Warren Sturgis McCulloch

• 1. Introduction

  ›

  Warren Sturgis McCulloch spent the spring of 1947 building a machine meant to help blind people read. It converted printed letters into tones, designed so the tone for each letter stayed the same no matter what angle the letter was viewed from. When the researcher Gerhardt von Bonin examined the design, he asked McCulloch something unexpected. "Is this a diagram of the fourth layer of the visual cortex of the brain?"

  The question wasn't far off. McCulloch had spent decades treating the wiring of the human brain as a kind of logic, something that could be modeled, tested, and eventually built. But he hadn't set out to become a scientist at all. What turned a young man once headed toward the Christian ministry into someone mapping neurons with mathematics? How did a single 1943 paper convince a generation of researchers that thinking itself could be written down as equations? And how did the same man who traced brain circuits with strychnine also find time to write sonnets and design a dam? This documentary follows those threads.

• 2. From Pulpit to Laboratory

  ›

  Warren Sturgis McCulloch was born on the 16th of November 1898, in Orange, New Jersey. His brother went on to become a chemical engineer. Warren, though, originally planned to enter the Christian ministry. As a teenager, he moved in circles with the theologians Henry Sloane Coffin, Harry Emerson Fosdick, Herman Karl Wilhelm Kumm, and Julian F. Hecker. The Quaker Rufus Jones mentored him as well.

  He studied philosophy and psychology at Yale University after attending Haverford College, earning a Bachelor of Arts degree in 1921. Further study at Columbia brought him a Master of Arts in psychology in 1923. In 1927 he earned his medical degree from Columbia University's College of Physicians and Surgeons. He then interned at Bellevue Hospital in New York. After that, he worked under Eilhard von Domarus at the Rockland State Hospital for the Insane.

  By 1934 McCulloch had returned to academic life, joining Yale's Laboratory for Neurophysiology. He stayed there until 1941. That year he moved to Chicago and joined the Department of Psychiatry at the University of Illinois at Chicago. He served as a professor of psychiatry there and directed the Illinois Neuropsychiatric Institute until 1951. From 1952 he worked at the Massachusetts Institute of Technology in Cambridge, alongside Norbert Wiener.

  McCulloch also helped build the community around this emerging science. He was a founding member of the American Society for Cybernetics and served as its second president during 1967-1968. He mentored the British operations research pioneer Stafford Beer as well. In Chicago, though, he had already met someone who would shape his most influential work: a young logician named Walter Pitts.

• 3. The Grammar of Neurons

  ›

  In 1919, McCulloch turned his attention mainly toward mathematical logic. By 1923 he was attempting to build a logic of transitive verbs. His goal was to define what he called a psychon, a least psychic event: a binary, all-or-nothing occurrence with a necessary cause. He believed these events could combine into complex logical statements about their own causes. In 1929 he noticed they might match how neurons fire, either completely or not at all.

  Working with Walter Pitts of the University of Chicago, McCulloch published "A Logical Calculus of the Ideas Immanent in Nervous Activity" in 1943. The paper appeared in the Bulletin of Mathematical Biophysics, volume 5, pages 115-133. It described how memories might form through neural networks containing loops, or synapses that could be altered. It also showed that such looped networks could encode any statement in first-order logic with equality, and that the reverse held true as well. Later assessments would call it a seminal contribution to neural network theory, the theory of automata, the theory of computation, and cybernetics.

  McCulloch spent years afterward developing triadic relations, extending the logic of two-way relationships into three-way ones. An example would be "A gives B to C," or "A perceives B to be C." He was convinced this kind of logic was necessary to understand how the brain actually worked.

  Later, working with Manuel Blum, McCulloch asked whether a neural network could stay "logically stable." That meant performing the same Boolean function even as individual neurons' firing thresholds shifted. The question had a very physical inspiration: how the brain keeps functions like breathing steady under caffeine or alcohol. Both substances shift firing thresholds across the entire brain. By 1947, McCulloch and Pitts turned this same logical toolkit toward a much harder question. How does the brain recognize something it has only ever seen from a different angle?

