SNARC was the first artificial neural network computer, built in 1951.
SNARC stands for Stochastic Neural Analog Reinforcement Calculator. Marvin Minsky and Dean Edmonds constructed it at Harvard University while Minsky was a graduate student. The machine used about 3,000 vacuum tubes and simulated a small network of 40 artificial neurons.
Its purpose was to model how a rat might learn to navigate a maze. The system used random electrical signals and adjustable connections to represent reinforcement, making it an early physical demonstration of machine-like learning inspired by the brain.
SNARC predates the perceptron, which Frank Rosenblatt developed in 1957. That distinction matters because the two machines are often confused in popular summaries of neural-network history. SNARC was an analog, experimental device rather than a modern programmable digital neural network, but it helped establish the idea that a machine could change its behavior through feedback.