesses written down before I read them, are on the reading list, and the primers for them are in A Map Before the Reading and The Rest of the Map.
A map of a whole literature is the same move as Domain Cartography: a body of reading is a system you didn't build, and you chart it before you work inside it. The reason for charting at all is in Making Complexity Visible.
The branches show up in other posts too. The robot bug in The Hacker Ethic Meets Cybernetics is the robotics row, look, compare, nudge, look again. Shannon's row is the subject of my review of Gleick's The Information. And Level Up the Game is about what happened when I tried to read this field with an agent beside me.
The whole tree in words
The same tree as the drawing, era by era, with every link's strength and the reason when it is partial or contested.
Before the name 1788 to 1934
Watt's governor
Spinning flyballs rise as the engine speeds up and close the steam valve, so the engine holds its own speed.
Watt adapted a flyball governor millwrights already used on windmills; he did not invent it. Older still: the float regulator in Ktesibios's water clock, third century BCE, and Drebbel's self-regulating furnace, around the 1620s.
Read first: Otto Mayr, The Origins of Feedback Control (1970).
Maxwell, On Governors
The first mathematics of when a regulator settles down and when it hunts back and forth.
Grew from Watt's governor.
Wiener names this paper in the introduction to Cybernetics as the first significant paper on feedback mechanisms.
Read first: James Clerk Maxwell, On Governors (1868).
Written in: logarithms, polynomials and complex numbers, calculus, differential equations.
I've begun this one, and it sent me to learn its math.
Feedback amplifiers
Feed part of the output back, inverted, and you trade raw gain for steadiness; Nyquist showed how to tell when such a loop will ring.
Grew from Maxwell, On Governors (partial): the same stability problem met again in electronics, at Bell Labs.
Black had the idea on a ferry in 1927 and published it in 1934; Nyquist's stability test is 1932; Bode's book is 1945. Black was not part of the cybernetics movement, but Wiener drew on the engineering (Mindell).
Read first: Harold S. Black, Inventing the negative feedback amplifier, IEEE Spectrum (1977).
Written in: logarithms, polynomials and complex numbers, differential equations.
Servomechanisms at MIT
A theory of motors that follow a command, and the wartime fire-control work that grew from it.
Grew from Maxwell, On Governors (partial): part of the control engineering line Bennett and Mindell trace.
Hazen later ran the wartime fire-control division that funded Wiener and Bigelow's anti-aircraft predictor. Jay Forrester trained in Brown's Servomechanisms Laboratory.
Read first: David A. Mindell, Between Human and Machine: Feedback, Control, and Computing before Cybernetics (2002).
Written in: differential equations.
Milieu intérieur to homeostasis
A living body keeps its insides steady while the world outside changes.
Bernard's fullest statement is from 1878. Cannon first printed "homeostasis" in 1926. Rosenblueth worked with Cannon at Harvard, which is how this line reaches Wiener; Ashby later named his machine the homeostat.
Read first: Walter B. Cannon, The Wisdom of the Body (1932).
Ampère's cybernétique
A name in a classification of all the sciences for the art of governing a state.
A namesake, not a parent: Wiener coined the word on his own in 1947 from the Greek for steersman. No line is drawn from it.
Read first: André-Marie Ampère, Essai sur la philosophie des sciences, the second part, where cybernétique is named (1843).
The roots 1943 to 1956
Behavior, Purpose and Teleology
Purpose described in machine terms: behavior steered by feedback about how far it still is from its goal.
Grew from Milieu intérieur to homeostasis: Rosenblueth came from Cannon's lab; Servomechanisms at MIT: Wiener and Bigelow's wartime anti-aircraft predictor.
Read first: Rosenblueth, Wiener and Bigelow, Behavior, Purpose and Teleology (1943). Here: the primer for Rosenblueth, Wiener and Bigelow, 1943.
Written in: words and diagrams.
A Logical Calculus of neurons
Treat a neuron as an on-off switch, and a network of them can compute logic.
Read first: McCulloch and Pitts, A Logical Calculus of the Ideas Immanent in Nervous Activity (1943).
Written in: sets, logic and machines as tables, neural networks.
Craik, The Nature of Explanation
A mind carries small working models of the world and runs them to see what will happen.
The British root: the Ratio Club counted Craik, who died in 1945, as a founding influence.
Read first: Kenneth Craik, The Nature of Explanation (1943).
The Macy conferences
Ten meetings where engineers, physiologists, mathematicians and anthropologists tried to talk about circular causality in one shared language.
Grew from Behavior, Purpose and Teleology; A Logical Calculus of neurons: McCulloch chaired them.
The first was called "Feedback Mechanisms and Circular Causal Systems in Biological and Social Systems". Von Foerster edited the transcripts of the last five. Shannon and Licklider came as guests; Lewin only to the earliest, before his death in 1947.
Read first: Steve J. Heims, The Cybernetics Group (1991).
Written in: words and diagrams.
Wiener, Cybernetics
One mathematics of feedback and messages for machines, animals and societies, and the name for it.
Grew from Behavior, Purpose and Teleology; The Macy conferences; Maxwell, On Governors: cited by Wiener as the first paper on feedback.
Read first: Norbert Wiener, Cybernetics: Or Control and Communication in the Animal and the Machine, the introduction (1948). Here: the primer for Norbert Wiener, 1948.
Written in: calculus, probability, information and signals.
I read this one as a teenager.
A Mathematical Theory of Communication
Information can be measured in bits, and every channel has a top speed for sending it reliably.
Grew from Feedback amplifiers (partial): the same Bell Labs; Shannon cites Nyquist's earlier telegraph work.
A sibling of Cybernetics more than a child: Shannon came to Macy as a guest, and in 1956 he warned against the cybernetics "bandwagon". The 1949 book with Weaver is "The" Mathematical Theory, the paper is "A".
Read first: Claude Shannon, A Mathematical Theory of Communication (1948). Here: my review of Gleick's The Information.
Written in: logarithms, polynomials and complex numbers, probability, information and signals.
Von Neumann's automata
A logic of machines that compute, copy themselves and stay reliable though built from unreliable parts.
Grew from A Logical Calculus of neurons.
Given at the Hixon Symposium in 1948. His game theory with Morgenstern (1944) is a neighbour that grew up at RAND, not a cybernetic child.
Read first: John von Neumann, The General and Logical Theory of Automata (1951).
Written in: sets, logic and machines as tables.
Ratio Club, Grey Walter's tortoises
A London dining club for the British cyberneticians; Walter's small robot tortoises wander, seek light and find their way home to recharge.
Grew from Craik, The Nature of Explanation; Wiener, Cybernetics (partial): the club formed the year after Wiener's book, around the same questions.
Read first: W. Grey Walter, An Imitation of Life, Scientific American (1950).
Written in: programming.
Ashby, requisite variety
Only variety can absorb variety: a regulator needs at least as many ways to respond as the world has ways to disturb it.
Grew from Milieu intérieur to homeostasis: his homeostat, built in 1948; Ratio Club, Grey Walter's tortoises; The Macy conferences (partial): a guest in 1952.
Design for a Brain (1952) has the homeostat; An Introduction to Cybernetics (1956) has the law.
Read first: W. Ross Ashby, An Introduction to Cybernetics, chapter 11 (1956). Here: the primer for W. Ross Ashby, 1956.
Written in: sets, logic and machines as tables, logarithms, polynomials and complex numbers.
First branches 1950s to 1970s
Modern control theory
Steering written as state, estimate and optimal action: filters that guess what you cannot see, and plans that minimize cost over time.
Grew from Servomechanisms at MIT; Wiener, Cybernetics (partial): the Kalman filter builds on Wiener's filtering mathematics, but Kalman worked outside the movement.
Bellman's dynamic programming (1957) came out of RAND and operations research: a neighbour that joins this branch, not a cybernetic child. In the Soviet Union, Pontryagin's optimal control was filed under technical cybernetics.
Read first: Karl J. Åström and Richard M. Murray, Feedback Systems: An Introduction for Scientists and Engineers (2008).
Written in: calculus, differential equations, linear algebra.
Every Good Regulator
Whatever regulates a system best has to carry a model of that system.
Grew from Ashby, requisite variety.
Read first: Conant and Ashby, Every Good Regulator of a System Must Be a Model of That System (1970). Here: the primer for Conant and Ashby, 1970.
Written in: sets, logic and machines as tables, probability, information and signals.
