9 November 2026 · History · about 14 minutes

The Steersman and the Noise: Norbert Wiener, Cybernetics, and the Macy Conferences

Norbert Wiener taught machines to steer themselves, gathered the best minds in postwar America to build a science of it, and then spent the rest of his life asking the question the science could not answer: who holds the tiller.

A Yellow Cover

An anti-aircraft gun cannot aim at where a plane is; by the time the shell arrives, the plane has moved. The gun must aim at where the plane will be. And the plane is flown by a pilot who is trying, with every nerve in his body, not to be where the shell will be. So the gunner is not targeting a machine but a mind, and the question of how to hit that mind is a question about the relationship between past behaviour and future intent.¹

In the late winter of 1942, an unusual document arrived at the offices of the National Defense Research Committee in Washington. It was bound in yellow, ran to several hundred pages, and was stamped classified. The mathematics inside was of a kind that only a handful of people in the country could follow, and those who tried and failed gave the report its nickname: the Yellow Peril.²

Its author was Norbert Wiener, a mathematician at MIT, and he is the subject of this essay: the man who worked out how a machine could steer itself, and then spent the rest of his life asking who would steer the machines. Working with the engineer Julian Bigelow, Wiener built a statistical predictor that treated the pilot's flight path as a time series to be analysed and extrapolated. The theory behind it was general enough to describe any process of communication corrupted by noise, any effort to extract signal from interference, and it laid part of the groundwork for what we now call information theory.³ But the army found the results mixed: on two test courses the predictor performed only comparably to simpler systems, and the project ended in January 1943. The Yellow Peril was classified, filed, and not published openly until 1949.⁴

What Wiener and Bigelow had noticed, working on the predictor, went beyond engineering. The gunner tracking a plane and the pilot evading the shell were each using feedback: information about the gap between where they were and where they wanted to be, used to close that gap. This was how the thermostat and the steam governor worked, and the loop by which the body holds its temperature and its balance. Feedback was not a trick of machinery but a principle of purposeful behaviour in general, in animals and machines alike.⁵

Purpose Without a Soul

In January 1943, Arturo Rosenblueth, Wiener, and Bigelow published a short paper called "Behavior, Purpose and Teleology" in the journal Philosophy of Science.⁶ It is one of the stranger and more consequential papers in the history of science.

For centuries, the idea of purpose had been a problem for science. Machines did not have purposes; they had causes. Only living things, and especially minds, seemed to pursue goals. How do you explain behaviour directed toward a future state without invoking something non-mechanical, a soul or a will?

The three authors proposed an answer. Any system whose current state is modified by the gap between that state and its goal is behaving purposefully, in a precise sense, with no metaphysics required. A missile that corrects its course from radar returns is behaving purposefully. So is a cat reaching for a mouse. The difference is of degree, not of kind. The paper argued in effect that teleological behaviour is synonymous with behaviour controlled by negative feedback, and the authors believed they were rescuing the concept of purpose from vitalism on one side and dismissal on the other.⁷ It unified the vocabulary of engineering with the vocabulary of biology, and suggested that the same mathematics could describe both.

That same year, the neurophysiologist Warren McCulloch and the runaway teenage logician Walter Pitts published a paper of matching ambition: "A Logical Calculus of the Ideas Immanent in Nervous Activity." They proposed that neurons could be modelled as logical switches, and that networks of such switches could compute any logical function. They were proposing, in essence, that the brain was a kind of computer.⁸ McCulloch and Pitts were working at the University of Illinois College of Medicine in Chicago and did not know Wiener when the paper was written. In 1943, Jerome Lettvin introduced Pitts to Wiener, and the meeting went so well that Pitts moved to Greater Boston to work at MIT.⁹

So two of the founding ideas of artificial intelligence appeared in print in the same year, feedback-governed purposive behaviour and formal models of neural computation, in overlapping but distinct worlds, without any of the authors quite knowing what the others were doing.

