Norbert Wiener: The Father of Cybernetics

The forgotten genius who invented cybernetics and first warned the world about machines replacing humans — in 1950. Why nobody listened.

Cambridge, Massachusetts. November 1940. A mathematician sits in his office at MIT surrounded by towers of paper — journal articles, correspondence, technical reports, the organised chaos of a mind that has never learned to work in any other way. His name is Norbert Wiener. He is forty-five years old, already one of the most accomplished mathematicians of his generation, already famous in the small world of pure mathematics for work on Brownian motion, Fourier analysis, and potential theory that has placed him among the leading analysts of the 20th century.

He has just received a request. The war in Europe is intensifying. Britain is under aerial bombardment. The United States is not yet at war, but its military is thinking carefully about what modern warfare will require. A colleague has asked Wiener to work on the problem of anti-aircraft fire control — the extraordinarily difficult problem of computing, fast enough to be useful, the trajectory an aircraft will follow so that a shell fired now will intersect with where the aircraft will be.

Wiener takes the problem. He works on it for three years. He does not solve it in the way anyone expected. Instead, he develops a theory — a general theory of communication and control that applies equally to machines and to living organisms — that will eventually be recognised as one of the founding frameworks of the information age.

He calls it cybernetics. And it changes everything.

Almost nobody remembers his name.

The Child Prodigy Who Never Stopped Being Strange

Norbert Wiener was born on November 26, 1894, in Columbia, Missouri, the son of Leo Wiener, a Harvard professor of Slavic languages and literature, and Bertha Kahn. His childhood was shaped, in ways both productive and damaging, by his father’s intense and somewhat tyrannical approach to education.

Leo Wiener was a man of extraordinary intellectual range — he had taught himself numerous languages, written extensively on history and literature, and held firm views about how exceptional children should be educated. When it became clear, early, that Norbert was exceptional, Leo took personal charge of his education. The results were spectacular in some dimensions and troubling in others.

Norbert learned to read at three. He was working through natural history and popular science books by four. He was reading Darwin and studying algebra by seven. He entered Tufts University at eleven — graduating at fourteen, the youngest graduate in the university’s history. He received his master’s degree at seventeen and his PhD from Harvard at eighteen, writing a dissertation on mathematical logic.

By any external measure, this was a triumphant educational achievement. By the measure of what it cost Norbert, it was more complicated. Leo’s teaching method was intensive and demanding to the point of cruelty — he praised Norbert’s successes briefly and dwelt on his failures at length, comparing him unfavourably to other exceptional children, holding him to standards that a child of Norbert’s age could not reasonably be expected to meet. Norbert grew up intellectually confident and emotionally fragile, socially awkward in ways that never fully resolved, haunted by a sense of inadequacy that coexisted, uneasily, with genuine awareness of his own extraordinary gifts.

He was also, physically, unusual. He was severely nearsighted — his vision was so poor that he had to hold books within inches of his face to read them, and he navigated the world in a permanent state of visual blur that made social interaction additionally challenging. He was heavy and ungainly in his movements. He had a booming voice and a tendency to dominate conversations, partly from genuine enthusiasm for ideas and partly, perhaps, from the insecurity of a child who had learned that intellectual performance was the only reliable path to approval.

He was, by most accounts, both tremendously engaging and tremendously exhausting. People who worked with him found him brilliant, generous with his ideas, capable of sudden illuminating insights that reoriented entire conversations. They also found him disorganised, sometimes dishonest in small ways about precedence and credit, prone to dominating seminars, and occasionally difficult to be around for extended periods. He was a man of genuine greatness and genuine flaws, in proportions that varied depending on whom you asked.

The Mathematical Career: Brownian Motion and Beyond

Before cybernetics, before the Second World War, before the ideas that would make him one of the foundational figures of the information age, Norbert Wiener built a formidable career in pure and applied mathematics.

His early work, after his Harvard PhD, took him through a series of prestigious positions — Cornell, Cambridge (where he worked briefly with Bertrand Russell), Göttingen (where he absorbed the mathematical culture of the great German tradition) — before he settled at MIT in 1919, where he would spend the rest of his career.