The perceptron
A net of artificial neurons that learns its own weights from examples instead of being wired by hand.
Grew from A Logical Calculus of neurons.
Hebb's learning rule (1949) fed in from neuropsychology, a neighbour rather than a cybernetic child. Minsky and Papert's Perceptrons (1969) is often blamed for the long neural-net winter after; how much it caused is argued.
Read first: Frank Rosenblatt, The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain (1958).
Written in: linear algebra, neural networks.
The name "artificial intelligence"
The Dartmouth summer project gave thinking machines a new name, and with it a new field built on symbols rather than feedback.
Grew from Von Neumann's automata (partial); A Logical Calculus of neurons (partial).
McCarthy, in his own words: one of the reasons for inventing the term was "to escape association with 'cybernetics'", and to avoid having either to accept Wiener as a guru or to argue with him. This is where the parent's name starts to drop out of the catalog.
Read first: John McCarthy, review of The Question of Artificial Intelligence (1988).
Written in: sets, logic and machines as tables.
Information theory
Coding, compression and channel capacity grew into their own field, and kept their distance from cybernetics on purpose.
Grew from A Mathematical Theory of Communication.
Read first: Claude Shannon, The Bandwagon (1956).
Written in: logarithms, polynomials and complex numbers, probability, information and signals.
Man-Computer Symbiosis
A person and a computer coupled in a tight loop, each doing what it is good at; then the research money that built interactive computing.
Grew from The Macy conferences: a guest in 1950; Wiener, Cybernetics: a member of Wiener's weekly circle at MIT.
He became the first director of ARPA's computing office in 1962. Engelbart's augmentation work drew money from there, but his own stated roots are Vannevar Bush and Korzybski, so his line to cybernetics is partial at most.
Read first: J. C. R. Licklider, Man-Computer Symbiosis (1960). Here: The Hacker Ethic Meets Cybernetics.
Written in: words and diagrams.
Management cybernetics, the VSM
Every viable organization does five jobs, and each of its parts is itself a viable system doing the same five.
Grew from Ashby, requisite variety: Beer was Ashby's friend and built the VSM on requisite variety.
Beer came out of operations research, a wartime discipline that grew up alongside cybernetics rather than from it.
Read first: Stafford Beer, The Viable System Model: Its Provenance, Development, Methodology and Pathology (1984). Here: the primer for Stafford Beer, 1984.
Written in: sets, logic and machines as tables.
Project Cybersyn
A telex network, software and an operations room to steer Chile's nationalized industries in near real time, ended by the coup.
Grew from Management cybernetics, the VSM.
Read first: Eden Medina, Cybernetic Revolutionaries: Technology and Politics in Allende's Chile (2011). Here: the podcasts on the gentler shelf.
Soviet cybernetics
Condemned as reactionary pseudoscience, rehabilitated a year later, then the plan for a national computer network to run the economy, which was never built.
Grew from Wiener, Cybernetics: Kitov and Lyapunov read Wiener.
The 1954 Short Philosophical Dictionary called it "reactionary pseudoscience"; the 1955 article by Sobolev, Kitov and Lyapunov brought it back. Gerovitch, From Newspeak to Cyberspeak (2002), is the fuller history.
Read first: Benjamin Peters, How Not to Network a Nation: The Uneasy History of the Soviet Internet (2016).
General systems theory
The same patterns of organization show up in cells, societies and machines, so study the patterns.
Grew from The Macy conferences (partial): sibling movements: Gerard sat at Macy, but Bertalanffy treated cybernetics as one special case of his theory and kept them apart.
The society was conceived in 1954 and founded in 1956; it is now the ISSS.
Read first: Ludwig von Bertalanffy, General System Theory (1968).
Written in: words and diagrams.
System dynamics
Stocks, flows and delays simulated on a computer, to see why firms and cities keep producing behavior nobody chose.
Grew from Servomechanisms at MIT: Forrester trained in the Servomechanisms Laboratory; Wiener, Cybernetics (contested): George Richardson argues system dynamics grew from a separate servomechanisms thread of feedback thought, distinct from cybernetics; Forrester never attended Macy.
Read first: Jay W. Forrester, Industrial Dynamics (1961).
Written in: programming, differential equations.
The Limits to Growth
A system dynamics model of the whole world economy, run forward to see where growth hits its limits.
