The History of AI

The McCulloch-Pitts neuron

Era 2 · AI is born (1943–1979) › Neural networks and learning

Lineage
Neurons and networks
Kind
Publication

Who

University of Illinois Chicago

What happened

The neurophysiologist Warren McCulloch and the young logician Walter Pitts publish "A Logical Calculus of the Ideas Immanent in Nervous Activity" in the Bulletin of Mathematical Biophysics.

Why it mattered

They model the neuron as a logical unit that fires or not depending on the sum of its inputs, and prove that a network of such units can compute any logical function. They use Carnap's logical language and the notation of Principia Mathematica.

The full story

McCulloch had spent years looking for a theory of the brain that could account for reasoning. Pitts, a self-taught young man with a deep knowledge of Principia Mathematica, supplied the missing logic. Together they started from what neurophysiology knew about neurons: they fire or they don't, all or nothing, and they need a minimum number of stimuli to add up before they do.

With those simplifications, each neuron becomes a logical proposition that is true when it fires. A neuron that fires only if it receives two inputs at once performs a conjunction; one that fires with either of them, a disjunction; and an inhibitory connection acts as negation. The paper proves that a network of these units can represent any expression of propositional logic and, with loops that keep activity going, remember past states.

The model does not learn: the thresholds and connections are fixed. Its importance was conceptual. It showed that a network of simple, brain-inspired elements has the computing power of logic, and it joined neurology, Turing's logic, and machine design in a single language.

Von Neumann adopted these neurons as idealized components in the EDVAC report, and Rosenblatt's perceptron started from them, adding adjustable weights.

Three neurons. The first receives a and b and fires if the sum reaches 2, which gives AND. The second fires if it reaches 1, which gives OR. The third receives a through an inhibitory connection and fires only if a is off, which gives NOT.

With different thresholds, the same unit computes AND, OR, and NOT.

Under the hood

Each unit receives inputs worth 0 or 1, of two kinds: excitatory and inhibitory. It fires, that is, its output is 1, if no inhibitory input is active and the number of active excitatory inputs reaches its threshold θ.

With two inputs and θ = 2, the unit computes "a AND b"; with θ = 1, "a OR b"; with a single inhibitory input and θ = 0, "NOT a." Since any formula of propositional logic can be written with AND, OR, and NOT, a network of these units can compute any logical function of its inputs.

Time advances in steps: the output at time t + 1 depends on the inputs at t. A network with loops can sustain activity that remembers what happened earlier.

Common misconceptions

  • What people often say: The McCulloch-Pitts neuron already learned from experience.

    In fact: Its thresholds and connections are fixed; the network is designed by hand, like a circuit. Weights adjusted from examples arrived with Rosenblatt's perceptron in 1958.

Builds on

  • Principia Mathematica (1910)

    Documented direct influence

    The paper uses their notation and their logical language

  • Recurrent neural circuits (August 1933)

    Documented direct influence

    Lorente de Nó's recurrent circuits sparked McCulloch's interest in circular causality. The 1943 paper does not list Lorente de Nó in its references

  • The all-or-none principle in nerve (February 1914)

    Documented direct influence

    The paper opens this way: because nervous activity is all-or-none, neural events can be treated with propositional logic

  • The Entscheidungsproblem (1928)

    Documented direct influence

    Hilbert and Ackermann's textbook is one of the three works the paper lists in its references

  • On Computable Numbers (November 30, 1936)

    Documented direct influence

    The paper argues that a net with a tape and the means to read and write it computes the same things as a Turing machine

Leads to

  • First Draft of a Report on the EDVAC (June 30, 1945)

    Documented direct influence

    Von Neumann cites the model as a key result and adopts its neural notation

  • The perceptron (1957)

    Documented direct influence

    Rosenblatt starts from the McCulloch-Pitts threshold neuron and discusses their model in the 1958 paper

Threads that pass through here

Sources

How to cite

“The McCulloch-Pitts neuron.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/mcculloch-pitts-neuron

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline