In 1958, Frank Rosenblatt described a machine that learned from examples instead of following fixed rules. His perceptron looked at shapes through a grid of 400 photocells and learned to tell them apart, and the Mark I hardware that followed now sits in the Smithsonian. Every neural network we use today still starts from this one idea.
1. One neuron, three steps
Each input gets a weight, the neuron adds them up, and it fires only if the sum clears the threshold. The weights are the whole memory of the machine.
2. A straight line decides
A perceptron is a line drawer. It sorts points into two groups with one straight cut, and it learns by moving that line until the groups sit on the right sides.
3. Learning is nudging the weights
The perceptron guesses, checks the answer, and nudges each weight a little toward the right one. Rosenblatt proved that if a straight line can separate the groups, this loop is guaranteed to find it.
4. The line that cannot be drawn
XOR outputs 1 only when the inputs differ, and no single straight line can separate its four points. Minsky and Papert proved this in 1969, and the field went quiet for years.
Key takeaway
The perceptron gave us the two ideas every modern network still runs on: weighted sums and learning from mistakes. Its limit was just as valuable as its success, because it showed exactly what one layer cannot do.
What is Next in This Series
Next is Part 2, The Multilayer Perceptron, Visualized, where stacking neurons and adding backpropagation finally cracks XOR. For the deeper story of how these ideas grew into modern LLMs, see How LLMs Actually Work.
Originally published on LogicStack. Run the code examples right in your browser with our free interactive Python and SQL playgrounds — no setup, no sign-up.
Top comments (0)