SNARC — the first neural network in hardware

SNARC: neural learning in copper and tubes

In 1951, Marvin Minsky and Dean Edmonds, two 24-year-old graduate students at Princeton University, built the Stochastic Neural Analog Reinforcement Calculator — SNARC. It was the first machine that physically implemented a neural network: not simulated in software, but built from real electrical components, including 40 Hebb synapses and a surplus autopilot mechanism from a B-24 bomber.

How SNARC worked

SNARC simulated a rat navigating a maze. Each synapse was represented by a capacitor that could adjust its "weight" based on reinforcement signals — more or less in accordance with Donald Hebb's learning rule from 1949, which states that connections between neurons that fire together are strengthened. The machine had no central processor; learning was distributed across the network of 40 analog synapses.

Why it matters

SNARC proved that the theoretical McCulloch-Pitts neuron from 1943 could actually be built as a physical learning machine. It demonstrated that adaptive behavior — learning from experience — could emerge from a network of simple components without explicit programming. This was a radical idea in an era where "computing" still meant following a predetermined sequence of instructions.

Minsky's later influence

Marvin Minsky went on to become one of the founding fathers of AI. He co-founded the MIT AI Lab in 1959 with John McCarthy, wrote the influential book Perceptrons (1969) with Seymour Papert — which, ironically, helped trigger the first AI winter by demonstrating the limitations of simple perceptrons — and made major contributions to knowledge representation, robotics, and the theory of mind. He remained a central figure in AI until his death in 2016.

Sources

• Minsky, M. (1954). Neural Nets and the Brain-Model Problem. PhD thesis, Princeton University.
• Crevier, D. (1993). AI: The Tumultuous History of the Search for Artificial Intelligence. Basic Books.
• Wikipedia — SNARC
• Wikipedia — Marvin Minsky