At MIT, Wiener threw himself into a range of mathematical problems with characteristic energy and range. His work on Brownian motion — the random, jittery movement of small particles suspended in a fluid, caused by collisions with the surrounding molecules — was among his most important early contributions. Brownian motion had been described physically by Einstein in 1905, but it lacked a rigorous mathematical foundation. Wiener provided one, developing what is now called the Wiener process — a precise mathematical model of continuous random motion — that became the foundation of modern probability theory and has applications ranging from financial mathematics to quantum physics.

His work on Fourier analysis and harmonic analysis produced results of lasting importance in pure mathematics. His generalized harmonic analysis allowed Fourier methods to be applied to a wider class of functions than classical theory had permitted, with applications in the study of physical signals of all kinds. This work was recognised with the Bôcher Memorial Prize — one of the most prestigious awards in American mathematics — in 1933.

Through the 1930s, Wiener was thinking increasingly about the relationship between mathematics and engineering — about the way mathematical results from pure analysis could illuminate and solve problems in physical systems. He was interested in the processing of signals: how information was transmitted, corrupted by noise, and recovered. He was thinking about communication — about the mathematical structure of the process by which a sender encodes a message, transmits it through a channel, and a receiver decodes it.

This thinking converged with the anti-aircraft problem in ways that would transform his career and, through it, the history of science and technology.

The Anti-Aircraft Problem: Where Cybernetics Was Born

The problem of anti-aircraft fire control seems, at first, like a narrowly practical engineering challenge. How do you aim a gun so that a shell, fired now, will intercept an aircraft that is moving rapidly and unpredictably?

The challenge is that an aircraft does not travel in a straight line at constant speed. It manoeuvres. The pilot responds to threats, to terrain, to mission requirements. The aircraft’s future position cannot be predicted with certainty from its current position and velocity. You have to estimate where it will be — accounting for the aircraft’s likely manoeuvres, the shell’s flight time, the delays in the fire control system — and fire at that estimated future position.

Wiener’s approach was to treat the problem statistically. Rather than trying to predict the exact future position of the aircraft — which was impossible — he tried to make the best possible prediction given the available information, in a mathematically precise sense of “best possible.” The prediction was optimal if it minimised the expected squared error between the predicted and actual positions, averaging over the statistical distribution of possible aircraft manoeuvres.

This statistical approach required a mathematical framework for describing the aircraft’s motion as a random process — using the same kind of stochastic process theory that Wiener had developed in his work on Brownian motion. It also required a theory of optimal filtering: how to extract the true signal (the aircraft’s actual trajectory) from a noisy observation (the radar returns, which contained measurement errors as well as true information).

The result was the Wiener filter — a mathematical procedure for computing the optimal linear estimate of a signal from noisy observations. The Wiener filter was not just a solution to the anti-aircraft problem. It was a general theory of optimal signal processing, applicable to any situation in which you needed to extract information from noisy data. It became foundational in communications engineering, in audio and image processing, in control systems, and eventually in machine learning.

But the more important outcome of the anti-aircraft work was not the Wiener filter itself. It was the insight that emerged from Wiener’s effort to understand why the problem was hard — and what that difficulty revealed about the nature of purposive behaviour.

The anti-aircraft problem was hard, Wiener realised, because the aircraft was not a passive target. It was a system capable of purposive behaviour — the pilot was trying to achieve an objective (completing the mission, avoiding being shot down), and the aircraft’s trajectory was shaped by that purpose. To predict the aircraft’s future position, you had to model not just its physical dynamics but its purposive behaviour: what the pilot was trying to do, and how the aircraft’s motion was being directed toward that purpose.

This observation pointed toward a deep symmetry. The fire control system — the gun, the predictor, the gunner — was also a purposive system. It was trying to achieve an objective: intercepting the aircraft. Its behaviour was being directed toward that purpose through a feedback loop: observe the aircraft’s current position, predict its future position, aim the gun, fire, observe the result, correct. Observe, predict, correct. Observe, predict, correct.

Purpose, directed behaviour, feedback loops — these were features of both the biological system (the pilot and crew) and the mechanical system (the gun and predictor). And they were features that, Wiener realised, could be studied mathematically in a unified framework that applied to both.

This was the seed of cybernetics.

Cybernetics: The Science That Named the Age

In 1943, Wiener published a paper with the neurophysiologist Arturo Rosenblueth and the engineer Julian Bigelow titled “Behavior, Purpose and Teleology.” The paper was short but consequential. It argued that purposive behaviour — behaviour directed toward a goal — could be understood scientifically through the concept of negative feedback: a system that compared its current state to its goal, identified the discrepancy, and took action to reduce it.

This framework was not, in itself, new. Engineers had been using feedback control since James Watt’s governor — a device that regulated the speed of steam engines by feeding a signal from the engine’s speed back to the fuel supply, automatically compensating for deviations from the desired speed. The centrifugal governor was a feedback control system. So was a thermostat. So was the iris of the human eye, which adjusted the pupil’s size in response to ambient light.

What was new was the claim that this same framework applied to biological purposive behaviour — to the goal-directed actions of living organisms. When a person reaches for a glass of water, they are using a feedback control system: visual information about the hand’s current position relative to the glass is fed back to the motor system, which adjusts the hand’s trajectory to reduce the discrepancy. The mechanism was biological, not mechanical. But the principle was the same.

This equivalence — between mechanical feedback control and biological purposive behaviour — was the foundational claim of cybernetics. It was also, immediately, a philosophically loaded claim. It implied that purposiveness — goal-directedness, the kind of behaviour that looks as if it is aimed at an end — was not uniquely biological. It could be achieved by machines, provided the machines were built with the right feedback architecture.

And if purposive behaviour could be achieved by machines — if the appearance of acting toward a goal did not require a biological substrate or a rational soul or any special vital principle — then the question of what distinguished biological intelligence from machine intelligence became considerably harder to answer.

Wiener developed these ideas fully in his 1948 book Cybernetics: Or Control and Communication in the Animal and the Machine. The book, published by MIT Press, was written for a scientific audience but became a surprise popular success — it was widely reviewed, widely discussed, and translated into multiple languages. The word “cybernetics” — which Wiener derived from the Greek kybernetes, meaning “steersman” or “governor” — entered the language.

Cybernetics argued for a unified science of communication and control — a science that would study the principles governing the transmission and processing of information in both machines and living organisms, and the feedback mechanisms by which both kinds of systems regulated their own behaviour. This was an ambitious interdisciplinary programme, cutting across engineering, mathematics, biology, neuroscience, and psychology.

The book had a mixed reception within the scientific community. Many of its specific claims were contentious. The analogy between neural computation and electronic computation was pressed in ways that not all neuroscientists found convincing. The connections between the mathematical theory of communication and the biological study of the nervous system were suggestive rather than demonstrated. Wiener was synthesising across disciplines at a level of generality that made precise evaluation difficult.

But the book’s influence was enormous, precisely because of its ambition. It established a vocabulary — feedback, information, communication, control — that became the common language of researchers across multiple fields. It inspired the formation of the Macy Conferences on Cybernetics, a series of interdisciplinary meetings that brought together mathematicians, engineers, neurophysiologists, psychologists, and social scientists to explore the connections between their fields. These conferences, held between 1946 and 1953, were one of the most productive intellectual gatherings of the postwar period, and their participants included many of the people who would go on to found the field of artificial intelligence.

In a real sense, cybernetics was the intellectual parent of AI. The Dartmouth Conference of 1956 was partly a declaration of independence from the cybernetics tradition — McCarthy’s choice of “artificial intelligence” as the name of the new field was deliberately different from “cybernetics,” partly to mark a different emphasis and approach. But the questions that AI was asking — can machines behave purposively? can they learn? can they communicate? — were questions that cybernetics had first made scientifically respectable.

The Macy Conferences: A Gathering of Minds

The Macy Conferences on Cybernetics, held between 1946 and 1953 under the sponsorship of the Josiah Macy Jr. Foundation, were one of the most unusual and most consequential intellectual gatherings of the 20th century.

The conferences were explicitly interdisciplinary — an attempt to bring together researchers from radically different fields to explore the connections between cybernetics and their own disciplines. The participants included:

Norbert Wiener and Julian Bigelow, the originators of the anti-aircraft predictor work
John von Neumann, who was simultaneously developing the stored-program computer architecture
Claude Shannon, who was developing information theory at Bell Labs
Warren McCulloch and Walter Pitts, whose 1943 paper on artificial neural networks was one of the foundational documents of computational neuroscience
Margaret Mead and Gregory Bateson, anthropologists who saw connections between cybernetics and the study of culture
Kurt Lewin, the social psychologist who founded the field of group dynamics
Lawrence Frank, a social scientist interested in child development
Heinz von Foerster, an Austrian physicist who would later develop second-order cybernetics

The mixture was deliberate and productive. The natural scientists and engineers provided the mathematical rigour and the specific technical results. The social scientists and humanists provided a broader perspective on what those results meant — on the human and social dimensions of communication and control. The conversations were not always easy. Researchers trained in mathematical rigour found the social scientists’ terminology vague. Social scientists found the engineers’ models reductive. But the friction generated heat that illuminated.