The Word

Wiener had a name in mind for the science he was building. He took it from the Greek kybernetes, the steersman of a ship. James Clerk Maxwell had used the Latin derivative governor in an 1868 paper on feedback in steam engines, a paper Wiener cited as a predecessor. The French physicist André-Marie Ampère had proposed cybernétique as a name for the science of government in the early nineteenth century, but the word had been forgotten. Wiener revived it and gave it a new meaning: the science of control and communication in the animal and the machine.¹⁰

In 1948, he published a book with that subtitle. Cybernetics was first issued in Paris by Hermann et Cie, then in a matching American edition by John Wiley in New York. The French-set first edition was full of typographical errors, which persisted until a corrected second edition appeared in 1961.¹¹ It nevertheless went through at least five American printings in 1948 alone, and Time Magazine reviewed it on December 27 that year under the headline "In Man's Image."¹²

The public fascination was real and a little anxious. If feedback explained purposive behaviour, and if the brain was a logical machine, then the difference between a calculating machine and a human being was perhaps a matter of complexity rather than of essence.

Before the Meetings

The seed had been planted before the book. In May 1942, the Josiah Macy Jr. Foundation sponsored a small conference in New York on hypnotism and the conditioned reflex, known as the Cerebral Inhibition Meeting. Among those present were McCulloch, Rosenblueth, the anthropologists Gregory Bateson and Margaret Mead, Lawrence K. Frank, and the psychoanalyst Lawrence Kubie. Rosenblueth presented the upshot of his conversations with Wiener and Bigelow: a general concept of goal-directed circular-causal processes with negative feedback, applicable to organisms and machines alike.¹³

The effect on the social scientists was immediate. Mead later recalled that she did not notice she had broken a tooth until the conference was over.¹⁴ McCulloch and Bateson began lobbying Frank Fremont-Smith, the Foundation's conference organiser, for a postwar follow-up. Randall Whitaker, writing for the American Society for Cybernetics, has argued that the agenda of the later meetings was already recognisable in 1942, and that the ten conferences that followed added little to those basic outcomes.¹⁵

A Hotel Room in New York

The Macy Foundation had an unusual idea about scientific progress. Its format brought preliminary, unpublished research to an audience drawn from many fields, and discussion was deliberately prolonged and unruly. The aim was not to advance a discipline but to find the vocabulary that might connect several.¹⁶

The first of the conferences met in New York in March 1946, under the title "Feedback Mechanisms and Circular Causal Systems in Biological and Social Systems."¹⁷ Fremont-Smith opened it: "Each group, when it comes together, is an experiment. If it excites you all enough to want to meet again, we will plan for further meetings."¹⁸

They found it interesting enough. The core group included Wiener; McCulloch, who would chair all ten conferences; Bigelow; Rosenblueth; John von Neumann; Bateson and Mead; the neurophysiologist Ralph Gerard; the ecologist G. Evelyn Hutchinson; Kubie; the psychologist Molly Harrower; the sociologist Paul Lazarsfeld; and Pitts.¹⁹ The title changed at nearly every meeting as the participants tried and failed to agree on what they were studying. Not until the sixth conference, in March 1949, did it acquire the name that stuck, when Heinz von Foerster, a young Austrian biophysicist at his first Macy meeting, suggested Wiener's word "cybernetics" on the grounds that it was mercifully shorter. The proposal was approved, and Wiener, by von Foerster's later recollection, was so moved that he left the room to hide his tears.²⁰

The name arrived more than halfway through. Six of the ten meetings had taken place before anyone called them cybernetics conferences.

What They Argued About

The first five conferences were never transcribed, so they are known only through retrospective summaries and participants' notes. The last five were published, edited by von Foerster with Mead and Hans-Lukas Teuber.²¹

The scholar Jean-Pierre Dupuy made a taxonomy of what was discussed across the decade. The largest topic by far was whether a logic-machine model fitted both brain and computer; then human and social communication; then analogies between organisms and machines; then neurosis and the pathology of mental life. Only five discussion units in his count went to what he calls "cybernetics machines", the robots and automata that in the popular imagination defined the whole enterprise.²² The newspapers in 1948 and 1949 were writing about thinking machines; the group itself spent most of its time arguing about neurons, perception and psychoanalysis.

