In 1958 Frank Rosenblatt, at Cornell Aeronautical Laboratory, publishes "The Perceptron" and implements the Mark I Perceptron: the first neural network able to learn from examples via a supervised learning rule adjusting weights proportionally to error — the birth of machine learning as practice. It can learn linearly separable patterns without explicit programming of rules. Optimism is checked in 1969 when Minsky and Papert mathematically prove single-layer perceptron limitations, triggering the first AI winter.