Paper 02
Beginner

The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain

Turing asked if machines could think. Rosenblatt built one that could learn. The Perceptron is the grandfather of every neural network alive today — the first machine that adjusted itself based on experience, rather than following rules someone wrote by hand.

The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain

Frank Rosenblatt · 1958 · Cornell Aeronautical Laboratory


“The perceptron is… the first machine which is capable of having an original idea.” — New Yorker, 1958 (slightly over-enthusiastic, but you get the point)

In 1950, Turing asked: can machines think?

In 1958, Rosenblatt built a machine that could learn.

Not follow instructions. Not compute a formula someone had written down. Learn — look at examples, make mistakes, adjust, and get better. The way a student does.

The machine was called the Perceptron. It was physically built — a room-sized contraption with motors, wires, and 512 photocells — and it could learn to recognise which side of a card a mark was placed on. That sounds trivial. It was revolutionary.

Every neural network in existence today — the ones powering Google Translate, ChatGPT, DALL-E, AlphaFold — is built on the idea Rosenblatt demonstrated with this machine.


What is in this paper?

SectionWhat you will learn
Historical ContextPost-Turing world, neuroscience in the 1950s, what “learning” meant before this paper
The ProblemWhy rule-based AI failed, and what Rosenblatt set out to build instead
The Core IdeaWeighted inputs, thresholds, the biological neuron analogy
How It WorksForward pass, prediction, error, weight update — step by step
The MathematicsDot products, the perceptron update rule, worked numerical example
The CodeBuild a perceptron from scratch in 25 lines of Python
Why It MatteredThe birth of connectionism, what this enabled
LimitationsThe XOR problem that broke the perceptron, the first AI winter
What Came NextHow backpropagation fixed what the perceptron could not do

Paper at a glance


Start reading

Begin with Historical Context →


Previous paper: Computing Machinery and Intelligence — Turing (1950) ← Next paper in the timeline: Backpropagation (1986) →

Discussion

Questions about this paper? Spotted something unclear? Start a discussion below — powered by GitHub, no separate account needed.