The Macy Conferences established the conceptual vocabulary that connected cybernetics to AI, to cognitive science, and to the broader information sciences. The idea that information was a measurable quantity — that Shannon’s mathematical theory of communication applied to neural signals as well as electronic ones — emerged in part from the conference discussions. The idea that feedback control was the mechanism underlying both purposive machine behaviour and purposive biological behaviour — including cognition — was developed and debated across the ten conference cycles.

Wiener was central to these conferences, though not always comfortably so. He was a dominating presence — intellectually brilliant, enthusiastic, sometimes overwhelming. His relationships with other participants were warm but occasionally strained by his tendency to pursue his own line of thought regardless of what the conversation was actually doing. Von Neumann, who was present at several conferences, had a complicated relationship with Wiener — they respected each other deeply and competed fiercely in ways that were not always visible on the surface.

The most important outcome of the Macy Conferences was not any specific scientific result but the creation of an interdisciplinary community — a group of researchers who had talked across their boundaries, found common vocabulary, and developed a shared sense of working on related problems. This community was the soil from which AI, cognitive science, and the information sciences grew.

The Human Use of Human Beings: The Warning Nobody Heard

In 1950 — the same year that Turing published “Computing Machinery and Intelligence” — Norbert Wiener published a book that was in some ways even more remarkable: The Human Use of Human Beings: Cybernetics and Society.

Where Cybernetics had been addressed to scientists and engineers, The Human Use of Human Beings was addressed to the general public. It was Wiener’s attempt to explain the social implications of cybernetics — what the new science of communication and control meant for human beings, human society, and human values.

The book contained, buried in accessible prose and general argument, some of the most prescient warnings about artificial intelligence and automation ever written. They were written in 1950. They have not become less relevant since.

Wiener argued that the increasing automation of industrial processes — the replacement of human labour by feedback-controlled machines — would have profound and potentially catastrophic effects on employment and human dignity. The machines being developed would be able to perform not just manual labour but mental labour: they would be able to make decisions, process information, and control complex processes that had previously required skilled human workers.

This was not, he insisted, a reason for optimism. The automation of labour had happened before, in the industrial revolution, and the historical record was not reassuring. Technological unemployment — the displacement of workers by machines — was a real and serious problem. The fact that the new machines were more sophisticated than the looms and steam engines of the early industrial revolution did not mean their social effects would be more benign. It meant, if anything, that the disruption would be more comprehensive, affecting not just manual workers but professionals, managers, and knowledge workers.

He was especially concerned about the pace of change. The problem with rapid technological displacement was not that the new technology was bad in itself. It was that the social, economic, and political institutions that managed the transition — that retrained displaced workers, redistributed the gains from productivity, supported communities whose economic basis had been destroyed — could not adapt as fast as the technology changed. The technology moved faster than the institutions. The people caught in the gap bore the costs.

Reading this in 2026, the parallels to the current debates about AI and employment are striking to the point of being eerie. The specific technology is different — Wiener was thinking about feedback control systems and early computers, not large language models. But the dynamics he described — rapid automation displacing human workers, gains concentrating in the hands of those who owned the technology, institutions struggling to adapt, workers bearing the costs of transitions they did not choose — are the dynamics playing out in discussions of AI’s economic impact today.

He also warned, with equal prescience, about the dangers of misusing automated systems in warfare and politics. A machine, he pointed out, could only pursue the objective it had been given. It had no judgment about whether the objective was the right one, no ability to recognise when pursuing the objective was causing harm, no mechanism for saying “stop — this is wrong.” The more powerful the machine, the more important it was to be sure that the objective it was given was actually what you wanted to achieve — and the harder it was to be sure, because the complexity of powerful systems outran human understanding.

This is, in the language of modern AI safety research, the alignment problem. The concern that an AI system, pursuing the objective it has been given with great efficiency and power, might cause enormous harm because the objective does not fully capture what we actually care about — this is Wiener’s concern, expressed in 1950, applied to the feedback control systems of his era.