The argument that ran through the whole series was between McCulloch and Pitts, who wanted to model mind as a network of discrete logical switches, and Gerard, who kept pointing out that the brain is a wet, continuous, biochemical organ that does not look much like a logic circuit. At the first conference Gerard said the brain's operations were far more analogical than digital. At the seventh, in March 1950, he was making the same point again, to the visible frustration of the mathematicians. It was never resolved.²³

The psychoanalysts presented a different difficulty. Kubie's presentations on neurosis and compulsion were met with criticism from Pitts, who could find no coherent theory in psychoanalysis, and from Bateson, who could find no objectivity. Kubie kept coming back, and his persistence was itself an argument: the science of mind could not be reduced to circuit diagrams without losing something, even if no one in the room could say what.

Claude Shannon came as a guest to three of the last four conferences, and his information theory, published in 1948, gave the engineers a precise language for signals, noise and what a channel can carry.²⁴ Von Neumann was a core member who behaved like an outsider, attending irregularly and pursuing his own agenda; by the eighth conference he had effectively dropped out. He still cast a long shadow. At the sixth, a message from him proposing that the brain's complexity could not be accounted for by the number of its neurons set off a debate that ended only when Pitts found an error in von Neumann's arithmetic.²⁵

Wiener attended his last conference in March 1950, the seventh, and did not come to the eighth in 1951 or any after it. By then he was busy with something the other participants were not doing, which the section after next describes. The circle itself broke two years later.

The Break

After the war, Wiener had helped MIT recruit a research group around McCulloch and Pitts, formalising a collaboration that had begun when Pitts moved to Cambridge in 1943. Then, in 1952, Wiener suddenly turned against McCulloch and broke off relations with everyone connected to him, Pitts included. He sent a message to Jerome Wiesner, director of the Research Laboratory of Electronics at MIT: "Please inform [Pitts and Lettvin] that all connection between me and your projects is permanently abolished. They are your problem. Wiener."²⁶

The biographers Flo Conway and Jim Siegelman, in Dark Hero of the Information Age (2005), argue that the break was engineered by Wiener's wife Margaret, who bore an intense hostility toward McCulloch and his bohemian circle. By their account she told Wiener a damaging and apparently false story about members of the group and the Wieners' daughter. Wiener chose to believe it.²⁷

McCulloch never understood why the friendship ended. Pitts, who had come to regard Wiener as a father, was devastated and went into a long decline. In 1959 he was a co-author of "What the Frog's Eye Tells the Frog's Brain", which showed that the frog's retina was itself doing sophisticated processing before any signal reached the brain, a result awkward for a picture of neurons as simple logical switches. Pitts burned his unfinished doctoral dissertation on probabilistic three-dimensional neural networks and years of unpublished work. He died in 1969 of conditions associated with alcoholism.²⁸

The personal wreckage was real, and the break had intellectual consequences. It divided the two most important centres of cybernetic thinking in America just as the field was finding its identity, and it contributed to the fragmentation that left cybernetics without an institutional home. The final Macy conference, in April 1953, was held at Princeton rather than New York. The discussion was animated but, in Teuber's judgement, empty; he threatened to resign if the transcripts were published, and in the end only the papers appeared.²⁹

The Rebel Scientist

Wiener, meanwhile, had been doing something his colleagues at the conferences mostly had not: saying in public what his science implied.

