What were the first big milestones in AI?

The story of today’s AI runs through a handful of dated turning points. This first page covers the early ones, before computers were fast enough to make neural networks pay off. The term itself came a little before our timeline starts: the field was named at the 1956 Dartmouth Summer Research Project, the workshop widely treated as the founding event of artificial intelligence.^[1] The funding proposal behind it rested on one sweeping conjecture: “every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”^[1] Nearly seventy years on, that conjecture is still unresolved rather than settled either way.

• 1958 — The perceptron. Frank Rosenblatt built the perceptron at the Cornell Aeronautical Laboratory and demonstrated it publicly in July 1958, on an IBM 704, “a 5-ton computer the size of a room” fed by punch cards.^[2] It was the first trainable neural network, a machine that adjusted itself from examples rather than being programmed by hand.^[2] The underlying algorithm dates to 1957, and the physical Mark I machine was demonstrated in 1960.^[3] Coverage at the time did not undersell it: the New York Times called it an “embryo of computer designed to read and grow wiser,” and Rosenblatt himself predicted a machine “capable of perceiving, recognizing and identifying its surroundings without any human training or control.”^[2] The AI winters that followed are, in part, the story of that prediction meeting reality.

• 1986 — Backpropagation. David Rumelhart, Geoffrey Hinton, and Ronald Williams published “Learning representations by back-propagating errors” in the journal Nature. The method gave a practical way to train networks with hidden layers, which a single perceptron could not handle.^[4]

Between these dates the field went through long stretches of disappointment, sometimes called AI winters, when funding and interest dried up. The ideas were sound, but the hardware and the data needed to make them work had not arrived yet. That changed in the years covered on the next page.