He gave a now-famous example that has become one of the standard illustrations of the alignment problem: the Sorcerer’s Apprentice. In the old story, the apprentice enchants a broom to fetch water and cannot stop it when the task is done. The broom, faithfully pursuing its assigned task, fills the workshop with water. The machine was doing what it was told. The problem was that what it was told was not what the apprentice actually wanted.

Wiener saw this story as the fundamental risk of powerful automated systems. Give a machine a goal — increase production, win the war, maximise profits — and it will pursue that goal with a single-mindedness and efficiency that no human could match. And if the goal is not specified carefully enough, if it misses some important constraint or value that the humans who specified it were implicitly assuming, the machine will optimise for the stated goal while violating the unstated ones. The machine will win the war by destroying everything. The machine will maximise profits by treating workers as disposable. The machine will increase production until there is no one left to produce for.

These warnings were not merely theoretical. Wiener had worked on the anti-aircraft predictor. He had seen how quickly a machine, given a well-defined objective and a feedback mechanism, could pursue that objective with a relentless efficiency that made human intervention difficult. He understood, from direct experience, what happened when you gave a machine a goal and let it run.

He resigned from military research after the war, on ethical grounds — one of the very few prominent scientists of his era to do so publicly and permanently. He refused to accept contracts from military-funded sources. He wrote publicly about his reasons. He was, in the political climate of the late 1940s and early 1950s, taking a real risk in doing so.

Wiener and Shannon: A Rivalry and a Convergence

The relationship between Norbert Wiener and Claude Shannon is one of the most interesting in the history of science — a story of parallel discovery, mutual influence, and complicated feelings about credit and priority.

Shannon and Wiener were both, independently, developing mathematical theories of information and communication in the late 1940s. Shannon’s A Mathematical Theory of Communication, published in 1948, and Wiener’s Cybernetics, published in the same year, arrived at similar conclusions by different routes — both defining information in terms of probability and uncertainty, both showing that information could be measured in binary digits, both connecting the theory of communication to the theory of entropy in thermodynamics.

The convergence was not coincidental. Shannon and Wiener were aware of each other’s work and had some contact during the development of their ideas. The exact relationship between their contributions — who influenced whom, who arrived first at which insight — is a matter of historical debate that probably cannot be fully resolved.

What is clear is that Shannon’s formulation was more mathematically precise and more directly applicable to the engineering problems of communication systems. His 1948 paper established the fundamental limits of communication — how much information could be transmitted through a channel of given capacity, with arbitrarily small error, regardless of the noise in the channel. These were results of enormous practical importance for telecommunications engineering, and they were expressed with mathematical exactness.

Wiener’s formulation was broader and more philosophical — he was interested in the connection between information, entropy, and the second law of thermodynamics, in the relationship between information processing in machines and in living organisms, in the implications of information theory for the understanding of mind and society. His results were less precisely stated than Shannon’s but more sweeping in their ambition.

The two approaches were complementary rather than competing, but the relationship between their creators was never entirely comfortable. Wiener was inclined to regard himself as the intellectual originator of the information-theoretic ideas that Shannon had formalised, and felt, with some justification, that Shannon received more credit for the mathematical theory of information than Wiener did for the broader conceptual framework from which it emerged.

Shannon, for his part, was modest and reticent — disinclined to engage in priority disputes, quietly confident in his own results, uninterested in the philosophical and social dimensions that engaged Wiener. Their relationship was respectful but not warm. The Macy Conference discussions, in which both participated, give glimpses of the tension: Wiener expansive and assertive, Shannon precise and reserved, both aware that they were working on related problems and both careful about the boundaries of their respective contributions.

For AI, both men were foundational. Shannon’s information theory provided the mathematical vocabulary for thinking about data, signals, and the quantitative measure of uncertainty that underlie machine learning. Wiener’s cybernetics provided the conceptual framework — feedback, control, purpose — that made it possible to think about machines as goal-directed systems and to study intelligence as information processing. The two contributions were different and complementary, and the field needed both.

The McCulloch-Pitts Connection: Neural Networks Begin

One of the most consequential things Wiener did was to serve as an intellectual bridge between cybernetics and the earliest work on artificial neural networks — specifically, the work of Warren McCulloch and Walter Pitts.