Grew from System dynamics.
Read first: Meadows, Meadows, Randers and Behrens, The Limits to Growth (1972).
Written in: programming, differential equations.
Ecosystems as loops
Energy and nutrients move through an ecosystem in circuits, and the circuits hold it steady.
Grew from The Macy conferences: Hutchinson was a core member; Howard Odum was his student.
Contested: whether an ecosystem is cybernetic at all. Engelberg and Boyarsky said no in 1979, since nothing in it holds a set point; Patten and Eugene Odum answered yes in 1981.
Read first: G. Evelyn Hutchinson, Circular Causal Systems in Ecology (1948).
Bateson: the double bind, an ecology of mind
Families and cultures run on loops of communication, and information is a difference that makes a difference.
Grew from The Macy conferences: Bateson credited the conferences for his turn to systems.
Read first: Gregory Bateson, Steps to an Ecology of Mind (1972). Here: the primer for Gregory Bateson, 1970, Culture Is the Oldest Information Technology.
Written in: words and diagrams.
Second order 1967 to 1990s
Von Foerster, the observer inside
There is no description without someone describing: second-order cybernetics is the cybernetics of observing systems.
Grew from The Macy conferences: he edited the transcripts; Ashby, requisite variety: Ashby worked at his Biological Computer Laboratory in Illinois.
Read first: Heinz von Foerster, Cybernetics of Cybernetics (1979). Here: the primer for Heinz von Foerster, 1979.
Written in: words and diagrams.
Mead, Cybernetics of Cybernetics
A society that studies self-correcting systems should turn the same tools on itself.
Grew from Bateson: the double bind, an ecology of mind; The Macy conferences.
Given in 1967 at the first symposium of the American Society for Cybernetics; von Foerster gave it its title.
Read first: Margaret Mead, Cybernetics of Cybernetics (1968). Here: the primer for Margaret Mead, 1968.
Written in: words and diagrams.
Autopoiesis
A living thing is a network that keeps making the parts that make the network, and draws its own edge while it does.
Grew from Von Foerster, the observer inside: Maturana worked at the Biological Computer Laboratory.
De máquinas y seres vivos is dated 1972 in some sources and 1973 in others.
Read first: Varela, Maturana and Uribe, Autopoiesis: The Organization of Living Systems (1974). Here: the primer for Varela, Maturana and Uribe, 1974.
Written in: words and diagrams.
Radical constructivism
Knowledge is built to fit experience, not copied from a world outside.
Grew from Von Foerster, the observer inside.
Read first: Heinz von Foerster, On Constructing a Reality (1973). Here: the primer for Heinz von Foerster, 1973.
Written in: words and diagrams.
The Embodied Mind
Cognition is enacted by a body acting in a world, not computed on a picture of it.
Grew from Autopoiesis.
Read first: Varela, Thompson and Rosch, The Embodied Mind (1991).
Written in: words and diagrams.
Pask's conversation theory
Knowledge is made in conversation: two participants learning each other until they agree, and machines built to take part.
Grew from Von Foerster, the observer inside: Pask worked at the Biological Computer Laboratory; Management cybernetics, the VSM: Beer was his friend and collaborator.
Musicolour, his light show that answered a musician, dates from 1953; the conversation theory books from 1975 and 1976. He advised Cedric Price on the Fun Palace, which was never built. Glanville was his student.
Read first: Gordon Pask, Developments in Conversation Theory, Part I (1980). Here: the primer for Gordon Pask, 1980, Arguing and Agreeing, a freewriting page.
Written in: words and diagrams.
Whole Earth, counterculture
Cybernetic ideas carried into the counterculture, and from there into the personal computer.
Grew from Bateson: the double bind, an ecology of mind: Brand interviewed Bateson and Mead in 1976; Bateson was a Whole Earth hero.
Read first: Fred Turner, From Counterculture to Cyberculture (2006).
Written in: words and diagrams.
Luhmann's social systems
Society as communication producing more communication, a system that draws its own boundary.
Grew from Autopoiesis (contested): Luhmann borrowed autopoiesis for society; Maturana himself rejected applying it to social systems.
Read first: Niklas Luhmann, Social Systems (1984).
Written in: words and diagrams.