• 4. Teaching a Machine to Recognize

  ›

  In their 1947 paper, "How We Know Universals: The Perception of Auditory and Visual Forms," McCulloch and Pitts tackled exactly that question. A square looks different from different angles and under different lighting, yet a viewer still recognizes it as the same shape. A spoken phoneme works the same way, sounding different at different volumes and pitches while remaining recognizable. The paper described this as recognizing an object despite the action of some symmetry group upon it.

  Their first proposed solution was computing an invariant by averaging a neural network's response across that entire symmetry group. Their second was a negative feedback circuit that produced a fixed, canonical representation. To recognize a square, the circuit could move the eye until the object's "center of gravity of brightness" reached the middle of the visual field. This converted the object into a fixed form the brain could compare against its own stored representations.

  That same 1943 paper had already argued that a Turing machine's program could be implemented inside a finite network of formal neurons. In this view, the neuron itself became the brain's basic logic unit. From 1952, at MIT's Research Laboratory of Electronics, McCulloch's team tested these ideas by studying the visual system of the frog. They found that the frog's eye already organized and interpreted visual information before sending it onward, rather than simply relaying a raw image.

  That research became the 1959 paper "What The Frog's Eye Tells The Frog's Brain," co-written with Jerome Lettvin, H. R. Maturana, and Walter Pitts. It appeared in the Proceedings of the I.R.E., volume 47, issue 11.

• 5. Mapping the Living Brain

  ›

  Applying strychnine to a single point on the brain, McCulloch found, produced excitation at other points entirely. He used this effect, known as strychnine neuronography, as a method for mapping the brain's internal connections.

  Alongside Bailey and Bonin, McCulloch ran a series of studies during the 1940s tracing connections in the brains of macaques and chimpanzees. Those studies identified links consistent with what researchers now understand as the VOF, a pathway confirmed later by modern methods.

  In his final years, McCulloch would take up an entirely different kind of collaboration, with a researcher named Roberto Moreno-Díaz.

• 6. Modeling Memory in Loops

  ›

  With Roberto Moreno-Díaz, McCulloch modeled memory as a pattern of oscillations circling through a loop of neurons. Together they worked out how many distinct oscillation patterns a neural network of a given size could sustain, a count credited to Schnabel in 1966.

  They also proved a universality theorem. For any oscillation pattern a smaller network could produce, they showed a larger network with binary inputs could reproduce it given the right input.

  In his last years during the 1960s, McCulloch also returned to loops, oscillations, and triadic relations with Moreno-Díaz. He studied the reticular formation with a researcher named Kilmer, and worked out dynamic models of memory with Da Fonseca. This decade of work was drawn together into a single 1968 paper, his last major synthesis.

• 7. When the Brain Disagrees With Itself

  ›

  McCulloch called it a heterarchy of motives: contradictory information and motives that could not be arranged into one strict chain of command. Instead, he believed, they followed a looser, more tangled order. He proposed that the brain manages this through what he called "poker chip" reticular formations, a kind of democratic, evenly distributed neural network.

  The idea addressed a specific puzzle: how does the brain commit an animal to just one course of action when the situation itself is ambiguous? Working with Kilmer, McCulloch designed a prototype network called RETIC, built from "12 anastomatically coupled modules stacked in columnar array." It could switch between stable, unambiguous modes even when its inputs stayed unclear.

  His principle of "Redundancy of Potential Command" outlived him in other researchers' hands. Von Foerster and Pask developed it further in their study of self-organization. Pask also carried it forward into his own Conversation Theory and Interactions of Actors Theory.

  One of his own essays, "Recollections of the Many Sources of Cybernetics," would not reach print until 1974, in the journal ASC Forum.

• 8. Poetry, Dams, and Family

  ›

  At his farm in Old Lyme, Connecticut, McCulloch turned engineer on the side, drafting plans for buildings and a dam. Alongside his scientific papers, he also wrote sonnets.

  In 1924 he married Ruth Metzger, known to those close to her as "Rook." The couple went on to raise three children together.

  He died in Cambridge, Massachusetts, on the 24th of September 1969. His 1965 book, Embodiments of Mind, gathered decades of his thinking under one cover. In 1993, a fuller volume, The Complete Works of Warren S. McCulloch, brought his writing to readers who had never met him.