The bombs dropped on Hiroshima and Nagasaki in August 1945 shocked many scientists into moral reflection. For Wiener the shock was ethical and personal. In January 1947 he published a short letter in The Atlantic Monthly under the title "A Scientist Rebels," announcing that he would no longer allow his work to be used by what he called "irresponsible militarists" to build weapons, and calling on other scientists to do the same. The letter was prompted by a request from a researcher at an aircraft corporation for access to his wartime work on guided missiles. He refused, made the refusal public, and from then on refused government funding and military contracts for the rest of his life.³⁰

Two years later he did something odder, for a mathematician. On 13 August 1949 he wrote to Walter Reuther, president of the United Auto Workers. An industrial company had asked him to advise on automatic control mechanisms; he had "turned down unconditionally" the request. Such machinery, he told Reuther, "will undoubtedly lead to the factory without employees," and "in the hands of the present industrial set-up, the unemployment produced by such plants can only be disastrous." His proposal was not that the unions resist the machines. It was that they get there first: "steal a march upon the existing industrial corporations" and take part in producing the machines themselves, "to secure the profits in them to an organization dedicated to the benefit of labor."³⁷

In 1950 he put the argument in a book, The Human Use of Human Beings: Cybernetics and Society. It is remembered now for its warnings about automation, which were specific and grim. Machines capable of feedback and learning would make whole categories of human labour redundant. The automatic machine was, he wrote, "the precise equivalent of slave labor," and any labour that competed with it would have to accept the economic conditions of slave labour. The unemployment that followed would make "the depression of the thirties" seem a pleasant joke. He wrote it in the future tense.³¹

This was not a popular thing to say in 1950, in a country enjoying a boom, and it was dismissed as alarmism. But notice what Wiener had done. Cybernetics asked how a system steers itself. Wiener was asking a different question, which his science could not answer: who holds the tiller, and who gets the catch. He had decided that the second question mattered more than the first, and that the answer could not be left to the people who happened to own the machines.

What Was Cybernetics?

By the time the Macy conferences ended in 1953, cybernetics had produced several important things and failed to produce one.

What it produced: a common vocabulary for feedback, information and control across disciplines that had had no way to talk to each other. The concept of feedback, now so embedded in ordinary speech that its origin is invisible, spread from Wiener's work. The idea that the brain might usefully be compared to a computing machine, however inexact, opened research programmes in neurophysiology and psychology. The McCulloch-Pitts neuron became the ancestor of the perceptron and, at a long remove, of the neural networks that now dominate machine learning.³² The conferences gave all of this a social location: a room where a biologist could argue with an engineer and an anthropologist could ask what a logical calculus had to do with how children learn.

The arguments have not ended. Whether large neural networks are learning or pattern-matching, whether feedback and reward are adequate concepts for intelligent behaviour, whether the brain is better understood as a logic circuit or a continuous dynamical system: these are recognisably the arguments Wiener, McCulloch, Pitts, Bateson and Gerard were having in New York hotel rooms between 1946 and 1953. The mathematics has grown more powerful. The question underneath has not changed.

What cybernetics failed to produce was the unified science Fremont-Smith had hoped for. As the American Society for Cybernetics has observed, the conferences were less a birthplace than a breeding ground, and no one came out of them claiming to have been transformed into a "cybernetician."³³ The disciplines pulled apart. By the mid-1950s information theory was Shannon's field and computer science was becoming its own; the ideas that would become cognitive science were finding homes in psychology and linguistics, though the field's own journal and society did not arrive until the late 1970s.³⁸ Cybernetics, the name and the aspiration, was left without a home. McCulloch's summary of ten years' consensus, written for the final conference, was: "Our most notable agreement is that we have learned to know one another a bit better, and to fight fair in our shirt sleeves."³⁴

And the one thing Wiener added, which his colleagues mostly did not, is the reason the failure matters now. The others wanted to know whether a machine could have a mind. He wanted to know what a machine with something like a mind would do to the people who did not own it. A science that could explain control, and could not say who should have it, was going to be used by whoever got there first. The Macy conferences never put that question on the agenda. The companies building the machines today still prefer not to.

Wiener died in Stockholm on 18 March 1964, aged sixty-nine, of a heart attack. He had received the National Medal of Science from President Johnson at the White House two months earlier.³⁵ His final book, God and Golem, Inc., published in the year of his death, explored the theological implications of machines that learn and reproduce; it won the National Book Award in 1965.³⁶ He had taught machines how to steer themselves. He was rather less confident about the people choosing the destination.