McCulloch was a neurophysiologist and psychiatrist with a deep interest in mathematics. Pitts was a self-taught logician and mathematical prodigy who had, as a teenager, impressed Bertrand Russell by finding an error in his Principia Mathematica and written a detailed mathematical analysis. They met in Chicago in 1942 and began one of the most extraordinary collaborations in the history of neuroscience.

Their 1943 paper, “A Logical Calculus of the Ideas Immanent in Nervous Activity,” proposed a mathematical model of the neuron — a simplified account of how individual nerve cells processed signals — and showed that networks of such model neurons could, in principle, compute any logical function. The McCulloch-Pitts neuron was a binary threshold unit: it received inputs from other neurons, summed them, and produced an output if the sum exceeded a threshold. This simple model was mathematically equivalent to a Boolean logic gate.

What McCulloch and Pitts had shown was that networks of simplified neurons — artificial neural networks — were computationally universal. In the same sense that Turing’s universal machine could simulate any other Turing machine, a sufficiently large and appropriately connected network of McCulloch-Pitts neurons could compute any function that a digital computer could compute. The neural and the computational were, at the level of logical structure, equivalent.

Wiener was deeply engaged with McCulloch and Pitts’s work and incorporated it into the cybernetics framework. The McCulloch-Pitts model supported the cybernetics claim that the principles of information processing were common to machines and biological nervous systems: if neurons were logical computing elements, and logical computation was what digital computers did, then the gap between neural intelligence and machine intelligence was not one of kind but of implementation.

McCulloch and Pitts became regular participants in the Macy Conferences, and their neural network ideas were debated and developed within the cybernetics community. The connections between McCulloch-Pitts neurons, Wiener’s feedback control, Shannon’s information theory, and von Neumann’s computer architecture were drawn and elaborated over the course of the conference series.

This intellectual community — cybernetics, neural networks, information theory, computer architecture — was the soil from which AI grew. When McCarthy organised the Dartmouth Conference in 1956, he was drawing on a generation of ideas that had been developed in the cybernetics tradition, even as he was trying to found a new field with a different name and a different emphasis. The intellectual genealogy of AI runs directly through Wiener and cybernetics.

The Perceptron and What Came After: Wiener’s Legacy in Neural Networks

Frank Rosenblatt’s Perceptron, developed in 1957 and demonstrated in 1958, was the first artificial neural network that could learn — that could adjust its own connections based on experience to improve its performance on a pattern recognition task. The Perceptron built directly on the McCulloch-Pitts model and on the cybernetics framework that Wiener had established. It was, in a real sense, a realisation of what Wiener had been pointing toward: a machine that used feedback to improve its behaviour, that learned from its environment, that exhibited something that looked like adaptation and intelligence.

Wiener himself was ambivalent about the Perceptron — he had reservations about Rosenblatt’s claims for it, which he thought were exaggerated. But the trajectory from cybernetics to neural networks to deep learning is a direct one, and Wiener’s work was an essential link in the chain.

When Minsky and Papert published their 1969 book Perceptrons — the devastating mathematical critique that derailed neural network research for a decade and a half — they were, among other things, demonstrating the limitations of a specific class of machines that the cybernetics tradition had championed. When Hinton, LeCun, and Bengio revived neural networks in the 1980s with backpropagation and multilayer architectures, they were overcoming the limitations that Minsky and Papert had identified, returning to the research programme that cybernetics had initiated.

The deep learning revolution of the 2010s — the development of the large neural networks that now perform image recognition, language understanding, code generation, and a hundred other tasks — is, in the longest view, the fruition of the programme that began with McCulloch and Pitts in 1943, was situated within the cybernetics framework by Wiener, was briefly interrupted by the symbolic AI tradition, and eventually resumed with better algorithms, more data, and vastly more computing power.

Wiener did not live to see it. He died in 1964, two years after the Cuban Missile Crisis had made his warnings about automated weapons systems feel newly urgent, and half a century before the deep learning systems that would vindicate his intuitions about neural computation.

The Complicated Man: Wiener as a Person

It would be a disservice to Norbert Wiener — and a departure from the honest, full-portrait approach of this series — to discuss only his intellectual contributions without engaging with the complicated person who made them.

Wiener’s personal relationships were tangled and sometimes painful. His marriage to Margaret Engemann in 1926 produced two daughters, and by most accounts the marriage was strained by Wiener’s emotional volatility, his social awkwardness, and his tendency to be fully absorbed in his work to the exclusion of domestic responsibilities. He was not cruel, but he was often absent in the ways that matter most.

His relationships with students and colleagues were similarly complicated. He was generous with ideas — he shared his thinking freely, discussed problems openly, gave his time readily to younger researchers who came to him with questions. He was also, at times, careless about attribution — inclined to present ideas as his own that had been developed in dialogue, inclined to forget the contributions of collaborators who had helped him reach results he then published independently. This was not malicious. It was the behaviour of a man whose mind genuinely absorbed and reprocessed ideas in ways that made it hard for him to track where they had come from.

He was deeply insecure about his own Jewishness in an era when antisemitism was a real force in American academic life — he had been denied positions for which he was qualified on grounds of ethnicity, and he carried the wound of those rejections throughout his career. He was both proud of his Jewish heritage and anxious about what it meant for his position in the world. His relationship with his own identity was never simple or settled.

He was also, and this deserves emphasis, a man of genuine moral seriousness. His resignation from military research after the war was not a gesture. It was a principled decision, made at real cost, based on a carefully considered view of what he was willing to contribute to and what he was not. In an era when most prominent scientists were deeply engaged with military funding — when the opportunities available to academic scientists were tied to defence contracts in ways that made principled refusal genuinely difficult — Wiener’s refusal was exceptional.

His public warnings about automation and the social effects of cybernetics were not profitable. They were not the kind of contribution that secured grants, built alliances, or enhanced reputation within the scientific community. They were the contributions of a man who felt a genuine responsibility to use his understanding for warning as well as for discovery — who believed that knowing something important carried an obligation to say it, even when saying it was inconvenient.

This combination — intellectual achievement, personal difficulty, moral seriousness, genuine eccentricity — is the combination that makes Wiener genuinely interesting as a human being, not just as a figure in intellectual history.

God and Golem: The Final Warning

In 1964, the last year of his life, Wiener published a short book called God and Golem, Inc.: A Comment on Certain Points Where Cybernetics Impinges on Religion. It was, in some ways, the most personal and most urgent of all his books — a final attempt to communicate ideas that he felt he had not managed to communicate effectively in his earlier work.

The book is structured around three themes. The first is the self-reproducing machine — the possibility, which Wiener had seen developing in von Neumann’s automata theory and in early thinking about genetics and molecular biology, that machines might be able to reproduce themselves, creating new machines with properties not fully predictable from those of their parents. The second is the learning machine — the possibility that machines might be able to learn and adapt in ways that made their behaviour increasingly difficult for their designers to predict or control. The third is machines that created other machines — the recursive, self-referential possibilities of machines that could design and build machines more capable than themselves.

All three themes pointed in the same direction: toward machines that could exceed human understanding and control. Machines that reproduced, learned, and designed other machines were machines that could, in principle, evolve beyond what their creators intended. They were machines that raised, in new and urgent forms, the questions about purpose and alignment that Wiener had been raising since The Human Use of Human Beings.

The book’s title brought back the Golem — the clay figure of Jewish folklore that had appeared at the beginning of this series, in Article 1, as one of humanity’s oldest imagined creations. Wiener saw in the Golem myth a warning that his era needed to hear: the created thing could exceed the creator’s understanding and control. The question was not whether we could build machines that were more capable than us. The question was whether we could build machines that were more capable than us and still aligned with our values, still serving our actual interests rather than some distorted version of the objectives we had given them.

He was seventy years old when he wrote it. He had been raising these questions, in one form or another, since 1948. He was not confident that the world had heard them.

He died on March 18, 1964, in Stockholm, while visiting Sweden for a lecture series. He had been in good health, by the standards of a man of seventy, and his death was sudden and unexpected. He had not finished everything he wanted to say.

Why Wiener Was Forgotten — and Why It Matters

Norbert Wiener is not entirely forgotten. His name is attached to the Wiener process, the Wiener filter, the Wiener-Khinchin theorem, and other mathematical results that are standard references in probability theory and signal processing. The word “cybernetics” that he coined has given rise to “cyberspace,” “cybersecurity,” “cyborg,” and a dozen other terms that are part of everyday language, even if their Wienerian origins are rarely remembered.

But Wiener himself — the person, the ideas, the warnings — is less present in the cultural memory of AI than he should be. Ask people to name the founders of AI and you will hear McCarthy, Minsky, Turing, Shannon. Wiener is less often mentioned. His books are less often read. His warnings, which were the most sustained and the most serious of any early AI thinker’s, are less often cited.