Morin's complex thought
Think the parts and the whole together, and let order and disorder, observer and observed, stay in the same picture.
Grew from Von Foerster, the observer inside: Morin draws on von Foerster's order from noise; Information theory (partial): he calls his method a reorganization of systems theory, cybernetics and information theory.
Read first: Edgar Morin, Restricted Complexity, General Complexity (2006). Here: the primer for Edgar Morin, 2006, Finding Edgar Morin.
Written in: words and diagrams.
On the shelf today now
Robotics and control
Simple loops between sensing and acting, layered up, give a body sensible behavior in a messy world.
Grew from Ratio Club, Grey Walter's tortoises: Brooks names Grey Walter's robots as built on cybernetic principles; Modern control theory.
Braitenberg directed a Max Planck Institute for Biological Cybernetics; his Vehicles (1984) is on the gentler shelf.
Read first: Rodney Brooks, Intelligence Without Reason (1991). Here: David Silver's robot bug.
Written in: programming.
Control loops in software
A controller watches the actual state of a cluster, compares it with the desired state, and acts on the difference, forever.
Grew from Modern control theory: the Kubernetes docs explain controllers as control loops, with a thermostat; Every Good Regulator (partial): readers draw this line, this site included; Kubernetes descends from control engineering and Google's Borg, and shares only the root word, helmsman.
Philipp Janert's Feedback Control for Computer Systems (2013) is the workbench book for this shelf.
Read first: The Kubernetes project, Controllers (current docs). Here: The Regulator Was Fed Intents, A Loop Is a Path That Closes.
Written in: programming.
Neural nets, deep learning
Many layers of simple units, trained by sending the error backward, learn what nobody programmed into them.
Grew from The perceptron.
Rumelhart, Hinton and Williams made backpropagation famous; Linnainmaa (1970) and Werbos (1974) had it first.
Read first: Rumelhart, Hinton and Williams, Learning representations by back-propagating errors (1986).
Written in: calculus, linear algebra, neural networks.
Reinforcement learning
An agent learns to act by trial and error, from reward, with nobody showing it the right answer.
Grew from Modern control theory: Bellman's optimal control, one of the three threads Sutton and Barto trace; Ratio Club, Grey Walter's tortoises (partial): Grey Walter's learning tortoise and Shannon's maze mouse appear as early precursors.
Sutton and Barto: "we had a sense that reinforcement learning had been thoroughly explored in the early days of cybernetics and artificial intelligence. On closer inspection, though, we found that it had been explored only slightly."
Read first: Sutton and Barto, Reinforcement Learning: An Introduction, chapter 1 (2018).
Written in: probability, information and signals, neural networks.
Entropy, coding, compression
The measure of information under every network, every compressed file and the loss function most models train on.
Grew from Information theory.
Data science's main ancestor is statistics, not cybernetics. What it inherited from this family is Shannon's measure: entropy, cross-entropy, bits.
Read first: James Gleick, The Information (2011). Here: my review of The Information.
Written in: logarithms, polynomials and complex numbers, probability, information and signals.
I've read Gleick's The Information.
Interaction design, HCI
Design the loop between a person and a system, so the system can take part in a conversation instead of only taking orders.
Grew from Man-Computer Symbiosis; Pask's conversation theory; Whole Earth, counterculture (partial): Turner traces the counterculture into personal computing; the line into design practice is looser.
Read first: Dubberly and Pangaro, How cybernetics connects computing, counterculture, and design (2015). Here: the primer for Dubberly and Pangaro, 2015.
I've watched some of Paul Pangaro's videos.
Organizational design
Design an organization so its parts can act on their own and the whole still holds together.
Grew from Management cybernetics, the VSM; Project Cybersyn (partial): the cautionary and the inspiring case at once.
Conway's law (1968), that systems copy the communication structure of the teams that build them, sits on this shelf but arrived on its own; no cybernetic line is documented.
Read first: Dan Davies, The Unaccountability Machine (2024). Here: Gentler On-Ramps to Cybernetics.
Systems thinking
See the stocks, flows, delays and loops behind a problem before trying to push on it.
Grew from The Limits to Growth; General systems theory (partial).
Senge's The Fifth Discipline (1990) came out of Forrester's group at MIT, so its link to cybernetics is as contested as system dynamics' own.
