Portfolio · Computing Lineage
Floor 3 — McCulloch & Pitts: The Neuron Made Into a Gate
In 1943, two men wrote down a mathematics of the neuron. They chose the threshold. The biology was already shouting that the neuron was continuous — and they heard it, and chose the threshold anyway.
The floor
In 1943, Warren McCulloch and Walter Pitts published A Logical Calculus of the Ideas Immanent in Nervous Activity. It is one of the foundational papers of artificial intelligence. In it, they modelled the neuron as a threshold gate: inputs arrive, they are summed with weights, and if the sum exceeds a threshold the neuron fires. Otherwise it stays silent.
This model is what every artificial neural network, from the Mark I Perceptron to GPT-5, is a descendant of. The architecture on the chip this sentence was typed on owes its shape to the choice made on this floor.
What was picked
The neuron as a Boolean gate. Biology reduced to a step function. A real neuron was permitted to have weights, permitted to have a threshold, permitted to sum — but was not permitted to be in between. It either fired, or it did not.
McCulloch and Pitts knew what they were doing. The paper is explicit about the abstraction. They wanted a model that was formally tractable, that could be composed like logic gates, that Turing could have proven things about. They chose a version of the neuron that obeyed the rules of Floor 1.
It worked. It gave us a mathematics we could build with. Nobody should be angry at these men.
But they made the choice. And the field, for eighty years, has been building on it.
What could have been picked
The actual neuron was never a threshold gate, and everyone knew it.
By 1943, physiologists had been measuring graded membrane potentials for decades. Hodgkin and Huxley were already closing in on their continuous differential-equation model of the action potential — it would land in 1952, and would win a Nobel Prize for describing the neuron as a smoothly varying voltage responding to continuously summed inputs over time.
Even more: the neuron's firing rate carries information. A neuron that fires ten times a second means something different from one that fires a hundred times a second. The Boolean gate throws that away. The biological neuron holds it.
The continuous neuron was not speculative. It was on every biology bench in the world. The fork was taken in full sight of the alternative.
What we missed
Eighty years of "neural" networks that are not, in the biological sense, neural. A discipline that had to re-invent continuous activations (sigmoid, tanh, GELU) bolt-by-bolt, each time justifying itself against the ghost of the threshold. A hundred papers a month about how to push "almost-firing" information through a stack of gates that were designed to discard it.
And, downstream, a vocabulary problem. We call these things neurons, and then we are surprised when they don't do what neurons do. They were never meant to. They were meant to do what a 1943 logician could prove theorems about.
The alternate timeline is an artificial nervous system where the activation is a position, the threshold is an emergent approximation for when the output needs to be read discretely, and the system's work is continuously happening even when nothing has crossed a line.
We have every piece of that. We just can't quite get the field to put the threshold down.
What the next floor will ask
If the neuron can be continuous, why is the machine around it still shaped like a tape?
That's Floor 4.