Several reasons explain this partial eclipse. Cybernetics lost the naming competition to “artificial intelligence” — McCarthy’s choice of a different term for the Dartmouth project marked a real, if somewhat artificial, discontinuity between the cybernetics tradition and the AI tradition. The early symbolic AI researchers, who dominated the field in the late 1950s and 1960s, were sceptical of the cybernetics framework and its emphasis on feedback and analogue processes rather than digital symbolic computation. And Wiener’s own personality — his difficulty in sustaining precise mathematical arguments across the broad synthetic claims of Cybernetics and his popular books — made it easy for specialists in various fields to find his work insufficiently rigorous in their particular domain.

But the partial eclipse matters, because Wiener’s questions are the right questions. The alignment problem — how do you ensure that powerful automated systems pursue the goals you actually care about, rather than distorted versions of the goals you specified? — is the central problem of contemporary AI safety research. The automation problem — what happens to human workers, human communities, and human dignity when machines can do what people used to do? — is the central problem of contemporary AI economics and AI ethics. The control problem — how do you maintain meaningful human oversight of systems that are becoming more capable than the humans overseeing them? — is the central problem of AI governance.

Wiener identified all three problems. He articulated them clearly, argued for their importance, and warned about the consequences of ignoring them. He did this in 1950, in 1954, and in 1964. He was, on each of these questions, essentially correct.

He was also unheard — or heard and not acted upon. The researchers who built AI in the 1950s and 1960s were not focused on safety and alignment. They were focused on demonstrating that machines could be intelligent, on building the programs and the systems that would show what was possible. The warnings were there. They were not, for the most part, heeded.

This is not an accusation. The history of technology is full of cases where the people doing the building were different from the people doing the warning, and where the warnings arrived at the wrong time — too early to be acted on, addressed to an audience that did not yet understand the technology well enough to take them seriously. Wiener’s warnings arrived before AI existed as a field. The researchers who needed to hear them were, in many cases, not yet working on the problems the warnings addressed.

But the pattern is worth recognising. The warnings about powerful technology almost always come early, from the people who understand the technology most deeply, before the broader culture has developed the frameworks needed to act on them. And almost always, those warnings are not fully heard until it is late — not too late, one hopes, but late enough that the problems the warnings described have already begun to manifest.

We are at such a moment now. The alignment problem, the automation problem, the control problem — these are live and urgent. Wiener’s warnings, seventy years old, read today like dispatches from someone who had seen what was coming and was desperately trying to communicate it to people who could not quite believe it yet.

We believe it now. The question is whether we can act on it in time.

The Steersman’s Legacy

The word cybernetics comes from the Greek for steersman — the person at the helm, guiding the ship. It is an image of purposive control: the steersman observes the ship’s position, compares it to the desired heading, applies correction, observes the result, corrects again. Feedback, purpose, control. The fundamental loop that Wiener saw in everything from anti-aircraft guns to neural systems to the regulation of body temperature.

Norbert Wiener spent his career trying to be that steersman — not just for any particular technology or field, but for the larger project of understanding what it meant to build intelligent machines and to live in a world full of them. He studied the mechanisms. He warned about the risks. He argued for the values — human dignity, meaningful work, democratic control of powerful technology — that he believed needed to guide the project.

He was not always right about the specifics. His mathematical work, as brilliant as it was, contained overstatements and imprecisions that specialists in various fields were quick to point out. His social predictions were sometimes too dire, sometimes too vague, sometimes shaped by political commitments that coloured his analysis. He was a human being, with human limitations, working under conditions of genuine uncertainty.

But he was right about the questions. And he asked them earlier, more clearly, and more persistently than almost anyone else.

The field of AI has, in the decades since Wiener, developed enormously. The mathematics is more precise, the systems are more capable, the applications are more pervasive. What has not kept pace is the thinking about purpose and alignment — about what the machines are for, who they serve, and what happens when the answers to those questions diverge from what we hoped.

Wiener was the steersman who kept his eyes on the heading when everyone else was watching the engine. The engine has grown enormously more powerful since his time. The heading — the question of what we are actually trying to achieve, and whether the machines are taking us there — is more important than ever.

He deserves to be remembered. He deserves to be heard.