Read first: Donella Meadows, Thinking in Systems (2008). Here: first on the gentler shelf.
I've read some of Donella Meadows.
Family and systemic therapy
Treat the family's pattern of communication, not one person in it, as what keeps a problem going.
Grew from Bateson: the double bind, an ecology of mind: the Mental Research Institute in Palo Alto grew beside Bateson's project.
Read first: Bateson, Jackson, Haley and Weakland, Toward a Theory of Schizophrenia (1956).
Written in: words and diagrams.
Cognitive science, extended mind
A mind can reach past the skull into tools, notebooks and the people around it.
Grew from The Embodied Mind; Craik, The Nature of Explanation (partial): cognitive science's own histories count the 1940s work among its precursors; Bateson: the double bind, an ecology of mind (partial): Bateson's blind man and his stick is the precursor often noted; whether Clark and Chalmers cite it is unchecked.
Read first: Andy Clark and David Chalmers, The Extended Mind (1998).
Written in: words and diagrams.
The free energy principle
To stay alive a system must carry a model of its world and act to keep its predictions coming true.
Grew from Every Good Regulator: Friston cites Conant and Ashby and calls his result consistent with the good regulator theorem; Craik, The Nature of Explanation (partial).
Read first: Anil Seth, The Cybernetic Bayesian Brain (2015).
Written in: probability, information and signals.
Complexity science
Many simple agents interacting give rise to patterns no single agent holds: markets, ecologies, cities.
Grew from Wiener, Cybernetics (contested): the Santa Fe Institute's founders came from physics and Los Alamos and do not claim cybernetics; Pickering and others read complexity science as its continuation; Morin's complex thought (partial): Morin's general complexity is the philosophical line; he credits the Santa Fe kind and argues it needs his.
Read first: Melanie Mitchell, Complexity: A Guided Tour (2009). Here: Making Complexity Visible, the cybernetics reading path.
Written in: programming, logarithms, polynomials and complex numbers.
Sources
An agent checked the links against these in October; the papers themselves are cited in the entries above.
| Source | What it supports |
|---|---|
| Steve J. Heims, The Cybernetics Group (MIT Press, 1991) | the Macy conferences and who sat at them |
| Ronald R. Kline, The Cybernetics Moment (Johns Hopkins University Press, 2015) | how the field scattered into information theory, AI and systems science |
| Andrew Pickering, The Cybernetic Brain (University of Chicago Press, 2010) | the British line: Grey Walter, Ashby, Beer, Pask; complexity as continuation |
| Eden Medina, Cybernetic Revolutionaries (MIT Press, 2011) | Project Cybersyn |
| Slava Gerovitch, From Newspeak to Cyberspeak (MIT Press, 2002) | Soviet cybernetics |
| Benjamin Peters, How Not to Network a Nation (MIT Press, 2016) | the Soviet national network plans |
| David A. Mindell, Between Human and Machine (Johns Hopkins University Press, 2002) | feedback and control before cybernetics |
| Stuart Bennett, A History of Control Engineering (IEE, 1979 and 1993) | the governors and the control line |
| Otto Mayr, The Origins of Feedback Control (MIT Press, 1970) | the oldest regulators |
| George P. Richardson, Feedback Thought in Social Science and Systems Theory (University of Pennsylvania Press, 1991) | system dynamics as a separate thread |
| Debora Hammond, The Science of Synthesis (University Press of Colorado, 2003) | general systems theory as a sibling movement |
| Fred Turner, From Counterculture to Cyberculture (University of Chicago Press, 2006) | Whole Earth and the road to personal computing |
| Husbands and Holland, "The Ratio Club," in The Mechanical Mind in History (MIT Press, 2008) | the British dining club |
| John McCarthy, review of The Question of Artificial Intelligence | why the new field took a new name |
| Kubernetes documentation, Controllers and Overview | control loops and the helmsman |
| Sutton and Barto, Reinforcement Learning, section 1.7 | the threads reinforcement learning grew from |
| Friston, "Life as we know it" (2013) | the free energy principle citing the good regulator |
| Brooks, "Intelligence Without Reason" (1991) | Grey Walter in the robotics line |
| American Society for Cybernetics, Cybernetics of Cybernetics | Mead's address |
| ISSS, history | the founding of the general systems society |